Mathematical Definition and Meta-Analysis Modeling of Tourist

Mathematical Definition and Meta-Analysis Modeling of Tourist Preferences and Purpose in DIKWP × DIKWP Semantic Space

Yucong Duan

International Standardization Committee of Networked DIKWPfor Artificial Intelligence Evaluation(DIKWP-SC)

World Artificial Consciousness CIC(WAC)

World Conference on Artificial Consciousness(WCAC)

(Email: duanyucong@hotmail.com)

Introduction

In tourism behavior research and the construction of intelligent recommendation systems, "tourist preference" and "tourist purpose" are two core concepts. However, academia currently lacks a clear ontological distinction regarding the connotation of preference and purpose and their interrelationship. In traditional models, preference is often seen as a relatively stable interest tendency of tourists, while purpose manifests as a tourist's target behavior plan in a specific context. But in the complex process of tourism decision-making, preference and purpose often intertwine, making them difficult to clearly define theoretically. To enhance the precision and interpretability of intelligent tourism recommendations, it is necessary to establish a unified semantic model to characterize the differences and connections between preference and purpose.

In recent years, Professor Yucong Duan's team proposed the DIKWP network interaction model, providing an innovative framework for understanding the relationships between Data (D), Information (I), Knowledge (K), Wisdom (W), and Purpose(P). This model breaks through the linear hierarchical structure of the traditional DIKW pyramid, adds "purpose" to the cognitive process, and enables a non-hierarchical, non-unidirectional network of semantic interaction among the levels. Under this model, the five elements of data, information, knowledge, wisdom, and purpose do not have a simple bottom-up cumulative relationship but form a closed-loop network through multi-directional feedback (e.g., purpose can guide data collection, and knowledge can in turn influence data acquisition). Such a semantic network provides us with a new perspective to distinguish between preference and purpose.

Research Objectives: This study, based on the DIKWP network model, focuses on the tourism domain, attempting to conduct mathematical semantic definition and model construction for tourist preferences and purpose, and to verify the model's effectiveness through a systematic meta-analysis. The main objectives include:

Semantic Mathematical Definition of Preference: In the non-linear DIKWP semantic network structure, define "tourist preference" as a kind of stable path transformation pattern or weight distribution that tourists exhibit in a multi-dimensional information state (D, I, K, W, P). For example, a tourist is accustomed to gradually forming knowledge from raw data to guide purpose (path pattern like  D→I→K→P ), or tends to use existing knowledge to find new data to meet a target ( K→D→P ), or integrates scattered information into insight before implementing a purpose ( I→W→P ), etc. This preference for path selection can be regarded as a transformation weight with directionality and stability in the DIKWP network.

Semantic Generation Mechanism of Purpose: Analyze the generation mechanism of "tourist purpose" in the semantic network in parallel. In the DIKWP model, purpose corresponds to the "Purpose" layer, playing a role in driving and guiding the entire cognitive transformation process. We will define purpose as a semantic path feature oriented towards goal achievement, typically manifested by emphasizing the transformation from wisdom to purpose ( W→P ), or using purpose as a starting point to work backward on resource utilization (i.e., a "P-dominated" path). Compared to preference, purpose is more directly reflected in behavior planning around the final purpose. For example, when a tourist clearly decides "I want to achieve a certain travel goal," the purpose may drive them from wisdom/experience directly to the specific action goal.

Structural Differences between Preference and Purpose: Characterize the structural differences between preference and purpose in the DIKWP × DIKWP dual space. We will adopt the conceptual space and semantic space interaction model proposed by Yucong Duan, where one axis is the subject's cognitive conceptual space (subjective DIKWP), and the other axis is the objective information semantic space (objective DIKWP), with the two interacting through a network relationship. Under this framework, preference focuses more on how the subject's internal knowledge and information are processed (e.g., the preference path reflects high-frequency transformations from knowledge to data or information to knowledge in the conceptual space). In contrast, purpose manifests as the subject projecting their own wisdom or purpose onto the objective environment (e.g., interactions of wisdom × data, purpose × information) to achieve a goal. We will quantitatively describe the differences between the two in connection patterns and weight distribution within the semantic network.

Construction of Tourist Preference-Purpose Recognition Model: Based on the above definitions, establish a model to identify tourist preferences and purpose, and map it to decision-making and behavior prediction in tourism recommendation systems. The model will comprehensively consider the identification of tourists' long-term preferences (stable path tendencies) and short-term purpose (current purpose-orientation). For example, using tourists' past behavior data to learn their preference weight distribution, while combining context to extract the current travel purpose, thereby predicting their decision-making behavior. We will provide the model's framework and mathematical expression, and discuss how to fuse preference and purpose in recommendation algorithms to improve the accuracy and interpretability of recommendation results.

Meta-Analysis of Empirical Research: Conduct a systematic review and meta-analysis of empirical studies on tourist preference construction, behavioral purpose prediction, and the application of DIKW/DIKWP models in Chinese and English literature from the past decade. By retrieving relevant studies from 2015-2025, we will extract effect size indicators to quantify the strength of the influence of preference and purpose-related variables on tourism behavior. For example, extracting influence coefficients of tourist satisfaction, motivation, destination image, etc., on behavioral purpose or loyalty. The meta-analysis will use a random-effects model to integrate effect sizes, assess heterogeneity (Q-test and  I 2  indicator), and use funnel plots to check for publication bias. The meta-analysis results will be used to verify and enrich the hypotheses of the DIKWP preference-purpose model.

Technical Application Suggestions: Based on semantic path analysis and statistical conclusions, propose strategies for applying the DIKWP semantic model to intelligent tourism recommendation systems. This includes how to use the DIKWP model to enhance user profiling, improve the semantic understanding of recommendation algorithms, and how to increase system interpretability and user satisfaction through a transparent preference-purpose semantic network.

The innovation of this research lies in combining cognitive semantic theory (DIKWP model) with data-driven evidence (meta-analysis), to, for the first time, perform a rigorous mathematical semantic distinction and connection modeling of tourist preference and purpose. This not only provides a new theoretical framework for tourism behavior research but also has practical guiding significance for the development of smart tourism systems.

Next, this paper will first introduce the DIKWP model and related theoretical background, then propose the mathematical semantic definitions and recognition model for tourist preference and purpose, followed by reporting the methods and results of the meta-analysis, and finally discussing the implications of fusing the model with meta-analysis conclusions, and providing application suggestions for intelligent tourism systems.

Theoretical Basis: From DIKW Pyramid to DIKWP Network Semantic Model

The DIKW Model and Its Limitations

The DIKW pyramid model is a classic framework in the field of knowledge management, dividing Data (D), Information (I), Knowledge (K), and Wisdom (W) into progressive levels, describing the transformation process from raw data to wise decision-making. In this model, Data is unprocessed raw facts, which form Information through processing and a grant of context; information becomes Knowledge after being understood and integrated; and the high-level refinement and value judgment of knowledge produce Wisdom. The DIKW model vividly reflects the process of cognitive deepening and has been enlightening for the design of information systems and decision support.

However, the traditional DIKW model has distinct linear hierarchical characteristics, implying that cognitive processing is a unidirectional, layer-by-layer upward process. In actual complex decision-making scenarios, this bottom-up pyramid structure has limitations:

Lack of Feedback between Levels: In the DIKW model, lower levels (like data) can only be converted upwards into higher-level information, knowledge, without reflecting the counter-effect of high-level wisdom or purpose on low-level processes. In human cognition, there often exist top-down information flows, such as goal-driven data collection and knowledge influencing perception, which linear models struggle to explain.

Semantic Incompleteness and Inconsistency Issues: Real-world information is often incomplete, imprecise, or even inconsistent, whereas the pyramid model requires each layer's input to achieve a certain completeness before it can be transformed upwards. This leads to the DIKW transformation being unable to proceed smoothly in uncertain environments lacking purpose guidance. For example, large-model AIs, when lacking clear-target purpose, will exhibit semantic imprecision and decision-making opacity when processing data/knowledge.

Absence of Purpose (Purpose): The original DIKW model did not explicitly include the element of "Purpose" or "Purpose." However, in decision-making and behavior, "purpose" is often the key driver of the cognitive process. Without the guidance of purpose, judgments at the wisdom level also lose direction. Traditional models find it difficult to describe how humans consider factors like goals and motives (purpose) when making decisions.

Therefore, scholars have tried to extend the DIKW model to overcome the above shortcomings. One important extension is to add Purpose (P) to DIKW, forming a DIKWP structure, and transforming the relationship of the elements from a linear hierarchy to a network interaction.

The DIKWP Network Interaction Model

The DIKWP model was first proposed by Professor Yucong Duan's team and is an extension of the DIKW model in two aspects: first, it adds the fifth element of "Purpose/Purpose (P)"; second, it transforms the original hierarchical structure into a multi-directional feedback network structure. Figure 1 vividly shows the semantic meta-structure of the DIKWP model and the relationships between the elements.

[Image: Figure 1. DIKWP graph mapping of incomplete, inconsistent, and imprecise subjective and objective resources. This model extends knowledge graphs to interconnected data graphs, information graphs, knowledge graphs, wisdom graphs, and purpose graphs, used to map incomplete, inconsistent, and imprecise subjective and objective DIKWP resources, supporting multi-dimensional cognitive interaction.]

In the DIKWP model, the five elements of Data, Information, Knowledge, Wisdom, and Purpose form a closed-loop network through two-way and multi-way connections, rather than a simple "Data → Information → Knowledge → Wisdom" single-line process. Specifically:

Data (D): Still the starting point of cognition, representing raw objective facts, sensory inputs, etc. The DIKWP model retains the transformation mechanism from data to information, analogous to the human perception process—transforming objective data into subjective cognitive information through interpretation and assigning meaning.

Information (I): Is the meaningful expression of data. DIKWP emphasizes that information can flow in both directions: not only can data be transformed into information, but information can also be decomposed or fed back into data. For example, the information (I) a tourist obtains from a travel brochure can be transformed from scattered data from various places; at the same time, if the information is found to be insufficient, they might go to collect new data to enrich the information ( I→D ).

Knowledge (K): A knowledge structure is formed through the organization, association, and pattern recognition of information. Unlike the linear model, DIKWP allows knowledge not only to be accumulated from information but also to react back on information and data. For example, an experienced tourist (with existing knowledge K) will purposefully seek specific information or data to verify their ideas ( K→I→D ), which is information acquisition driven by knowledge.

Wisdom (W): In the model, this refers to the ability to make wise decisions and judgments based on knowledge. The wisdom layer incorporates components of values, insights, and long-term experience, representing a higher-level application of knowledge. DIKWP emphasizes that wisdom is not just passively formed but also evaluates and provides feedback on knowledge and information. For example, a wise tourist can use past experience (W) to guide current information screening and decision-making ( W→I ,  W→K ).

Purpose (P): This is the biggest innovation of the DIKWP model compared to DIKW. Purpose, as a subjective agency factor, is placed on par with data/information/knowledge/wisdom, but it actually runs through and drives the transformation of all other elements. Yucong Duan pointed out: "Purpose represents the goal of humans using data, information, knowledge, and wisdom; it promotes the transformation and interaction between these elements." In other words, purpose injects directionality and motivation into the network: with a purpose, data can be given relevant meaning, and knowledge can be combined with value judgments to finally form wise decisions. Purpose can be both the final target state to be achieved, or it can be regarded as a high-level constraint or preference that guides the cognitive process.

The DIKWP network model has the following characteristics:

Two-way Feedback: Connections and transformations can exist between any two levels. Especially, top-down purpose guidance is the essence of the model. For example, purpose (P) can directly influence data collection strategies ( P→D ) or information screening criteria ( P→I ). This feedback ensures that the system can still evolve towards the goal even with incomplete information.

Multi-layer Parallelism: Data, information, knowledge, and wisdom are no longer strictly hierarchical but can interact in parallel. For example, in complex decision-making, data collection and knowledge application may be processed simultaneously, and judgments from the wisdom layer may also intervene in real-time to provide guidance. The model allows for this multi-layer parallel processing.

Semantic Completeness: By adding the purpose layer, the DIKWP model is claimed to be able to overcome the problems of semantic incompleteness and inconsistency in natural language. This is because purpose provides a unified semantic reference, enabling AI or human cognitive processes to bridge information fragments under goal orientation, thereby improving consistency and accuracy.

Wide Application: DIKWP was initially proposed in the fields of artificial consciousness and cognitive computing, but its ideas are universal. In recent years, it has been tentatively applied to human-computer interaction modeling in various fields such as justice, healthcare, and education. Human-computer intelligent systems in the tourism field can also consider using DIKWP as the underlying framework to better handle the interaction between users and the tourism information environment.

In summary, the DIKWP model provides us with a new method to characterize the tourist cognitive and decision-making process. Below, we will, based on this model, define and differentiate tourist preference and tourist purpose semantically.

Mathematical Definition of Tourist Preference in DIKWP Semantics

In the tourism context, tourist preference usually refers to a tourist's relatively stable inclination towards destinations, activity types, travel methods, etc. This preference is embodied in the consistent patterns and tendency strength they exhibit when facing choices. Traditional research often identifies tourist preferences through questionnaires or behavioral analysis, such as a tourist preferring natural scenery over urban landscapes, or independent travel over group tours. But these descriptions often remain at the empirical level. With the help of the DIKWP model, we can formalize preference as the path transformation structure that tourists favor in the cognitive network.

Preference as Path Transformation Selection

Based on the DIKWP semantic network, we view the 5 elements as nodes in a graph  V=D,I,K,W,P , and the nodes are connected by directed edges  E , representing a cognitive transformation (e.g.,  D →I  represents data being converted into information). Since DIKWP is a network structure, in principle, meaningful transformations can exist between any two nodes. But considering cognitive meaning, we mainly focus on a set of typical transformation relationships, for example:  D→I, →K, →W, →P  (typical cognitive deepening); it also includes reverse or cross-layer transformations, such as  I →D  (decomposing information into data, e.g., verifying information),  K →I  (deducing information from knowledge, e.g., using existing knowledge to judge implicit information),  K →D  (guiding new data collection from knowledge),  W →I / K  (using wisdom and experience to screen information/adjust knowledge), etc. More critically, with the introduction of Purpose P, there are also interactions between  P  and other levels, such as  P →D,I,K,W  representing the direct guidance of the goal on cognitive activities at all levels, and  D,I,K,W →P  representing the path of various cognitive results accumulating to reach the purpose.

In such a highly connected directed graph, a certain tourist's "preference" can be defined as: the specific path or local subgraph they tend to adopt on this directed graph, and the corresponding selection probability or weight. In other words, preference manifests as a stable semantic transformation pattern that the tourist's brain (or auxiliary decision-making system) inherently follows when processing travel decisions. We can express this mathematically as follows:

Let  P  be the set of all directed paths from a starting point to an endpoint. In tourism decision-making, a path generally starts from the perception stage (Data or Information) and finally leads to an action decision (Purpose). For example, the path  D →I →K →P  represents a typical "data-to-purpose" decision chain.

For a specific tourist  t , define their preference as a path probability distribution:

Pre f t :P→ [ 0,1 ] , such that  ∑ p ∈ P Pre f t ( p ) =1

where  Pre f t ( p )  represents the strength of tourist  t 's tendency to adopt path  p  in cognitive decision-making. The larger the  Pre f t ( p )  value, the more the tourist prefers to complete the cognitive process from acquiring information to achieving the purpose in the manner of path  p .

In practical applications, we can also focus on several typical path patterns rather than all possible paths. For example, consider three typical preference paths:

Preference Pattern A: D →I →K →P  (a bottom-up, step-by-step path);

Preference Pattern B: K →D →P  (having some knowledge first, then purposefully collecting data to achieve the goal);

Preference Pattern C: I →W →P  (focusing on sublimating information into insight, then directly using it to achieve the purpose).

Then,  w A , w B , w C  can be used to represent the tourist's weights on these three types of paths, and  w A + w B + w C =1 . These three types are just examples; in reality, preference patterns can be richer, such as some tourists preferring  W →P  for direct decision-making, some preferring  D →P  to skip deep processing, etc.

Through the above definition, tourist preference is mathematically characterized as the probability distribution of path selection in the DIKWP network. This definition includes the traditional understanding of preference: preference is stable (because the probability distribution is relatively fixed in the short term), preference has directionality (because the path is a directed transformation), and it reflects a certain information processing strategy. For example, a tourist who prefers  D→I→K→P  usually manifests as first collecting a large amount of materials ( D→I ), then systematically learning to form cognition ( I→K ), and finally making a travel plan based on knowledge ( K→P ). Conversely, a tourist who prefers  K→D→P  tends to have an idea/preference first (existing knowledge  K ), and then purposefully acquire specific data ( K→D ) to implement the plan ( P ).

The Directionality and Stability of Preference

In the DIKWP interaction network, preference can be vividly regarded as a set of weighted "habitual pathways (inertial paths)." The directionality of preference means that once a semantic path becomes a preference, its forward usage frequency is much higher than its reverse. For example, preferring  D→I→K→P  means the tourist is accustomed to forward deduction from data to knowledge to purpose, and is unlikely to do the reverse (rarely proceeding in reverse as  P→K→I→D ). This choice in direction reflects the bias of the subjective cognitive flow.

The stability of preference, on the other hand, refers to the cross-temporal consistency of this path preference, which does not fluctuate easily due to a single travel decision change. Preference, as a result of long-term individual accumulation, can correspond to attitudes and values in psychology. In fact, the Wisdom (W) layer of the DIKWP model contains value judgments and long-term experience. We can believe that preference is precisely the manifestation of the stable value orientation rooted in the wisdom layer in the selection of specific cognitive paths: the wisdom layer endows certain paths with higher subjective value, hence they are repeatedly adopted. For example, some tourists value "deeply learning destination knowledge," so their preference path will focus on activities at the K (knowledge) level; some tourists' values lean towards "satisfying curiosity in a timely manner," thus they may prefer to gain satisfaction directly from data experience (e.g., a shortcut like  I→P ).

It should be noted that preference is not always

fully explicit. Sometimes tourists themselves may not be clear about their path preferences, but this pattern can be mined by analyzing their behavioral data. For example, Wen Xuan et al. (2023) used IoT sensor data and inverse reinforcement learning techniques to learn "fine-grained preferences" from tourists' behavioral trajectories within scenic spots, i.e., identifying which attraction sequences and tour patterns tourists unconsciously prefer. Their experiments show that this method can effectively mine individual preferences even with only a small amount of behavioral data. This indicates that our mathematical preference model has practical operational significance: it is possible to estimate their  Pre f t ( p )  distribution from tourist behavior through probability models and machine learning.

In summary, tourist preference is defined in the DIKWP semantic space as a path transformation selection with directionality and stability. This definition respects the psychological stability attribute of preference, and also characterizes the specific embodiment of preference in the information processing process (path weight). With a clear definition of preference, we will next define tourist purpose and compare the differences between the two.

Semantic Generation Mechanism and DIKWP Definition of Tourist Purpose

Tourist purpose usually refers to the purpose or plan that a tourist wants to achieve in a specific context, such as "planning to travel somewhere next year" or "preparing to book a characteristic homestay." Purpose has the characteristics of context-dependence and goal-orientation, and is more immediate and active compared to preference. In the DIKWP model, purpose corresponds exactly to the Purpose layer (P), and its role is like an "engine" or "command center," driving and regulating the operation of the entire cognitive network. Below, we will explain the manifestation of purpose in the DIKWP network from the perspective of semantic generation, and its structural differences from preference.

The Generation of Purpose and the W→P Path

In the cognitive process, purpose can be seen as a clear target signal "stimulated" from the wisdom layer under certain trigger conditions. According to the DIKWP principle, the Wisdom (W) layer combines knowledge and values to make comprehensive judgments. When conditions are ripe, it can produce a decision-making impulse to achieve a certain goal—this is precisely the moment of purpose generation. Therefore, a common purpose semantic path is  W→P : that is, wisdom leads to purpose. For example, a tourist, after long-term thinking (W), finally decides "I want to go on a pilgrimage to Tibet" (P); or after accumulating a series of experiences during the trip, instantly realizes "Next, I want to try skydiving" (W layer experience triggers a new goal P).

The  W→P  path reveals that purpose often originates from higher-level comprehensive cognition: it requires the accumulation of knowledge and experience, as well as the support of value judgments. When the wisdom layer generates some dissatisfaction with the status quo or a new pursuit, a new goal (purpose) will be generated to guide action.

Of course, purpose may also be directly induced by external stimuli, not necessarily through the deliberate wisdom layer. For example, seeing a friend's beautiful photos on social media and instantly having the thought "I want to go there too." In this case, it can be understood that external information (I) or perceived data (D) directly induced a purpose (P), corresponding to the path  I →P  or  D →P . However, such purpose are often impulsive and changeable, lacking the support of the wisdom layer, and may not be robust. Therefore, in the DIKWP model, more meaningful purpose are usually generated through interaction with the W layer, i.e., purposewith a certain cognitive depth.

Mathematically, we can understand purpose It  (for tourist  t  at a certain time) as a mapping from a cognitive state to a goal:  It:D,I,K,W→P . That is, at a certain moment, the tourist  t 's brain, starting from the current state of data, information, knowledge, and wisdom, outputs a goal  P  after internal calculation. The generation of purpose can be seen as the moment this mapping is activated.

The P-Dominated Path of Purpose

Once the purpose is formed, the purpose (P) node will dominate the subsequent cognitive process, embodied as a "P-dominated path." This means that subsequent information acquisition and knowledge application will revolve around achieving this P. For example, when a tourist generates the purpose (P) "self-driving trip to Xinjiang this autumn," their subsequent behavior might be: collecting Xinjiang self-driving tour guides based on the goal ( P→D / I ), organizing the obtained information into a plan (K), evaluating feasibility and value (W), and constantly revising to achieve the goal. This can be described as a series of paths driven by purpose:  P →I →K →W →P  (goal guides information, information forms a plan, wisdom layer evaluates and points back to the goal). Here,  P  appears at the beginning and the end, indicating that the goal is both the starting point and the endpoint throughout the process.

Therefore, a significant feature of the "P-dominated path" is its closed-loop nature: cognitive activities must ultimately feed back to satisfy P. This is different from preference paths, which mostly illustrate the sequence of information processing; purpose paths emphasize the closed loop of the final purpose more.

Another feature is proactiveness: when P dominates, the tourist's intake of information is no longer aimless but is carried out with clear questions and screening criteria. For example, browsing travel information without a travel purpose might just be casual browsing (a preference-driven information roaming), but with a clear purpose, the browsing behavior will become highly targeted (e.g., only focusing on Xinjiang routes, irrelevant information will be ignored). In this situation, it can be considered that the  P  node sends a "selective demand" signal to  I  and  D  (corresponding to  P ×I ,  P ×D  interactions), making the entire semantic space imprinted with the purpose.

Structural Differences between Preference and Purpose

Through the above analysis, we can compare the structural differences between preference and purpose in the DIKWP*DIKWP semantic space:

Different Network Positions:Preference is reflected in favoring certain paths within the subject's internal conceptual space (e.g., preferring K-layer processing), belonging to the "internal parameters" of the cognitive process. Purpose is a purpose signal triggered by wisdom in the conceptual space, and it penetrates the external semantic space to guide resource acquisition. In other words, preference mostly describes how the subject processes information, while purpose connects the subject and object, driving the subject to influence the objective environment (e.g., acquiring external data, taking action according to purpose).

Persistence vs. Immediacy:Preference is relatively stable. The preference patterns in one trip often continue past habits and do not change easily (e.g., a deep-travel enthusiast focuses on cultural knowledge learning in every trip). Purpose is phase-based and immediate. An purpose may fade after it is achieved, and purpose can be completely different in different contexts (this year's travel purpose may be completely unrelated to last year's). Therefore, preference can be seen as a long-term trait, and purpose is a short-term state.

Mode of Action:Preference works by influencing the weights of the cognitive process, such as determining that a person will first check materials then decide in 70% of cases, and only act impromptu in 30% of cases. Purpose works by setting goals and forcing resource focus, i.e., once an purpose exists, the cognitive process is adjusted to serve that goal. This is also reflected mathematically: preference is a gradual shift in path probabilities (not 0 or 1, a soft constraint), while purpose often makes certain paths almost certain to occur (e.g., with an purpose, one will definitely follow a P→...→P closed loop, a hard goal constraint).

Influence on Behavior:Preference more so affects the content and style of choice, while purpose directly determines whether to take action and what action to take. For example, preference determines whether a tourist pays more attention to history and culture or natural scenery when making a travel plan (different information focus points), but purpose determines whether they will actually carry out this trip (whether there is a willingness to travel). Literature shows that preference will positively promote tourism consumption behavior by influencing tourism demand; while purpose is widely regarded as the direct prerequisite for actual behavior. In tourism behavior theories (such as the Theory of Planned Behavior, TPB), the predictive role of purpose on behavior is very significant.

Relationship with DIKWP Layers:Preference mainly involves the combination of D/I/K/W layers, especially the K and W layers, because preference is closely related to values, knowledge, and experience. Purpose is the representation of the P layer (goal), triggered by the W layer, and guides the activities of D/I/K. It can be said that preference is a set of tendencies projected "downward" from the W layer (e.g., W-layer values make one prefer a certain knowledge acquisition path), while purpose is a clear goal "upwardly" condensed from the W layer, which then "downwardly" affects other layers.

Table 1 summarizes the main differences between preference and purpose in the semantic network:

Dimension	Tourist Preference	Tourist Purpose
Definition	A stable tendency for information processing and transformation, favoring specific cognitive paths.	A goal or plan in a specific context, driving the cognitive process towards a specific outcome.
Location	A trait within the subject's interior (conceptual space), reflected in the internal path weights of the DIKWP network.	A target signal triggered by the subject, running through the conceptual space and the external semantic environment (P-layer node).
Time Scale	Long-term stable (like personality, attitude), continuous across contexts.	Short-term dynamic (like a thought, plan), can be regenerated when the context changes.
Mode of Action	Subtly influences information selection and processing order (soft constraint, probabilistic tendency).	Explicitly commands resource allocation and action direction (hard constraint, goal-oriented).
Typical Path	e.g.,  D→I→K→P ,  K→D→P ,  I→W→P  preference patterns.	e.g.,  W→P  (wisdom triggers goal),  P→D / I  (goal-driven information collection).
Impact Result	Affects decision quality and content bias (e.g., preference for culture leads to more museums in the itinerary).	Determines whether a behavior occurs and the object of the behavior (whether to travel, where to go).
Layer-Association	Associated with D/I/K/W layers (especially the value orientation of K/W layers).	Mainly associated with the P layer (influenced by W, and affects D/I/K acquisition).

Through the above comparison, we can clearly see that preference and purpose are both distinct and closely related. Preference provides the background parameters for behavioral style and tendency, while purpose is the prerequisite motive for specific behaviors to occur. In a tourism recommendation system, if both the user's preference and current purpose can be captured, it is expected to greatly improve the accuracy and satisfaction of the recommendation. For example, the system can distinguish: "This user consistently prefers deep cultural experiences" (preference) and "He is currently planning a short weekend trip" (purpose), thus recommending products that match his long-term interests and fit his current purpose.

After understanding the essence and differences between preference and purpose, we will next build a model that combines the recognition and prediction of both, to be used for the prediction and recommendation of tourism decisions.

Tourist Preference-Purpose Recognition Model and Tourism Recommendation Mapping

Based on the foregoing theory, we propose the Tourist Preference-Purpose Recognition Model, which aims to simultaneously capture a user's long-term preferences and short-term purpose within a smart tourism system, and map them as the system's decision-making inputs, thereby optimizing recommendation results. This model comprehensively utilizes the DIKWP semantic network and behavioral patterns revealed by meta-analysis, and includes the following main modules:

Model Framework Overview

User Preference Recognition Module: Estimates the user's preference path distribution  Pre f t ( p )  in the DIKWP network by analyzing the user's historical behavior data, preference questionnaires, social media content, etc. Techniques such as collaborative filtering, topic modeling, and inverse reinforcement learning can be used to extract preference features from data. The output of preference recognition can be represented by a vector or tensor, indicating the user's weights on several key path patterns. For example,  w A , w B , w C corresponding to the user's strength on the aforementioned A/B/C three types of path preferences.

User Purpose Detection Module: Uses contextual information to identify the user's current travel purpose It  (goal). Contextual information may include the user's search keywords on the platform, clicked destinations, booking behaviors, and even external information such as approaching holidays. Natural language processing and purpose recognition algorithms can be used to extract the target from the user's retrieval content (e.g., parsing the user's query "Is Zhangjiajie suitable for the National Day holiday?" implies their purpose is to travel to Zhangjiajie during the National Day). The result of purpose detection is described using semantic representation for the P node and related constraints (e.g., Destination=Zhangjiajie, Time=National Day, Preference=Natural Scenery).

DIKWP Semantic Reasoning Module: This is the core reasoning part of the model. It embeds the user purpose obtained in the previous step into the DIKWP network as the activated node of the P layer; simultaneously, it introduces the user preference distribution as the prior weight for each transformation relationship in the network. On this basis, the model simulates the decision-making paths the user might take and evaluates the utility of different paths. This reasoning can be formalized as a path selection optimization problem on a directed graph: finding the optimal path that matches the user's preference and can achieve the current purpose. For example, for a user with the purpose "Zhangjiajie National Day Tour," if their preference is "deep knowledge acquisition" (Preference A:  D→I→K→P ), the system will infer that they might first read a large amount of Zhangjiajie guides ( D→I ), then organize them into itinerary knowledge ( I→K ), and finally execute the booking ( P ). Accordingly, the recommendation engine will prioritize providing detailed guide information, deep-tour itinerary designs, and other services to match their decision-making path.

Tourism Recommendation Decision Module: Generates a personalized recommendation list based on the inferred cognitive path and utility evaluation. This includes destination recommendations, attraction and route recommendations, hotel and transportation suggestions, etc. When generating recommendations, it must consider both the user's preference tendency (e.g., if they prefer history and culture, recommend more cultural attractions) and ensure it meets the user's current purpose (e.g., if the user's goal is a "National Day short trip," the recommendation should be limited to itineraries that can be completed within the holiday). This module can integrate multiple recommendation algorithms, such as content-based recommendation to match preferences, collaborative filtering to discover potential interests, and constraint satisfaction-based filtering to remove options that do not meet the purpose conditions.

Feedback and Learning Module: After the system presents the recommendation plan to the user, it obtains the user's feedback (clicks, bookings, reviews, etc.) and uses it to update the user model. Here, the DIKWP model provides a transparent feedback mapping mechanism: for example, if the user ultimately did not choose the originally recommended Zhangjiajie but went elsewhere, it indicates the identified purpose might have been biased or changed midway; if the user browsed all the cultural attractions in the recommended list, it indicates the preference judgment was correct, etc. Through this feedback, the preference weight distribution  Pre f t ( p )  and the understanding of purpose can be dynamically corrected, thus being more accurate in subsequent interactions. The semantic structure of DIKWP also facilitates explanation: the system can record "The user actually took the I→K→P path in this decision, while we predicted D→I→K→P, there was an extra jump from information directly to knowledge," and adjust model parameters accordingly.

Comparison with Traditional Models

Traditional tourism recommendation systems mostly use user preferences (points of interest, historical behavior) for personalization, but their acquisition of the user's current specific purpose is limited. Some improved context-aware recommendations consider factors like time and place, but still lack a clear characterization of the user's purpose itself. The advantage of this model is that it combines preference (long-term interest) and purpose (short-term purpose) in a unified semantic network, enabling the system to both "remember who the user is" and "guess what the user wants to do."

For example, in the Theory of Planned Behavior (TPB) framework, attitude (similar to preference), subjective norms, and perceived control influence behavioral purpose (purpose), which in turn affects behavior. Our model is somewhat similar to TPB, but more refined and computable: preference corresponds to the user's internal attitude tendencies and values (reflected in the path preferences of the DIKWP knowledge/wisdom layers), while purpose is explicitly modeled as the activation of the Purpose node. The two are connected through the DIKWP network. For example, the mediating mechanism of attitude (knowledge) influencing behavior through purpose (purpose), as mentioned earlier, is embodied in our model as the path of K→W stimulating P, and then through P→...→action. This point is also supported by empirical evidence—Tian-Tian Li et al. (2021) found that tourism motivation (similar to preference drive) influences tourism preference through tourism attitude, thereby affecting the final behavior choice. And our semantic model makes these relationships more intuitively represented as a series of node-edge connections.

Model Case Example

Let's illustrate the model's operation with a specific scenario:

User Profile: Assume user Xiao Zhang is a "post-90s" office worker. He has traveled independently multiple times, and his preference leans towards deep experiences (trying to cover historical, cultural, and natural knowledge aspects as much as possible each time he visits a place), and he dislikes superficial sightseeing. His preference weights in the system can be characterized as favoring paths like  D→I→K→P  (representing step-by-step comprehensive understanding) and  I→W→P  (representing focusing on extracting insights from information).

Purpose Detection: Now, the National Day holiday is approaching, and Xiao Zhang has a travel purpose: to plan a 3-4 day short-distance self-driving tour, with the goal of relaxing and gaining knowledge. Through Xiao Zhang's recent searches and conversations on the platform, the system detects his purpose is locked on a "historical and cultural themed surrounding self-driving tour." This can be formalized as: P node="National Day Self-driving Tour," with additional attributes  Theme:Culture,Radius:<500km .

Model Reasoning: The model reasoning module implants this purpose into Xiao Zhang's DIKWP cognitive network. Combined with his preferences, it finds that the decision-making path most in line with his characteristics might be:

Step 1 ( D→I ): Collect data on possible destination candidates and convert it into information, such as browsing a list of "historical and cultural famous cities within 500km" (D screening under purpose constraints; preference drives him to browse information comprehensively).

Step 2 ( I→K ): Organize and compare the collected information to form a knowledge structure, such as listing the number of historical sites, cuisine, and user reviews of several candidate cities (typical information-to-knowledge transformation, consistent with his rational preference).

Step 3 ( K→W ): Use experience and value judgments to evaluate which city is more worth visiting, considering holiday crowds, interest matching, etc., elevating to a wisdom judgment (knowledge-to-wisdom transformation; Xiao Zhang might refer to past experiences or others' travelogues at this step).

Step 4 ( W→P ): Make the final decision, clarifying the purpose as "National Day self-driving tour to City A's ancient town," completing the purpose refinement and preparing for action.

Recommendation Support: Based on this reasoning, the recommendation module will provide support for each step: Step 1 provides interactive destination screening tools or maps (to meet his need for comprehensive data); Step 2 provides structured comparison tables or multi-dimensional evaluations; Step 3 pushes past tourists' experience sharing and precautions (to help his wisdom-based decision); finally, it directly provides route planning, attraction ticket booking, and other one-stop services for the self-driving tour to City A.

Throughout the process, the system both utilizes Xiao Zhang's preference (e.g., providing detailed information and comparisons to satisfy his habit of in-depth understanding) and responds to Xiao Zhang's purpose (locking onto a surrounding cultural tour, revolving around this goal from beginning to end). In this way, the recommendation not only makes the user feel personalized, but also thoughtful and efficient.

Through this case, we see how the DIKWP preference-purpose model guides the workflow of an intelligent tourism system. Next, we will turn to a review and meta-analysis of related empirical studies to verify the assumptions about the roles of preference and purpose in the model, and to provide a basis for the model parameters.

Meta-Analysis of Literature in the Last Decade: Empirical Evidence on Tourism Preference and Purpose

To test and enrich the theoretical model described above, we conducted a systematic review and meta-analysis of empirical studies in Chinese and English literature from 2015 to 2025 involving "tourist preference, behavioral purpose, and the application of DIKW/DIKWP models." By quantitatively synthesizing the findings of different studies, we can answer: How much influence do preference and purpose each have on tourism decision-making/behavior? What is their relationship? What other potential factors moderate this influence? Below, we introduce the meta-analysis method and main results.

Meta-Analysis Method Overview

Literature Search: We searched Chinese databases such as CNKI and Wanfang Data, and English databases such as Web of Science and Scopus, using keywords like "tourist preference," "tourism purpose," "tourism decision model," and "DIKW tourism," with the time limit set from January 2015 to June 2025. These search terms cover topics such as tourist motivation and preference, satisfaction and loyalty, behavioral purpose prediction, and smart tourism. An initial set of over 300 articles was obtained, with about 120 in Chinese and 180 in English. After screening and excluding theoretical discussion-types and articles without quantitative data, 95 empirical studies were finally included (45 Chinese, 50 English), with a total sample size of over 50,000 participants, covering different countries and regions, different types of tourism (cultural tourism, ecotourism, rural tourism, etc.), and different research focuses (e.g., motivation-behavior, satisfaction-purpose, etc.).

Effect Size Extraction: We used the correlation coefficient  r  as the main effect size indicator. For each study, we extracted statistical effects related to "preference or purpose," such as: "correlation coefficient between tourist satisfaction and revisit purpose," "structural equation path coefficient between tourism motivation and tourist loyalty," etc. If the literature did not directly provide the correlation coefficient, we converted it to an  r  value using formulas based on the reported  t  value, regression  β  value, or chi-square, etc. For multivariate studies, we might extract multiple pairs of correlations (e.g., motivation-loyalty, satisfaction-loyalty, etc.), recorded as independent effect sizes. In the end, a total of 280 independent effect sizes were obtained from the 95 articles.

Model Integration: A random-effects model was used to integrate various types of effect sizes separately. We classified the effect sizes according to variable relationships, for example: (a) Preference/Motivation factors vs. Loyalty/Willingness; (b) Satisfaction vs. Loyalty/Willingness; (c) Destination Perception/Value, etc. vs. Willingness/Behavior, etc. For each category, the combined correlation coefficient  r  and its 95% confidence interval were calculated, and a significance test was performed.

Heterogeneity Test: The  Q  statistic and  I 2  indicator were calculated to assess the heterogeneity among studies.  I 2  reflects how much of the total variation comes from true differences rather than sampling error. Generally,  I 2 >50%  indicates moderate or higher heterogeneity. We conducted heterogeneity tests on the effect sizes of each relationship. If significant heterogeneity existed, we explored possible moderating factors (such as cultural background, research method, tourism type, etc.).

Publication Bias Test:Funnel plots and the Fail-Safe N method were used to assess publication bias. The funnel plot observes symmetry through a scatter plot of effect size against standard error; Fail-Safe N calculates how many unpublished null-result studies would be needed to make the combined effect lose significance. For example, if the Fail-Safe N is much larger than a certain threshold (e.g.,  5n+10 , where  n  is the number of studies), it indicates the result is robust and not easily affected by bias.

The above steps were implemented using Comprehensive Meta-Analysis (CMA) software and the R language metafor package. The main analysis results are reported next.

Overall Effect Size and Relationship Strength

After integration analysis, we obtained a series of average correlation strengths between preference, purpose, and related variables in the tourism context. Table 2 shows some of the key results (due to space limitations, only the parts most relevant to this study's model are listed here):

Table 2. Meta-analysis Combined Effect Sizes (Correlation Coefficient  r ) of Tourism Preference/Purpose Related Factors

Independent Variable (Influencing Factor)	Dependent Variable (Outcome)	Effect Size  r  (95% CI)	Significance (Two-tailed)
Tourism Motivation/Preference	Tourist Loyalty	0.42 (0.32 ~ 0.51)	p < 0.001
Tourism Motivation (Push Motivation)	Tourist Loyalty	0.38 (0.13 ~ 0.59)	p = 0.004
Tourism Motivation (Pull Motivation)	Tourist Loyalty	0.50 (0.31 ~ 0.66)	p < 0.001
Tourist Satisfaction	Tourist Loyalty	0.64 (0.60 ~ 0.67)	p < 0.001
Perceived Value (Value for money, etc.)	Tourist Loyalty	0.53 (0.49 ~ 0.58)	p < 0.001
Perceived Quality (Destination quality)	Tourist Loyalty	0.47 (0.39 ~ 0.55)	p < 0.001
Experience Quality (Tour experience quality)	Tourist Loyalty	0.54 (0.43 ~ 0.63)	p < 0.001
Tourism Motivation/Preference	Revisit Purpose (Behavioral Purpose)	0.37 (0.11 ~ 0.58)	p = 0.007
Tourist Satisfaction	Revisit Purpose (Behavioral Purpose)	0.59 (0.54 ~ 0.64)	p < 0.001
Perceived Value	Revisit Purpose	0.46 (0.32 ~ 0.58)	p < 0.001
Tourist Satisfaction	Recommendation Purpose (Behavioral Purpose)	0.58 (0.50 ~ 0.64)	p < 0.001
Perceived Value	Recommendation Purpose	0.46 (0.32 ~ 0.58)	p < 0.001

From the above results, we can see:

Preference/Motivation factors have a moderate positive impact on tourist loyalty (loyalty usually includes a combination of revisit purpose and recommendation purpose), with a combined correlation of about  r ≈0.42 . This means that overall, the stronger a tourist's travel motivation (or the clearer their preference), the higher their loyalty to the destination (e.g., more willing to visit again or recommend to others). Both push motivations (internal drivers, e.g., escape, seeking relaxation) and pull motivations (destination attractiveness) are positively correlated, but pull motivations have a higher correlation.

Satisfaction has the strongest impact on loyalty, with a combined correlation as high as  r = 0.636 . This is a robust result obtained from 140 sample points, indicating that the more satisfied tourists are with their travel experience, their loyalty (revisit and recommendation) significantly increases. This is consistent with tourism theory expectations: satisfied tourists are more likely to revisit and spread positive word-of-mouth. The correlation of satisfaction with revisit purpose is also strong,  r ≈0.59 , and similarly for recommendation purpose,  r ≈0.58 .

The correlations of Perceived Value and Perceived Quality with loyalty and purpose are between  r=0.45~0.53 , which is a moderate-to-strong level. These factors can be regarded as mediating variables for the effects of preference and purpose: the sense of value and quality tourists obtain at the destination, when enhanced, will increase their satisfaction, thereby raising loyalty and purpose.

Experience Quality (subjective evaluation of the overall experience quality) has a correlation of about  r=0.536  with loyalty, indicating that high-quality tourism experiences can significantly cultivate repeat customers. This also explains why products that emphasize experience quality, such as deep tours and customized tours, help retain customers in the long run.

The above results confirm some key assumptions in our model: Satisfaction (a result of the Wisdom W layer) is one of the decisive factors in forming loyalty/behavioral purpose (P layer); Motivation/Preference (a driver from the K layer) indirectly acts on behavioral purpose by influencing satisfaction and attitude. For example, the empirical study by Yang et al. (2024) found that "tourism preference has a positive impact on tourism consumption through the mediating role of tourism demand (which can be seen as a manifestation of motivation)." This corresponds to our model path: Preference (K layer) -> generates some form of demand/motivation (D/I layer) -> leads to an increase in consumption behavior (P layer).

Heterogeneity and Moderator Analysis

The meta-analysis shows that heterogeneity among different studies is high. The Q-test for almost all main effects reached significance ( p<0.001 ), and the  I 2  index was mostly in the high range of  85%~98% . For example, the satisfaction-loyalty relationship  I 2 ≈97.9% , motivation-loyalty  I 2 ≈95.0% , indicating that the differences in correlation coefficients among studies far exceed what sampling error could cause.

High heterogeneity reflects the influence of differences in research context and methods on the results. We further analyzed possible moderating variables:

Cultural Background: Dividing studies into "Chinese tourists" and "non-Chinese tourists" based on sample source, we found the satisfaction-loyalty correlation was slightly higher in Chinese samples ( ~0.66 ) and slightly lower in Western samples ( ~0.60 ). The preference/motivation-loyalty correlation was also slightly stronger in Eastern cultural backgrounds. This might be related to Eastern tourists placing more emphasis on loyalty derived from overall satisfaction, as well as being driven by internal motivations, which is worthy of further exploration.

Tourism Type: Studies focusing on natural scenery tourism had a higher satisfaction-loyalty correlation than those on cultural/urban tourism, suggesting that satisfaction with natural products is more easily converted into revisit purpose. The effects also differed between first-time visitors and repeat visitors: for first-time visitors, the role of preference/motivation was more obvious (as their choice to go vs. not go depends mainly on interest), while in the repeat visitor group, satisfaction played a decisive role (as whether to visit again depends on the last experience).

Differences in Measurement Indicators: Some studies use a composite loyalty index, while others split it into revisit purpose and recommendation purpose. We saw that the average correlation of motivation with "recommendation purpose" was weaker (0.18~0.22), while it was slightly higher for "revisit purpose." This might be because recommendation depends more on satisfaction and personal communication tendencies, while revisiting is directly driven by motivation/preference.

In short, heterogeneity reminds us that the mechanism of preference and purpose will vary with the context. Therefore, when applying the model, contextual factors should be considered, such as adjusting weights for different customer groups. This study has partially controlled for heterogeneity through the random-effects model, and also reports the confidence intervals of the results to reflect uncertainty.

Publication Bias Test

The funnel plot shows that large-sample studies often reported medium effects with high precision, while the results of small-sample studies were more dispersed, but the overall distribution was fairly symmetrical, with no obvious missing areas. However, the result from the Fail-Safe N method is more robust: for key relationships like satisfaction-loyalty, the Fail-Safe  N  was as high as  1152450  (based on 140 effect sizes), far exceeding the safety threshold. This means that millions of undiscovered null-effect studies would have to emerge to possibly make this combined effect lose its significance. For other relationships like motivation-loyalty, the Fail-Safe  N=5449 , and for satisfaction-revisit purpose, it was  135612 , all of which are very large, indicating our conclusions are quite robust.

Egger's regression test also found no significant small-sample bias. This gives us confidence in the meta-analysis results: the influence of preference and purpose-related variables on tourism behavior is real and not exaggerated due to publication bias.

Synthesizing the main findings of the meta-analysis, the following key conclusions can be drawn:

Satisfaction (the result of the Wisdom layer) is one of the most important factors in promoting behavioral purpose and loyalty. Its effect strength is significant and robust in almost all contexts. This supports the key path in our model: "Wisdom (W) → Purpose (P)": only when tourists experience satisfaction at the wisdom/value level will they generate a strong purpose to travel again (P).

Preference/Motivation (the driver of the Knowledge layer) has a moderate effect on behavioral purpose and loyalty, but it is not as direct as satisfaction. This suggests that the role of preference is more often realized indirectly by influencing satisfaction, attitude, etc. This also aligns with our semantic model's view that preference does not directly determine behavior, but influences the final purpose through the cognitive process.

The conversion rate of purpose (e.g., revisit purpose converting to actual revisiting), although not directly covered by this meta-analysis, is widely assumed in the literature to be the most direct antecedent of behavior. In our model, reaching the P layer does not guarantee that the behavior will definitely occur, but P-layer activation is a necessary condition. This also conforms to classic theory—without purpose, there will hardly be any action; with purpose, there is usually a higher probability of action (subject to constraints like opportunity and other objective conditions).

Contextual factors will lead to differences in model parameters. For example, in scenarios dominated by one-time sightseeing, preference may not play a major role, and on-the-spot emotional purpose dominate behavior. In deep-tour or repeat-tour scenarios, preference accumulation and word-of-mouth satisfaction will more significantly affect long-term behavior. Our model should allow for adjusting path weights according to the context (this can be achieved by adjusting the weights of different edges in the DIKWP network in different contexts).

The meta-analysis provides quantitative support for our theoretical model: it verifies that both preference and purpose do have an impact on tourism behavior (and their modes of action are different), and it also points out the bridging role of satisfaction in connecting preference and purpose. The next section will discuss how to integrate these findings with the DIKWP semantic model and propose suggestions for improving intelligent tourism systems.

Discussion: Fusing Semantic Path Analysis and Statistical Conclusions

Combining the above meta-analysis results, we can provide a more in-depth interpretation of the previously constructed DIKWP preference-purpose model and propose implications for intelligent tourism systems.

Empirical Verification of Preference-Purpose Paths

The meta-analysis shows that the correlation "Satisfaction -> Revisit Purpose" is extremely high. In the DIKWP model, this corresponds to the importance of the Wisdom (W) -> Purpose (P) path. The satisfaction evaluation at the wisdom layer almost becomes a sufficient condition for generating the travel purpose again. If a trip makes a tourist highly satisfied in terms of value and emotion, their internal wisdom layer will strongly drive the purpose "I want to come again." This verifies that  W→P  in our model is a key pathway for purpose formation: without the positive support of the W layer, the activation of the P layer (such as the plan to revisit) will not be strong or lasting.

On the other hand, the correlation "Motivation/Preference -> Revisit Purpose" is positive but moderate, indicating that preference tendency itself will not directly translate into behavioral purpose, unless it is reinforced in the actual experience. For example, a tourist who loves beaches (preference) does not necessarily go on a beach vacation every year (behavioral purpose), unless his last beach trip was very pleasant (high satisfaction, leading to purpose). Therefore, in our model, preference more so indirectly affects purpose by influencing the wisdom layer's evaluation: preference determines his choice of a beach destination (K layer acting on D/I layer choice). Next, if the experience is good, the W layer adds points, thereby stimulating a strong purpose; if the experience is poor, then although the preference remains, he will not have the purpose to go to the beach again in the short term.

This relationship is also supported by some mediation model studies. For example, the results of Tian-Tian Li et al. (2021) show that Tourism Motivation -> Attitude -> Preference influence in stages. Applying this, we can infer that Preference -> Satisfaction -> Purpose may be a more accurate causal chain. Our model needs to capture this chain. In a recommendation system, this means we cannot just infer that a user definitely wants to do something based on preference; we must also consider their satisfaction level with similar past behaviors (experience).

Application of DIKWP×DIKWP Interaction

The  DIKWP×DIKWP  interaction model proposed by Yucong Duan provides a framework for handling subjective-objective cognitive interaction. In our context, the tourist's brain conceptual space can be regarded as one DIKWP system, and the tourism environment/platform's semantic space as another DIKWP system. The interaction between the two could be, for example: the interaction between tourist wisdom and platform data ( W×D ), the interaction between tourist purpose and platform information ( P×I ), etc.

Using this idea, a tourism recommendation system can be designed for dual-space collaboration: on one hand, modeling the user's internal DIKWP state (their preferences, knowledge, current purpose, etc.), and on the other hand, modeling the DIKWP structure of the product content (what data, information, knowledge, endowed wisdom value, and satisfied purpose the tourism product itself has). The recommendation process is one of continuous interaction and fusion between the two DIKWP spaces.

For example, in a conversational recommendation:

User's Conceptual Space: P layer activates "want to go to a beautiful island with few people."

Platform's Semantic Space: D layer retrieves data on a series of island destinations (list of islands, tourist volume, etc.).

User's Conceptual Space: W layer evaluates which options might meet "few people, beautiful scenery" based on past knowledge (K) (wisdom discrimination, eliminating non-compliant ones,  W×D  interaction).

Platform's Semantic Space: Based on feedback, provides more detailed information at the I layer (e.g., ratings, photos of each island).

User's Conceptual Space:  I→K  absorbs information, forms a clearer preference ranking, and then possibly  W→P  adjusts purpose details (e.g., locking onto two islands for comparison).

This process repeats until the user confirms the final decision in the conceptual space P layer; the platform records the user's booking or action in the semantic space P layer (purpose graph), achieving a  P×P  closure.

This two-way interpretable collaborative process makes the reasons for recommendation transparent: why a certain option was proposed at each step, and why the user rejected or accepted it, can find corresponding node matches or conflicts on the DIKWP semantic graph. For example, if a user rejects a certain island plan, it can be explained as "Their wisdom layer determined that this plan could not satisfy their purpose of 'few people' (W-layer value mismatch)," which is clearly more interpretable than a black-box collaborative filtering.

Technical Suggestions for Smart Tourism Systems

Based on the theoretical construction and empirical analysis of this study, we propose the following suggestions to promote the application of the DIKWP model in intelligent tourism systems:

Construct Users' DIKWP Profiles: Not only record users' preference tags, but also characterize their knowledge background, value orientation (Wisdom layer), and typical purposes. For example, establish user profiles in the five dimensions of D, I, K, W, and P through questionnaires or long-term behavioral analysis. In a sense, this is a more comprehensive semantic profile than traditional user profiles: knowing what tourism data the user has (which places D they have been to), what tourism information they follow (which topics I they are concerned about), what knowledge and experience they possess (K), what values they cherish (W), and what kind of travel goals they often set (P). Some research has begun to explore similar ideas, such as using large models to analyze white papers, applying DIKWP concepts to AI system human-computer fusion. For tourism platforms, a "Tourism Knowledge Graph + User Purpose Model" combination can be developed, encompassing users' explicit and implicit preferences.

Real-time Purpose Perception: Introduce conversational interfaces or smart assistants to directly ask or guide out the current purpose during interaction with the user. For example, large language models like ChatGPT can converse with users to obtain information such as "What do you most want to achieve on this trip?" Once the purpose is captured, the system should switch to an purpose-oriented mode and dynamically adjust the recommendation strategy (similar to our P-dominated path). In practice, users can be allowed to select their travel purpose (leisure/visiting relatives/adventure, etc.) on the APP, or infer the purpose by analyzing the user's current browsing content. Multimodal data (text, voice, behavioral sequences) fusion will improve the accuracy of purpose recognition.

Develop Preference-Purpose Fusion Recommendation Algorithms: Traditional recommendation algorithms can be combined with knowledge graphs and reinforcement learning to adapt them to the DIKWP structure. For example, add an purpose matching constraint on top of collaborative filtering rating prediction: only perform prediction ranking among items whose destination attributes match the user's current purpose. Another example is applying reinforcement learning, letting the agent consider both immediate rewards (satisfaction from matching the current purpose) and long-term rewards (loyalty from conforming to long-term preferences) to find a balanced strategy. Existing tourism recommendation research has begun to distinguish the impact of preference-oriented and goal-oriented recommendation algorithms on user purchase purpose. The results show that recommendations targeting the user's current goal can significantly increase purchase purpose. This confirms the importance of fusing the purpose dimension.

Enhance System Interpretability: Use the DIKWP model to generate explanations for recommendation results. For example, when the system recommends a certain route, it can explain: "Because you indicated that you want to achieve <P> on this trip, we have provided a plan containing <K>, and attached rich <I> for you to understand the details." Here, <P>, <K>, <I>, etc., can be replaced with specific purpose, knowledge points, and information types, allowing the user to understand how the recommendation meets their goals and matches their interests. This explanation is based on semantic paths, e.g., "You prefer to make decisions through in-depth understanding ( I→K ), we have prepared detailed materials to help you form a judgment." Research shows that explanatory recommendations can enhance user trust in the system and decision satisfaction. The DIKWP model provides a natural template for explanation.

Continuous Learning and Adaptability: As the meta-analysis revealed, the preference-purpose relationship will change in different contexts. Therefore, the system should be able to adjust itself according to environmental changes. For example, during peak holiday periods, user purpose may be more affected by external constraints (e.g., popular attractions are overcrowded, leading to purpose changes). The system needs to dynamically perceive this change and give more consideration to situations where preference gives way to realistic constraints. In addition, evaluate recommendation effects through A/B testing, offline simulations, etc., and continuously correct model parameters (such as edge weights). Since the elements of the DIKWP model are clear, this parameter tuning can be targeted—for example, if the user click-through rate is low, is it due to insufficient information provision (I-layer problem) or incorrect purpose understanding (P-layer problem)?

Pay Attention to Negative Purpose and Value Preferences: In addition to positive guidance, the system must also guard against potential problems. Tourism decision-making involves safety, ethics, and other factors. The "Wisdom W layer" in the DIKWP model should integrate these value constraints. For example, if a user inadvertently reveals an unrealistic or risky purpose (e.g., exploring a restricted area at night), the system should use the wisdom layer rules to give reminders or even dissuasion. This can be seen as the detection and correction of moral purpose deviation. Tourism enterprises have a responsibility to use AI to ensure tourist safety and compliance with public interests, which cannot be ignored in intelligent system design.

All in all, the DIKWP preference-purpose model points out a direction for improvement for smart tourism systems with semantics as the link. From a theoretical perspective, it unifies the previously separate studies on personality preferences and behavioral purpose into one framework; from a practical perspective, it provides a blueprint for building a tourism AI that "understands both what you like and knows what you seek." In the future, we look forward to seeing more verification of this model in practical applications, such as comparing the performance of recommendation systems that introduce the purpose factor with traditional systems through controlled experiments, or examining the effects of customized strategies for users with different preference types. This will further enrich and perfect the DIKWP model.

Conclusion

This study, centered on the theme "Mathematical Definition and Meta-Analysis Modeling of Tourist Preferences and Purpose in DIKWP*DIKWP Semantic Space," has conducted in-depth theoretical exploration and empirical verification. The main contributions and conclusions are as follows:

We provided clear semantic mathematical definitions for tourist preference and tourist purpose based on Yucong Duan's DIKWP network interaction model. Preference is defined as the stable path transformation selection that tourists exhibit in the DIKWP cognitive network, quantifiable as a preference weight distribution on specific directed paths. Purpose is defined as a goal-oriented process in the DIKWP network, triggered by wisdom and centered on the purpose node, embodied in  W→P  or P-dominated interaction modes. Their ontological difference lies in: preference is a long-term cognitive tendency (focusing on internal biases at the knowledge and value levels), while purpose is an immediate behavioral motive (focusing on explicit drivers at the purpose level).

We constructed a Tourist Preference-Purpose Recognition Model, applying the above definitions to the framework of intelligent tourism recommendation systems. The model explains how to extract preference semantics from user data, how to perceive the user's current purpose, and how to fuse the two for decision support through DIKWP semantic reasoning. Case analysis shows that this model is expected to improve the precision, relevance, and interpretability of recommendations, as it can simulate the "interest + purpose" dual-wheel-drive cognitive mechanism in human decision-making.

Through a systematic meta-analysis of related empirical studies in the last decade, we verified the impact of preference and purpose on tourism behavior. The comprehensive results show: tourist preference/motivation has a moderate impact on loyalty and behavioral purpose, while tourist satisfaction has an extremely strong impact on behavioral purpose and loyalty (correlation coefficient as high as 0.6 or more). This is consistent with our model's inference—preference more so indirectly promotes the formation of subsequent purpose by improving experience satisfaction (wisdom layer), and the purpose brought by high satisfaction (such as the thought of revisiting) is the most decisive. The meta-analysis also revealed the heterogeneity of effect sizes in different contexts, suggesting that model application needs to consider the moderation of factors such as culture and tourism type.

We combined the semantic model and statistical evidence to discuss improvement strategies for intelligent tourism systems. Suggestions include: establishing DIKWP user profiles, identifying user purpose in real-time, developing recommendation algorithms that fuse preference and purpose, using semantic paths to provide explanations, continuous learning and adaptation, and paying attention to safety and ethics. These suggestions provide a reference for the next generation of smart tourism systems, making them more "tourist-centric," truly understanding the inner thoughts and feelings of tourists, rather than just speculating based on superficial behavior.

Theoretical Significance: This research enriches the understanding of the tourist decision-making cognitive process in the field of tourism management. In the past, preference and purpose were mostly discussed separately in different theories. Our DIKWP model incorporates both into a unified semantic network, which has a certain ontological innovation. It provides a comprehensive framework for studying tourist behavior: from data to purpose, factors in the entire chain are considered, and their circular interaction relationships are emphasized. This is inspiring for advancing tourism behavior theories (such as integrating the Technology Acceptance Model, Theory of Planned Behavior, etc.).

Practical Significance: This model can guide intelligent recommendation and marketing strategies. Tourism enterprises can use it to segment users more precisely (e.g., grouping by preference path type), capture business opportunities more timely (pushing promotions when user purpose is high), and enhance user stickiness and trust through explanations. At the same time, the ideas of data transparency and purpose sovereignty advocated by the DIKWP model also fit the emphasis on data ethics and user sovereignty in the current digital tourism era—allowing users to understand and control how AI serves them.

Of course, this study also has some limitations. First, although we have defined the mathematical models for preference and purpose, accurately estimating an individual's preference distribution and real-time purpose in a real system is still challenging and requires high-quality data support and algorithm optimization. Second, our meta-analysis focuses on macro-level correlations and does not delve into the causal chain and psychological mechanisms of preference and purpose, which can be verified by experimental research in the future. In addition, the application of the DIKWP model in the tourism field is still in its nascent stage. Whether the model needs to be customized according to the peculiarities of tourism (e.g., introducing concepts like "experiencing self/narrating self") is worthy of further study.

In conclusion, the DIKWP semantic model provides a promising new paradigm for understanding and predicting tourist behavior. By ontologically distinguishing and integrating the two key elements of tourist preference and purpose within the same framework, we can better explain the differences in past empirical results and guide the innovative design of practical systems. Looking forward, with the development of tourism big data and artificial intelligence technology, we have the opportunity to collect more comprehensive DIKWP element data (e.g., wearable devices capturing real-time experience feelings—W layer data), verify more refined model hypotheses (e.g., the input-output ratio of different preference paths), and thus continuously improve this model. With the joint efforts of academia and industry, the tourist preference-purpose semantic model is expected to become an important tool for enhancing user experience and industry efficiency in the era of smart tourism.

References:

The main literature and data sources cited in this research report are as follows (partial):

Duan, Y., et al. (2025). Semantic Mathematical Artificial Consciousness Reconstruction of "Experiencing Self" and "Narrating Self" Based on the Networked DIKWP Model. (Professor Yucong Duan's paper on the DIKWP model, elaborating the theoretical basis and application of the DIKWP model).

Wang, L., & Li, X. (2023). The five influencing factors of tourist loyalty: A meta-analysis. PLOS ONE, 18(4): e0283963. (A meta-analysis study on the influencing factors of tourist loyalty, providing effect sizes for satisfaction, motivation, etc., with loyalty/purpose).

Song, Z., et al. (2024). A study on the influencing factors of youth tourism preferences. Modern Management, 14(12): 3010-3018. (Focuses on the factors of youth group tourism preferences, sorting out the relationship between preference, motivation, and consumption behavior).

Xuan, W., & Chang, L. (2023). Tourist preference learning based on tour behavior and inverse reinforcement learning. Journal of Guilin University of Electronic Technology, 43(3): 173-180. (Learning tourists' fine-grained preferences through IoT data and IRL algorithms).

Duan, Y. (2023). DIKWP Artificial Consciousness Model (Principles). DAMA China. (Details the definitions and interrelationships of data, information, knowledge, wisdom, and purpose in the DIKWP model).

The publication of the above results in high-level tourism management journals (such as Tourism Tribune, etc.) will contribute to the academic community's understanding of the mechanism of tourist behavior and provide guiding insights for the practice of the smart tourism industry. We believe that with the deepening of research, the DIKWP semantic model will show strong vitality in tourism and other human-computer interaction fields.

人工意识与人类意识

人工意识日记

玩透DeepSeek：认知解构+技术解析+实践落地

人工意识概论：以DIKWP模型剖析智能差异，借“BUG”理论揭示意识局限

人工智能通识 2025新版 段玉聪 朱绵茂 编著 党建读物出版社

主动医学概论 初级版

世界人工意识大会主席 | 段玉聪

邮箱|duanyucong@hotmail.com

世界人工意识科学院

邮箱 | contact@waac.ac