Tourism e-commerce big data mining method based on artificial intelligence

Abstract

The invention belongs to the technical field of tourism e-commerce, and particularly relates to an artificial intelligence-based tourism e-commerce big data mining method, which comprises the following steps of: obtaining user behaviors, cross-platform product supply, historical transaction characteristics and social public opinion emotion, and constructing a tourism interest dynamic evolution graph; quantizing a time sequence drift and community diffusion rule of user interests through a time sequence diagram neural network model, constructing a supply-demand matching degree prediction model based on cross-platform product supply and user behavior data, and combining with the updated evolution graph and through a generative adversarial network to obtain a supply-demand matching degree prediction model. And generating a personalized tourism product recommendation list and a dynamic pricing strategy, carrying out intelligent diversion and resource pre-distribution, synchronously collecting user interaction feedback and market conversion efficiency data, and through an online incremental learning algorithm, optimizing a time sequence diagram neural network model, and generating a tourism demand mining and market response evaluation report. Therefore, the problems of insufficient user interest dynamic feature capture, weak user demand mining ability and the like in the prior art are solved.

Description

Technical Field

[0001] This invention belongs to the field of tourism e-commerce technology, specifically relating to an artificial intelligence-based method for big data mining in tourism e-commerce. Background Technology

[0002] With the deep integration of the digital economy and the tourism industry, tourism e-commerce has entered a new stage of precision operation. The market urgently needs intelligent mining technology based on multi-source data, requiring capabilities such as dynamic perception of user interests, cross-platform supply and demand matching, and rapid response to trending events. Currently, tourism e-commerce data mining often focuses on single-dimensional data (such as historical transactions or isolated user behavior). While this can achieve basic recommendation functions, it lacks deep integration of multi-source heterogeneous data, including cross-platform product supply, social sentiment, and historical transaction characteristics. This makes it difficult to depict the temporal drift patterns and community diffusion characteristics of user interests, and to accurately capture the demand shifts triggered by sudden trending events. There is a significant gap between this and the technological requirements for building a dynamic and personalized tourism service system.

[0003] Traditional tourism big data mining methods suffer from key flaws: data dimensions are limited to a single source, failing to form a cross-dimensional perception network encompassing "behavior-supply-public opinion," resulting in one-sided user interest profiles and insufficient accuracy in supply-demand matching; models lack dynamic adjustment mechanisms, lagging parameter updates in response to the evolution of user interests over time, community dissemination, and the impact of trending events, making it difficult to adapt to rapid fluctuations in tourism demand; simultaneously, they lack online incremental learning capabilities, failing to continuously optimize models based on real-time user feedback and market conversion data, leading to insufficient market adaptability of recommendation lists and pricing strategies. With the upgrading of tourism consumption and intensified cross-platform competition, the market's demand for highly accurate, responsive, and adaptable intelligent mining technologies is increasingly urgent. However, existing technologies, due to insufficient data integration, weak dynamic adaptability, and lagging iterative optimization, are insufficient to support efficient operation and decision-making in complex tourism scenarios. Summary of the Invention

[0004] This application provides an artificial intelligence-based big data mining method for tourism e-commerce to address the problems of insufficient capture of dynamic user interest features and weak user demand mining capabilities in existing technologies.

[0005] The first aspect of this application provides an artificial intelligence-based big data mining method for tourism e-commerce, comprising the following steps: acquiring user behavior data, cross-platform product supply data, historical transaction feature data, and social media sentiment data; constructing a dynamic evolution map of tourism interests based on the historical transaction feature data and the social media sentiment data, quantifying the temporal drift and community diffusion patterns of user interests through a time-series graph neural network model, and simultaneously using an anomaly detection algorithm to identify interest leaps caused by sudden hot events, dynamically updating the weight parameters of the evolution map; constructing a supply-demand matching degree prediction model based on the cross-platform product supply data and the user behavior data, combining the dynamically updated dynamic evolution map of tourism interests, mining potential tourism demand and predicting regional product supply-demand gaps through a generative adversarial network, generating a personalized tourism product recommendation list and dynamic pricing strategy; performing intelligent traffic diversion and resource pre-allocation based on the personalized tourism product recommendation list and dynamic pricing strategy, simultaneously collecting user interaction feedback data and market conversion efficiency data, and optimizing the time-series graph neural network model based on the interaction feedback data and conversion efficiency data through an online incremental learning algorithm, generating a tourism demand mining and market response evaluation report.

[0006] Preferably, a dynamic evolution map of tourism interests is constructed based on the historical transaction feature data and the social media sentiment data, including: constructing a multi-source heterogeneous tourism interest dataset that integrates historical transactions and social media sentiment; extracting structural features from the multi-source heterogeneous tourism interest dataset using a graph neural network and fusing features using a cross-modal attention mechanism to generate a user-interest feature matrix; and constructing a dynamically weighted interest network based on the user-interest feature matrix using a graph attention mechanism and combining it with a time series analysis algorithm to extract the spatiotemporal evolution path and community propagation pattern of user interests to generate a dynamic evolution map of tourism interests.

[0007] Preferably, the temporal drift and community diffusion patterns of user interests are quantified using a temporal graph neural network model, including: constructing a temporal graph neural network model; inputting the dynamic evolution map of tourism interests into the temporal graph neural network model, and generating the temporal drift coefficient and community diffusion intensity of user interests through joint calculation using the temporal attention mechanism and graph diffusion mechanism in the model.

[0008] Preferably, the anomaly detection algorithm is used to identify interest shifts triggered by sudden hot events and dynamically update the weight parameters of the evolution graph, including: constructing an anomaly detection algorithm; based on the anomaly detection algorithm, real-time monitoring of the social media sentiment data, identifying abrupt changes in the popularity of interest topics, and determining the abrupt changes as sudden hot events; in the dynamic evolution graph of tourism interests, locating users and interest nodes associated with the sudden hot events, and extracting user interest subgraphs composed of these nodes and their connecting edges; using a graph structure time-series comparison algorithm, calculating the structural metric changes of the user interest subgraph before and after the occurrence of the sudden hot event, the structural metrics including changes in node centrality and edge weights; and adjusting the influence weights of corresponding nodes and the propagation probability parameters of edges in the dynamic evolution graph of tourism interests according to the structural metric changes.

[0009] Preferably, a supply-demand matching degree prediction model is constructed based on the cross-platform product supply data and the user behavior data, including: acquiring cross-platform product supply data and user behavior data; performing multi-dimensional analysis and vectorization on the cross-platform product supply data to generate standardized product supply feature vectors, and simultaneously performing sequence modeling and feature extraction on the user behavior data to generate user behavior feature vectors; concatenating or cross-calculating the product supply feature vectors and user behavior feature vectors through a deep matching learning algorithm, and performing deep fusion through a multi-layer nonlinear network, and constructing a training sample set based on historical user-product interaction records, using the prediction of interaction probability as the objective function, and iteratively optimizing the neural network using the backpropagation algorithm to complete the construction of the supply-demand matching degree prediction model.

[0010] Preferably, based on the interactive feedback data and conversion efficiency data, the time-series graph neural network model is optimized using an online incremental learning algorithm to generate a tourism demand mining and market response assessment report, including: constructing an online incremental learning algorithm; using the interactive feedback data and conversion efficiency data as real-time monitoring signals based on the online incremental learning algorithm to incrementally optimize the time-series graph neural network model; and generating a tourism demand mining and market response assessment report based on the optimized time-series graph neural network model.

[0011] A second aspect of this application provides an artificial intelligence-based big data mining system for tourism e-commerce, comprising: an acquisition module for acquiring user behavior data, cross-platform product supply data, historical transaction feature data, and social media sentiment data; a construction module for constructing a dynamic evolution graph of tourism interests based on the historical transaction feature data and the social media sentiment data, quantifying the temporal drift and community diffusion patterns of user interests through a time-series graph neural network model, and simultaneously using an anomaly detection algorithm to identify interest transitions caused by sudden hot events, dynamically updating the weight parameters of the evolution graph; and a prediction module for predicting the data based on the cross-platform product supply data and... The user behavior data is used to construct a supply-demand matching prediction model. Combined with the dynamically updated tourism interest evolution map, a generative adversarial network is used to mine potential tourism demand and predict regional product supply-demand gaps, generating a personalized tourism product recommendation list and dynamic pricing strategy. The generation module is used to perform intelligent traffic diversion and resource pre-allocation based on the personalized tourism product recommendation list and dynamic pricing strategy, and simultaneously collect user interaction feedback data and market conversion efficiency data. Based on the interaction feedback data and conversion efficiency data, the time-series graph neural network model is optimized through an online incremental learning algorithm to generate a tourism demand mining and market response evaluation report.

[0012] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement an artificial intelligence-based big data mining method for tourism e-commerce as described in the above embodiments.

[0013] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement an artificial intelligence-based big data mining method for tourism e-commerce as described in the above embodiments.

[0014] The fifth aspect of this application provides a computer program product, including a computer program or instructions, for implementing an artificial intelligence-based big data mining method for tourism e-commerce as described in the above embodiments.

[0015] Therefore, this application has the following beneficial effects:

[0016] This application's embodiments integrate user behavior data, cross-platform product supply data, historical transaction characteristic data, and social media sentiment data to construct a dynamic evolution map of tourism interests. This map comprehensively covers user interest dimensions and, combined with a time-series graph neural network model, accurately quantifies the temporal drift patterns and community diffusion characteristics of user interests. Simultaneously, it relies on anomaly detection algorithms to capture interest shifts triggered by sudden trending events in real time and dynamically update the map weights, avoiding the problem of recommended content being out of sync with real-time user needs and market trends. Based on this, a supply-demand matching prediction model built from cross-platform product supply and user behavior data, combined with generative adversarial networks, can not only accurately predict regional tourism product supply-demand gaps but also deeply mine potential needs not explicitly expressed by users. This generates personalized tourism product recommendation lists tailored to user needs and flexible dynamic pricing strategies, achieving efficient matching between supply and demand. Furthermore, intelligent traffic redirection and resource pre-allocation optimize tourism resource allocation efficiency. Simultaneously, online incremental learning algorithms continuously optimize the time-series graph neural network model based on user interaction feedback and market conversion efficiency data, improving user decision-making efficiency and platform market conversion capabilities. This provides tourism e-commerce platforms with operational support that dynamically adapts to market changes and significantly reduces ineffective resource consumption. This solves the problems of insufficient capture of dynamic features of user interests and weak ability to mine user needs in existing technologies.

[0017] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0019] Figure 1 A flowchart illustrating an artificial intelligence-based big data mining method for tourism e-commerce, according to an embodiment of this application;

[0020] Figure 2 This is a schematic diagram illustrating the construction of a dynamic evolution map of tourism interests according to an embodiment of this application;

[0021] Figure 3 This is a schematic diagram illustrating the construction of a supply-demand matching degree prediction model according to an embodiment of this application;

[0022] Figure 4 This is a schematic diagram of an artificial intelligence-based big data mining method for tourism e-commerce according to an embodiment of this application;

[0023] Figure 5 This is a schematic diagram of the structure of a tourism e-commerce big data mining system based on artificial intelligence, according to an embodiment of this application.

[0024] Figure 6 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0025] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0026] The following describes an embodiment of an AI-based big data mining method for tourism e-commerce, with reference to the accompanying drawings. Addressing the problem of insufficient capture of dynamic user interest features mentioned in the background art, this application provides an AI-based big data mining method for tourism e-commerce. In this method, by integrating user behavior data, cross-platform product supply data, historical transaction feature data, and social media sentiment data, a dynamic evolution graph of tourism interests is constructed, comprehensively covering the dimensions of user interests. Combined with a time-series graph neural network model, it accurately quantifies the temporal drift patterns and community diffusion characteristics of user interests. Simultaneously, relying on anomaly detection algorithms, it captures interest leaps triggered by sudden hot events in real time and dynamically updates the graph weights, avoiding the problem of recommended content being out of sync with real-time user needs and market trends. Based on this, a supply-demand matching prediction model constructed based on cross-platform product supply and user behavior data, combined with generative adversarial networks, can not only accurately predict regional tourism product supply-demand gaps but also deeply mine potential needs not explicitly expressed by users. This generates personalized tourism product recommendation lists tailored to user needs and flexible dynamic pricing strategies, achieving efficient matching between supply and demand. Furthermore, by optimizing the efficiency of tourism resource allocation through intelligent traffic redirection and resource pre-allocation, and simultaneously leveraging online incremental learning algorithms to continuously optimize the time-series graph neural network model based on user interaction feedback and market conversion efficiency data, the efficiency of user decision-making and the platform's market conversion capabilities are improved. This provides tourism e-commerce platforms with operational support that dynamically adapts to market changes and significantly reduces the consumption of ineffective resources. Thus, it solves the problems of insufficient capture of dynamic interest features and weak user demand mining capabilities in existing technologies.

[0027] Specifically, Figure 1 This is a flowchart illustrating an artificial intelligence-based big data mining method for tourism e-commerce, provided as an embodiment of this application.

[0028] like Figure 1 As shown, this AI-based big data mining method for tourism e-commerce includes the following steps:

[0029] In step S101, user behavior data, cross-platform product supply data, historical transaction feature data, and social sentiment data are acquired.

[0030] It is understood that the embodiments of this application, by acquiring user behavior data, cross-platform product supply data, historical transaction characteristic data, and social media sentiment data, form the core data foundation for subsequent big data mining work in tourism e-commerce. User behavior data can intuitively present users' dynamic preferences such as real-time browsing, collection, and inquiries, providing direct evidence for accurately capturing immediate tourism demand trends; cross-platform product supply data can integrate tourism resource information from multiple channels, breaking the limitations of incomplete resource coverage on a single platform and ensuring the integrity of supply-side data; historical transaction characteristic data can mine users' long-term consumption habits, price sensitivity, and category preferences, providing historical reference for analyzing the evolution of interests; and social media sentiment data can capture market hotspots and user sentiment trends in real time, helping to quickly respond to demand changes caused by sudden hotspots.

[0031] In step S102, a dynamic evolution map of tourism interests is constructed based on historical transaction feature data and social sentiment data. The temporal drift and community diffusion patterns of user interests are quantified through a time-series graph neural network model. At the same time, an anomaly detection algorithm is used to identify interest shifts caused by sudden hot events, and the weight parameters of the evolution map are dynamically updated.

[0032] Among them, the tourism interest dynamic evolution map refers to a structured association carrier that integrates historical transaction feature data and social sentiment data. It is used to integrate the core association information between users and interests, providing an analytical basis for the time series graph neural network model to quantify the time series drift pattern and community diffusion characteristics of user interests. At the same time, it supports the dynamic updating of weight parameters after the interest jump is identified through anomaly detection algorithms, providing dynamic data support for the subsequent accurate mining of tourism demand.

[0033] It is understood that this application's embodiments construct a dynamic evolution map of tourism interests as a structured core carrier integrating historical transaction feature data and social sentiment data. This integrates the correlation information between users and interests, providing a precise analytical foundation for time-series graph neural network models to quantify the temporal drift patterns and community diffusion characteristics of user interests. Simultaneously, it supports anomaly detection algorithms to identify dynamic updates of weight parameters after interest transitions, effectively carrying out dynamic data association and parameter iteration, avoiding the lag of traditional static analysis. This provides dynamic and highly adaptable data support for subsequent tourism demand mining and supply-demand matching, ensuring the real-time nature and accuracy of the mining results.

[0034] For example, a travel e-commerce platform uses a dynamic evolution graph of travel interests for its users aged 25-40, primarily young working professionals. The platform first integrates historical transaction data (such as multiple bookings of weekend getaways and family-themed hotels over the past year, with spending concentrated in the mid-to-high-end market) with social media sentiment data (recent surge in popularity of topics like "urban hiking" and "weekend camping aesthetics," and frequent mentions of "niche routes" and "lightweight gear" in user social media sharing). This constructs a structured association graph, linking users to interest nodes such as "nearby trips," "family trips," and "light hiking." Through a time-series graph neural network model, the platform quantifies the temporal drift trend of the group's interests from "family trips" to "light hiking," while also capturing the diffusion characteristics of interests formed through friend sharing and community discussions—multiple users added the "light hiking" interest tag due to recommendations from their social circles. When a suburban hiking route suddenly becomes a hotspot due to online celebrity check-ins, the anomaly detection algorithm quickly identifies the event, and the graph updates the weights of "light hiking" related nodes in real time. Based on the updated graph, the platform accurately recommends matching niche hiking routes, lightweight camping equipment, and nearby characteristic homestays to target users, significantly improving user click and conversion efficiency and completing a closed loop from interest capture to precise service.

[0035] In this embodiment, a dynamic evolution map of tourism interests is constructed based on historical transaction feature data and social sentiment data. This includes: constructing a multi-source heterogeneous tourism interest dataset that integrates historical transactions and social sentiment; extracting structural features from the multi-source heterogeneous tourism interest dataset using a graph neural network and fusing features using a cross-modal attention mechanism to generate a user-interest feature matrix; and constructing a dynamically weighted interest network based on the user-interest feature matrix using a graph attention mechanism and combining it with a time series analysis algorithm to extract the spatiotemporal evolution path and community propagation pattern of user interests to generate a dynamic evolution map of tourism interests.

[0036] Graph neural networks (GNNs) are deep learning models specifically designed for processing graph-structured data. They can automatically learn the feature representations of nodes (such as users and points of interest) and edges (relationships) in a graph, thereby capturing complex relational information. The formula is as follows:

[0037]

[0038]

[0039] in, Let L be the node feature matrix of the l-th layer; This is the structured feature matrix output by the (l+1)th layer; For activation functions; (This is the normalized adjacency matrix). Let be the trainable weight matrix of the l-th layer; l is the number of layers in the graph neural network; It is an adjacency matrix with self-loops; for The degree matrix; The operator for finding the square root of the inverted matrix.

[0040] It should be noted that cross-modal attention mechanism refers to a neural network component used to fuse features from heterogeneous data sources. It achieves adaptive weighted fusion of multi-source information by calculating the correlation weights between features from different modalities. The formula is as follows:

[0041]

[0042]

[0043] in, This is the output of the cross-modal attention mechanism; K is the normalization function; Q is the query vector; K is the key vector; (where K is the transpose of the key vector K). Let K be the dimension of the key vector. V is the square root of the dimension of the key vector; V is the value vector. This is the historical transaction mode feature matrix; This is a social media sentiment modality feature matrix. Let M1 be a trainable weight matrix that maps to Q; Let M2 be a trainable weight matrix that maps to K; M1 is a trainable weight matrix that maps to a portion of V; M2 is a trainable weight matrix that maps to a portion of V; This is the feature splicing function.

[0044] Graph attention mechanism refers to an algorithm that introduces attention weights into graph neural networks to adaptively calculate the importance of connections between nodes in the graph, thereby completing dynamic weighted information aggregation. The formula is:

[0045]

[0046]

[0047]

[0048] in, For nodes in the graph For nodes The original attention coefficient; For attention scoring functions; The weight matrix is ​​trainable. For nodes eigenvectors; For nodes eigenvectors; These are the normalized attention weights; For nodes Neighboring nodes The function to be normalized; It is an exponential function; For nodes The set of neighboring nodes; For nodes For neighboring nodes The original attention coefficient; For the aggregated nodes eigenvectors; This is the activation function.

[0049] Time series analysis algorithms refer to a class of statistical or machine learning methods used to analyze data sequences arranged in chronological order. They aim to extract time-series evolution patterns such as trends, cycles, seasonality, and abrupt changes from the data. The formula is:

[0050]

[0051]

[0052] in, for Order difference operator; It is the difference order; for Time series data values ​​at any given moment; For timestamps; for time Time series data after order difference; For constant terms; The order of the autoregressive term; For the first One autoregressive coefficient; for time Time series data after order difference; The order of the moving average term; For the first One moving average coefficient; for The random error term at time step; for The random error term at time step.

[0053] It is understood that the graph neural network in this application, as the core for processing multi-source heterogeneous tourism interest data, can mine the relationship between users and interest nodes and extract structured features, laying the foundation for subsequent feature fusion. The cross-modal attention mechanism focuses on the key correlations between historical transaction and social media sentiment data, weakens irrelevant interference, and efficiently fuses features to generate an accurate user-interest feature matrix. The graph attention mechanism constructs a dynamically weighted interest network by adaptively calculating the connection weights between nodes, ensuring the real-time nature and importance distinction of network associations. The time series analysis algorithm extracts the spatiotemporal evolution path and community propagation pattern of user interests from a time-series dimension, capturing the changing patterns of interests over time. The four methods work together to avoid the problems of weak data associations, poor modal fusion, static networks, and lack of time-series analysis in traditional data mining, providing full-process support for the accurate construction of a dynamic evolution map of tourism interests.

[0054] For example, such as Figure 2 As shown, when constructing a dynamic evolution graph of tourism interests, a certain tourism e-commerce platform first builds a multi-source heterogeneous tourism interest dataset that integrates historical transactions and social media sentiment. From historical transaction feature data, it selects users who have "booked mountain hiking routes or forest homestays three or more times in the past six months." Simultaneously, it captures comments and topics posted by this group on various social media platforms, such as "wanting to experience stargazing camping after hiking" and "camping equipment purchasing guide." The user's transaction records and corresponding social media posts are associated by user ID to form a multi-source heterogeneous dataset. Next, based on this dataset, a graph neural network is used to extract the connection structure features between users and interest tags such as "mountain hiking" and "stargazing camping." At the same time, a cross-modal attention mechanism is used to strengthen the association weight between "multiple bookings of hiking products" and "frequent discussions of camping," while weakening interference from irrelevant information. This generates a user-interest feature matrix that includes the user's preference strength for different interests and the correlation between interests. Finally, based on this matrix, a graph attention mechanism is used to assign dynamic weights to interest-related edges such as "mountain hiking - stargazing camping" and "camping - outdoor equipment" (e.g., the weight of the corresponding edge increases when the discussion popularity of "hiking + camping" rises). Combined with time series analysis algorithms, the evolution path of users' interests from "single hiking" to "hiking + camping" is sorted out, as well as the dissemination pattern of this interest combination among young outdoor enthusiasts, and finally a dynamic evolution map of tourism interests is generated.

[0055] In this embodiment of the application, the temporal drift and community diffusion patterns of user interests are quantified by a temporal graph neural network model, including: constructing a temporal graph neural network model; inputting the dynamic evolution map of tourism interests into the temporal graph neural network model, and generating the temporal drift coefficient and community diffusion intensity of user interests through joint calculation by the temporal attention mechanism and graph diffusion mechanism in the model.

[0056] Among them, the temporal graph neural network model refers to a deep learning architecture specifically designed for dynamic graph data. By jointly modeling the spatiotemporal dynamic changes of the graph structure, it quantifies the temporal drift pattern of user interests and the intensity of community diffusion from the input temporal graph sequence. The formula is:

[0057]

[0058] ,

[0059]

[0060]

[0061]

[0062]

[0063] in, The input is a sequence of time-series graphs; This represents the total number of time steps. For time step The graph adjacency matrix; This represents the total number of nodes in the graph; For time step The node feature matrix; Symbols for the set of real numbers; For node feature dimensions; For time step time step Temporal attention weights; It is an exponential function; It is a multilayer perceptron; For time step The hidden layer features of the nodes; The node features are obtained after temporal attention fusion; For time step index; For time step The normalized adjacency matrix; Let be the degree matrix at time step t; The node features after graph diffusion; For activation functions; The weight matrix for the graph diffusion layer can be trained; For time step node The timing drift coefficient; For time step node The intensity of community diffusion, For nodes The set of neighboring nodes, for The Line 1 Column elements.

[0064] It is understood that the embodiments of this application use a temporal graph neural network model to jointly model the spatiotemporal dynamic changes of the graph structure and capture user-interest relationships at different time steps. Simultaneously, it leverages the joint computation of temporal attention and graph diffusion mechanisms to generate the temporal drift coefficient and community diffusion intensity of user interests, covering the spatiotemporal dimensions of interest evolution. This avoids the limitations of traditional single static graph analysis or temporal analysis, providing direct support for accurately quantifying the temporal drift and community diffusion patterns of user interests, and facilitating the effective implementation of a dynamic evolution graph of tourism interests.

[0065] For example, a short-distance travel platform inputs a 4-week dynamic evolution graph of travel interests (including weekly user-interest association data) into a temporal graph neural network model. The model uses a temporal attention mechanism to calculate the attention weight (up to 0.72) of the "family travel" feature in week 1 to the "light hiking" feature in week 4, capturing the temporal association of user interests shifting from family travel to light hiking. Simultaneously, through a graph diffusion mechanism, it calculates the adjacency weight of the "light hiking" interest within the user community, obtaining its community diffusion strength (value 1.8). Finally, it generates a temporal drift coefficient (0.65) for the shift from family travel to light hiking and the community diffusion strength for light hiking. Based on this, the platform pushes light hiking routes + community group packages to users with drifting interests, increasing the weekly click-through rate of this type of product by 42%.

[0066] In this embodiment, an anomaly detection algorithm is used to identify interest shifts triggered by sudden hot events and dynamically update the weight parameters of the evolution graph. This includes: constructing an anomaly detection algorithm; based on the anomaly detection algorithm, real-time monitoring of social media sentiment data to identify abrupt changes in the popularity of interest topics and classify these abrupt changes as sudden hot events; locating users and interest nodes associated with sudden hot events in the dynamic evolution graph of tourism interests and extracting user interest subgraphs composed of these nodes and their connecting edges; calculating the structural metric changes of the user interest subgraph before and after the sudden hot event using a graph structure time-series comparison algorithm, where the structural metric includes changes in node centrality and edge weights; and adjusting the influence weights of corresponding nodes and the propagation probability parameters of edges in the dynamic evolution graph of tourism interests based on the structural metric changes.

[0067] Among them, anomaly detection algorithms refer to statistical or machine learning methods used to identify deviations from normal patterns in data. These algorithms are used to monitor sudden shifts in social media sentiment in real time to identify emerging trending events. The formula is as follows:

[0068]

[0069]

[0070] in, The Z-score standardized result of the popularity value of the topic of interest at time t is used to quantify the degree of deviation between the popularity at that time and the historical normal popularity. The real-time popularity value of the topic of interest at time t; This is at a historically normal level of heat. The length of the sliding window; This refers to the historical time period from time tw to time t-1. This represents the normal fluctuation range of historical popularity. For timestamps; The anomaly flag value at time t; This is an abnormal flag value; This is the threshold for anomaly detection; This is an abnormal flag value; For other situations.

[0071] It should be noted that the graph structure time-series comparison algorithm is a calculation method used to quantify the structural differences of dynamic graphs between adjacent time segments. It quantifies the structural changes in the graph by comparing metrics such as node centrality and edge weights of the user interest subgraphs before and after hot events. The formula is:

[0072]

[0073]

[0074] in, This refers to the degree of change in the strength of the correlation between nodes before and after a sudden hot topic event; The centrality value of node i after a sudden hotspot event occurs; The centrality value of node i before the occurrence of a sudden hotspot event; This refers to the degree of change in the strength of the correlation between nodes before and after a sudden hot topic event; The weight of the edge between node i and node j after a sudden hotspot event occurs; The weight of the edge between node i and node j before the occurrence of a sudden hot event; To illustrate the degree of difference in subgraph structure before and after a sudden hot topic event; The total number of nodes in the user interest subgraph; The set of nodes for the user interest subgraph; The value representing the change in centrality of node i; The total number of edges in the user interest subgraph; The set of edges for the user interest subgraph; For the edge The weight change value.

[0075] Understandably, in this embodiment, the anomaly detection algorithm monitors social media sentiment data in real time, identifies sudden changes in the popularity of interest topics and determines them as sudden hot events. It can quickly capture key fluctuation signals related to user interests and avoid the impact of missing sudden hot events on the evolution of tourism interests due to delayed identification. The graph structure time-series comparison algorithm calculates the changes in node centrality and edge weights of relevant user interest subgraphs in the dynamic evolution graph of tourism interests before and after the sudden hot event, quantifies the time-series differences in the subgraph structure, and provides a precise basis for adjusting the node influence weights and edge propagation probability parameters in the graph. The two types of algorithms work together to support the dynamic optimization of the dynamic evolution graph of tourism interests.

[0076] In step S103, a supply-demand matching prediction model is constructed based on cross-platform product supply data and user behavior data. Combined with the dynamically updated tourism interest evolution map, a generative adversarial network is used to mine potential tourism demand and predict regional product supply-demand gaps, generating a personalized tourism product recommendation list and dynamic pricing strategy.

[0077] Generative Adversarial Networks (GANs) are deep learning frameworks that simulate real-world data distributions by training a generator and a discriminator against each other. In this method, they are used to generate and mine unexpressed tourism demands based on users' latent interest features, thereby predicting regional product supply and demand gaps. The formula is as follows:

[0078]

[0079]

[0080] in, For generator; The random noise vector is input to the generator; These are the trainable parameters of the generator; This is a vector concatenation symbol; The vector of potential user travel demands output by the generator; (The generator outputs a predicted value for the supply and demand gap of regional tourism products). For discriminators; These are the trainable parameters of the discriminator; The probability value output by the discriminator; This serves as an identifier that indicates "the data is real business data"; The total loss function of the generative adversarial network; For expected operators; This serves as an identifier for "x follows the distribution of real business data"; Distribution of real business data; Wrap the symbol in the result of the operation; The operator for natural logarithms; The distribution of random noise; This represents the discriminator's judgment result on the generator's output data.

[0081] It is understood that the embodiments of this application utilize generative adversarial networks to mine potential tourism demand and predict regional product supply and demand gaps. By using random noise, the generated potential demand is ensured to be diverse, avoiding limitation to only popular demands and covering more implicit user preferences. It also breaks the limitation of relying on historical data, generating logically consistent demands that have not appeared in history, thus reflecting the randomness of real-world tourism scenarios. Simultaneously, it provides support for the subsequent generation of personalized tourism product recommendation lists and dynamic pricing strategies, making the output more adapted to actual market demand.

[0082] For example, the smart tourism platform integrates cross-platform supply data from homestay alliances, scenic spot self-operated systems, and rural tourism cooperatives in southern Anhui, as well as user behavior data within its own app, such as searching for "niche villages" and "intangible cultural heritage experiences," collecting bamboo weaving videos, and planning weekend getaways, to construct a supply-demand matching prediction model. Combined with a dynamically updated tourism interest evolution map—which shows that users in surrounding cities like Hefei and Nanjing have recently shown a 58% increase in interest in "intangible cultural heritage experiences + rural homestays" compared to the previous month—the platform initiates generative adversarial network computation. The generator incorporates user behavior characteristics and random noise to uncover potential combined demand for "bamboo weaving experiences + mountain view homestays"; the discriminator compares the results with historical real order data to optimize performance, simultaneously predicting a 38% supply-demand gap for weekend homestays around Hongcun Village in Yixian County. The platform then pushes homestay packages including bamboo weaving courses and child-friendly facilities to families with children, increasing package prices by 22% during peak periods and offering a 25% discount on weekdays. In the first week of implementation, bookings for this type of package increased by 82%, and the overall occupancy rate of rural homestays in Yixian County rose from 65% to 91%, effectively balancing regional supply and demand.

[0083] In this embodiment, a supply-demand matching degree prediction model is constructed based on cross-platform product supply data and user behavior data. This includes: acquiring cross-platform product supply data and user behavior data; performing multi-dimensional analysis and vectorization on the cross-platform product supply data to generate standardized product supply feature vectors; simultaneously performing sequence modeling and feature extraction on the user behavior data to generate user behavior feature vectors; using a deep matching learning algorithm, concatenating or cross-calculating the product supply feature vectors and user behavior feature vectors, and deeply fusing them through a multi-layer nonlinear network; constructing a training sample set based on historical user-product interaction records; using the prediction of interaction probability as the objective function; and iteratively optimizing the neural network using a backpropagation algorithm to complete the construction of the supply-demand matching degree prediction model.

[0084] Among them, deep matching learning algorithm refers to a matching model based on deep neural networks. It predicts the matching probability or interaction tendency between user features and product features by learning the high-order nonlinear interaction relationship between them. The formula is as follows:

[0085]

[0086]

[0087] in, For users Tourism products The probability of supply and demand matching; The activation function for the output layer; This is the weight matrix of the first fully connected neural network layer; This is the weight matrix of the second fully connected neural network layer; This is the weight matrix of the third fully connected neural network layer; The hidden layer activation function introduces nonlinearity to capture the complex nonlinear relationship between users and product features; For feature splicing symbols; This represents a user behavior feature vector. Provide feature vectors for products; This is the bias term for the first layer of the neural network; This is the bias term for the second layer of the neural network; This is the bias term for the third layer of the neural network; Index for users; The cross-entropy loss function of the model; Total number of training samples For the training sample set; For real-world interaction tags; This is the operator for the natural logarithm.

[0088] It should be noted that backpropagation is a neural network training method based on gradient descent. It calculates the gradient of the loss function with respect to the parameters of each layer of the network, and updates the weights layer by layer along the reverse direction of the gradient to iteratively optimize the network's predictive ability. The formula is:

[0089]

[0090]

[0091] in, For loss function For parameters The gradient; The total number of training samples; For a single training sample; For the training sample set; For users Tourism products The probability of supply and demand matching; For real-world interaction tags; Used as an identifier for the number of layers in a neural network; This represents the layer number where the parameters used to calculate the gradient are located. For the first Transpose of the weight matrix of a layered neural network For the first Layer trainable weights; This is the matrix transpose symbol; For indicator functions; For the first Hidden layers and layer outputs; For parameters The input feature vector is transposed; It is a general term for the trainable parameters of a model; This is the learning rate.

[0092] It is understood that the embodiments of this application, through deep matching learning algorithms, combine platform product supply data and user behavior data to deeply fuse product feature vectors and user feature vectors, and capture the high-order nonlinear correlation between the two, accurately predicting the probability of user-product interaction; breaking the limitations of traditional linear matching, mining implicit needs not explicitly expressed by users, making the matching results more in line with personalized needs, providing a core basis for subsequent recommendation list generation, calculating the loss value between the predicted results and real interaction data through backpropagation algorithm, back-deriving the gradient of the influence of each layer parameter (weight, bias) on the loss, and iteratively updating parameters in combination with the learning rate, so that the model continuously corrects the deviation, avoids getting trapped in local optima, and can adapt to the dynamic changes of tourism data, ensuring the continuous stability of model prediction accuracy, and providing reliable algorithmic support for dynamic pricing strategies.

[0093] For example, such as Figure 3As shown, when a certain cultural tourism platform constructs a supply and demand matching prediction model, it first obtains cross-platform raw product data such as mountain homestays, natural scenic spots, and short-distance transportation from the national cultural tourism resource database, chain hotel booking system, and intercity ticketing platform, as well as user behavior data such as clicking on "family homestays," searching for "nature study tours," and collecting "handmade pottery experiences." In the product supply feature vector generation, the homestay data is broken down into price (380 yuan / night), location (near mountain scenic spots), type (family study tour homestays), and facilities (including children's playgrounds and pottery workshops). The data, including inventory (5 rooms available for the weekend) and rating (4.6 points), is used to create a 64-dimensional standardized vector by encoding product type with One-Hot encoding, price and rating with Min-Max normalization, and location information with latitude and longitude encoding. The user behavior feature vector is modeled using a Transformer model to depict the temporal behavior of "searching for nature study tours → browsing mountain homestays → collecting pottery experiences". After capturing the dependency relationship between behaviors of "study tour needs guiding accommodation selection", the user preference features are extracted through global average pooling to generate a 64-dimensional dense vector. Subsequently, the deep matching learning algorithm first cross-calculates the two types of vectors to generate interactive features of "parent-child needs - homestay facilities", and then inputs them into a three-layer nonlinear network. The first layer fuses basic features through matrix operations, the second layer uses the ReLU activation function to capture higher-order association features such as "nature study tours + hands-on experiences", and the third layer outputs the matching probability between users and products. At the same time, a sample set is constructed using historical records of "users booking marked as 1 / not booking marked as 0", and the cross-entropy between the predicted probability and the true label is used as the objective function. The backpropagation algorithm starts from the output layer, first calculates the gradient of the loss with respect to the weights of the third layer (i.e., the product of the prediction bias and the output of the second layer), and then backpropagates the gradients of the weights of the second and first layers and the biases of each layer layer by layer. Combined with a learning rate of 0.001, all parameters are iteratively updated. After 120 iterations, the model converges, and the final matching prediction accuracy reaches 81%, and the user click conversion rate of recommended products increases by 30%.

[0094] In step S104, intelligent traffic redirection and resource pre-allocation are carried out based on the personalized tourism product recommendation list and dynamic pricing strategy. User interaction feedback data and market conversion efficiency data are collected simultaneously. Based on the interaction feedback data and conversion efficiency data, the time series neural network model is optimized through an online incremental learning algorithm to generate a tourism demand mining and market response assessment report.

[0095] User interaction feedback data refers to the real-time behavioral records generated by individual users after receiving recommended content, which are used to directly measure their acceptance and interest in the recommended items.

[0096] It should be noted that market conversion efficiency data refers to a set of quantitative indicators compiled from the business level to evaluate the overall business effectiveness of recommendation and pricing strategies.

[0097] Understandably, this application's embodiments use user interaction feedback data to reflect user acceptance of the recommendation list and pricing strategy, providing user-side data for online incremental learning, making model optimization more aligned with needs, and improving subsequent recommendation accuracy. Market conversion efficiency data can assess the actual effects of intelligent traffic redirection and resource pre-allocation, providing market-side data for model optimization and reducing resource misallocation. Both together support model iteration and provide core data for evaluation reports, helping to clarify needs and market responses, ultimately improving user experience and market effectiveness.

[0098] For example, a certain platform, relying on a pre-established personalized travel product recommendation list and dynamic pricing strategy, rapidly advanced intelligent traffic redirection and resource pre-allocation. In the intelligent traffic redirection phase, the platform matched the reach method to the core needs tags of different users in the recommendation list: for users marked as "families with children" on the list, a pop-up window on the app's homepage pushed bundled products from the list, simultaneously sending SMS messages containing pricing discounts; for the "young backpackers" group on the list, an entry point to the list was embedded in their browsing history page, and targeted push notifications were sent reminding them to book routes with discounted prices; for the "health and wellness" group on the list, a link in a WeChat official account article guided them to a dedicated list page, highlighting the cost-effectiveness of the pricing. During resource pre-allocation, the platform uses the product popularity ranking and pricing strategies reflected in the recommended list to lock in resources with partner merchants in advance. 40% of the rooms are reserved for the top 3 most popular homestays on the list, 25% of scenic spot tickets are reserved for peak seasons with increased pricing, and 10% of resources are reserved for less popular products with reduced pricing based on estimated demand. A real-time monitoring module is also in place; if a product's bookings reach 80% of the reserved amount, an additional 20% is reserved; if bookings are less than 30%, some resources are released to the public pool. During this process, the platform simultaneously collects user interaction feedback data such as clicks on the list, product favorites, and order cancellations, as well as market conversion efficiency data such as booking completion rate, resource redemption rate, and revenue growth for each product. These two types of data are input into an online incremental learning algorithm, which rapidly optimizes the time-series neural network model through gradient updates, correcting the previous mismatch between traffic redirection rhythm and user activity periods. The resulting tourism demand mining and market response evaluation report clearly presents the market feedback effects of different group needs and pricing strategies, providing precise guidance for subsequent work.

[0099] In this embodiment, based on interactive feedback data and conversion efficiency data, an online incremental learning algorithm is used to optimize the time-series graph neural network model and generate a tourism demand mining and market response assessment report. This includes: constructing an online incremental learning algorithm; using the online incremental learning algorithm, interactive feedback data and conversion efficiency data as real-time monitoring signals to incrementally optimize the time-series graph neural network model; and generating a tourism demand mining and market response assessment report based on the optimized time-series graph neural network model.

[0100] Among them, the online incremental learning algorithm refers to a training method that allows machine learning models to continuously and efficiently update their parameters using newly arrived real-time data without retraining the entire network. The formula is as follows:

[0101]

[0102]

[0103] in, Let be the incremental loss function at time t; For time steps; Let t be the batch size of the new data at time t; Let be the true label of the i-th sample at time t; It is the natural logarithm function; Let be the model's predicted value for the i-th sample at time t; This is the weight decay coefficient; These are the model parameters updated at time t; The learning rate for incremental learning; This is the gradient operator for the parameters at time t-1; The momentum coefficient; Update the difference in parameters between the previous two time steps.

[0104] It is understood that the embodiments of this application use an online incremental learning algorithm with real-time user interaction feedback and market conversion efficiency data as supervision signals to incrementally update model parameters without retraining the entire time-series graph neural network model. This avoids the high computing power and time consumption problems of full retraining, reducing operating costs; it also allows the model to continuously adapt to the dynamic changes of user interest time-series drift and community diffusion, maintaining prediction accuracy; and it can quickly adapt to streaming market data, providing timely model support for tourism demand mining and strategy optimization, and improving market response efficiency.

[0105] For example, during cherry blossom season, a travel platform pushed customized routes combining cherry blossom viewing with visits to lesser-known ancient towns to its target users and implemented dynamic pricing. After pre-allocating resources through intelligent traffic redirection, the system simultaneously collected user feedback data showing a 25% increase in click-through rates for the routes but only a 5% increase in final order conversion rates, as well as market conversion efficiency data showing a 40% increase in occupancy rates for guesthouses near the routes while sales of traditional cherry blossom viewing packages declined. Based on this real-time data, the platform then incrementally optimized the temporal graph neural network model using an online incremental learning algorithm, enabling the model to more accurately identify that user interests had shifted from simply viewing flowers to immersive cultural experiences. The system then automatically correlates the structured data, such as the interest diffusion intensity and supply-demand gap prediction output by the optimized model, with real-time market conversion indicators. It also generates a spring cultural tourism integration demand mining report and a cherry blossom season market response assessment report, which include data charts and conclusions, through a preset report template. The report clearly points out that there is a 30% supply-demand gap in composite products combining natural landscapes and cultural heritage. It also shows that the dynamic pricing strategy boosted per capita consumption by 15% in the ancient town module, but caused traffic loss in the traditional cherry blossom viewing module due to price sensitivity. This guides the precise adjustment of product mix and pricing strategy in the next cycle.

[0106] According to an embodiment of this application, a big data mining method for tourism e-commerce based on artificial intelligence is proposed. By integrating user behavior data, cross-platform product supply data, historical transaction feature data, and social sentiment data, a dynamic evolution graph of tourism interests is constructed, which can comprehensively cover the dimensions of user interests. Combined with a time-series graph neural network model, the method accurately quantifies the temporal drift patterns and community diffusion characteristics of user interests. At the same time, relying on anomaly detection algorithms, it captures interest leaps caused by sudden hot events in real time and dynamically updates the graph weights, avoiding the problem of recommended content being out of touch with users' real-time needs and market trends. On this basis, a supply-demand matching degree prediction model based on cross-platform product supply and user behavior data, combined with generative adversarial networks, can not only accurately predict the supply-demand gap of regional tourism products, but also deeply mine potential needs that users have not explicitly expressed. This leads to the generation of personalized tourism product recommendation lists that meet user needs and flexible dynamic pricing strategies, achieving efficient matching between supply and demand. Furthermore, by optimizing the efficiency of tourism resource allocation through intelligent traffic redirection and resource pre-allocation, and simultaneously leveraging online incremental learning algorithms to continuously optimize the time-series graph neural network model based on user interaction feedback and market conversion efficiency data, the efficiency of user decision-making and the platform's market conversion capabilities are improved. This provides tourism e-commerce platforms with operational support that dynamically adapts to market changes and significantly reduces ineffective resource consumption. Thus, it solves the problems of insufficient capture of dynamic user interest features and weak user demand mining capabilities in existing technologies.

[0107] The following will illustrate a big data mining method for tourism e-commerce based on artificial intelligence through a specific embodiment, such as... Figure 4 As shown, it includes:

[0108] A certain travel e-commerce platform first integrates multi-dimensional data, specifically including user behavior sequences such as clicks, browsing, favorites, and bookings within the platform; cross-platform product supply information such as homestays, natural scenic spots, short-distance transportation, and cultural experience projects obtained from multiple cooperative channels; historical transaction characteristic data such as consumption amount, purchase frequency, and preferred product types reflected in past orders; and social sentiment data such as comments, likes, and sentiment tendencies on travel-related topics on various social media platforms. After cleaning and deduplication, this data forms the basic dataset for subsequent analysis.

[0109] Based on this data, the platform first constructs a multi-source heterogeneous tourism interest dataset that integrates historical transactions and social media sentiment. This dataset matches records of "multiple bookings of natural landscape products" in historical transactions with content in social media sentiment that "frequently mentions 'light outdoor activities + handicraft experiences'". Then, it extracts the structural features of users and interest tags in this dataset using a graph neural network. At the same time, it uses a cross-modal attention mechanism to strengthen the association weight between "historical booking behavior" and "social discussion content", generating a user-interest feature matrix. Next, it uses a graph attention mechanism to assign dynamic weights to the association edges between different interests (for example, the association weight between "light outdoor activities" and "handicraft experiences" increases with the popularity of the discussion). Finally, it combines time series analysis algorithms to trace the spatiotemporal evolution path of user interests from "single sightseeing" to "landscape + experience", as well as the dissemination pattern of a certain type of interest among similar consumer groups, thereby generating a dynamic evolution map of tourism interests. Subsequently, the platform built a temporal graph neural network model, input the evolutionary graph into the model, captured the changes in interest at different time points through the temporal attention mechanism, and analyzed the propagation path of interest in the user group by combining the graph diffusion mechanism. Finally, it generated the temporal drift coefficient of user interest (the probability of some users shifting from "single sightseeing" to "landscape + experience" within 30 days reached 65%) and the community diffusion strength (the weekly propagation rate of the interest "light outdoor + handicraft experience" in the target group was 22%). Meanwhile, the platform deploys anomaly detection algorithms to monitor social media sentiment data in real time. When it detects that the discussion volume of the topic "light outdoor activities + intangible cultural heritage handicrafts" has increased threefold in a short period of time, it is determined to be a sudden hot topic event. Subsequently, in the dynamic evolution graph of tourism interests, the platform locates users who are interested in this topic and their corresponding interest nodes, and extracts a user interest subgraph composed of these nodes and connecting edges. Through a graph structure time series comparison algorithm, the platform calculates the structural metric changes of this subgraph before and after the event—the centrality of relevant interest nodes increases by 40%, and the edge weight between "light outdoor activities" and "intangible cultural heritage handicrafts" increases by 25%. Based on this, the platform adjusts the influence weight of the corresponding nodes in the evolution graph, and at the same time increases the propagation probability parameters of relevant edges, completing the dynamic update of the evolution graph.

[0110] Next, the platform began constructing a supply-demand matching prediction model. First, it analyzed cross-platform product supply data from multiple dimensions, breaking it down into dimensions such as price, product type, supporting facilities, real-time inventory, and user ratings. These dimensions were converted into numerical forms by numerically processing inventory quantities, normalizing price ranges, and embedding encoded product types, and then concatenated into a standardized product supply feature vector of uniform length. Simultaneously, it performed sequence modeling on user behavior data, using a Transformer model to capture the behavioral dependencies from "searching for light outdoor activities → browsing handicraft experiences → collecting intangible cultural heritage projects," extracting deep features representing user interest intentions and preference intensity, and summarizing them into a fixed-dimensional user behavior feature vector. Then, the platform used a deep matching learning algorithm to cross-calculate the product supply feature vector and the user behavior feature vector, generating "user interest-product attribute" interaction features, which were then input into a multi-layer nonlinear network containing a ReLU activation function for deep fusion. At the same time, a training sample set was constructed based on historical user-product interaction records. Using the prediction of user-product interaction probability as the objective function, the backpropagation algorithm was used to iteratively optimize the neural network parameters, completing the construction of the supply-demand matching prediction model. The platform then inputs the dynamically updated tourism interest evolution map into the model, and combines it with a generative adversarial network to uncover potential tourism demand for "light outdoor activities + intangible cultural heritage handicrafts" that users have not explicitly expressed. It also predicts that the regional booking demand for this type of product exceeds the existing supply by 30%, forming a supply-demand gap. Based on this, the platform generates personalized tourism product recommendation lists for different users (recommending "light outdoor activities + intangible cultural heritage handicrafts" combination packages to users with high interest drift coefficients, and recommending upgraded landscape projects to users with traditional sightseeing preferences). At the same time, it formulates a dynamic pricing strategy—slightly increasing the price of combination packages with large supply-demand gaps (not exceeding 15%), and decreasing the price of traditional sightseeing products with sufficient inventory by 10% to improve conversion rates.

[0111] Based on this recommendation list and pricing strategy, the platform conducts intelligent traffic redirection and resource pre-allocation. For users with high interest drift coefficients, in addition to pushing exclusive pop-up windows for package deals, the platform also displays real-life footage of intangible cultural heritage handicraft experiences within the packages in the recommended section of their APP homepage, along with short video introductions narrated by the artisans. For users with traditional sightseeing preferences, in addition to sending price discount links, the platform also provides practical suggestions such as "off-peak travel routes" and "best time to take photos of the scenery". In the resource pre-allocation phase, the platform not only coordinates with partner merchants in advance to reserve 40% of the experience slots for the "light outdoor + intangible cultural heritage handicraft" combination package and 25% of the inventory for traditional sightseeing products, but also establishes a real-time synchronization mechanism to provide merchants with hourly feedback on booking progress and user inquiries: if the booking volume of the combination package reaches 85% of the reserved quota, an additional 15% of the slots will be added immediately and merchants will be reminded to prepare sufficient experience materials such as clay and needles and thread, and a "limited slots" notice will be displayed on the recommendation page; if the booking volume of traditional products is less than 30%, some inventory will be released to the public domain display pool, and its ranking weight in the recommendation list will be reduced, while a limited-time benefit of "second person half price for two people traveling together" will be added. During this process, the platform simultaneously collects multi-dimensional data: interactive feedback data includes user click duration on the package deals, number of inquiries, order rate within 72 hours of saving, cancellation reasons (such as "experience time conflicts with itinerary", "price exceeds expectations", "intangible cultural heritage project type does not match"), and response speed to traditional product promotional activities; market conversion efficiency data includes booking completion rate of the two types of products, experience project redemption rate, average spending per person, resource idle time, revenue growth of partner merchants, and the individual conversion contribution value of the "light outdoor" and "intangible cultural heritage handicrafts" modules in the package deals. Subsequently, the online incremental learning algorithm built by the platform started the optimization process, using these real-time data as supervision signals to accurately adjust the model parameters. In response to feedback that "cancellations of package deals are mostly due to full bookings for intangible cultural heritage handicraft projects", the calculation weight of "intensity of interest diffusion in intangible cultural heritage handicrafts" in the time-series graph neural network model was increased. Combined with conversion data that "traditional products have low redemption rates but consultation volume increased by 18% after price reduction", the model's prediction bias on "time-series drift speed of interest in traditional sightseeing" was corrected, and the interest shift cycle of this group was adjusted from the original prediction of 60 days to 45 days. The entire optimization process did not require retraining all the model parameters, but was completed only through local gradient updates, which shortened the time by 80% compared to full training.Finally, based on the optimized time-series neural network model, the platform generated a tourism demand mining and market response assessment report: The demand mining section clearly shows that the weekly growth rate of demand for "light outdoor activities + intangible cultural heritage handicrafts" reached 35%, and users are more inclined to choose the experience time of 9:00-11:00 on weekends. It also points out the structural changes in traditional sightseeing demand, such as "the proportion of family groups decreased by 12% and the proportion of young solo travelers aged 20-30 increased by 18%". The market response section affirms the effectiveness of the combination package booking conversion rate being 20% ​​higher than that of traditional products and the average consumption per person increasing by 15%. It also points out the problem that the supply gap of intangible cultural heritage handicraft projects in combination packages is as high as 25%, and there is still 18% of inventory idle after the price of traditional sightseeing products is reduced. It also further analyzes the pattern in dynamic pricing, that "the conversion rate is optimal when the combination package is increased by 10%, and the order rate drops significantly when it exceeds 12%". These conclusions provide a clear direction for the platform's next round of adjustments: on the one hand, it will expand the supply of package experiences by partnering with three new intangible cultural heritage workshops and launch a "weekend priority reservation" service; on the other hand, it will launch a "single lightweight package" for traditional sightseeing products, further reducing the price by 8%, while also offering a "check-in stamp collection and exchange for merchandise" activity to cater to the needs of young solo travelers.

[0112] In summary, this invention achieves multiple beneficial effects by integrating multi-dimensional data to construct a dynamic evolution map of tourism interests, combining it with a time-series neural network to quantify and dynamically update interest patterns, then using a supply-demand matching model to mine potential needs and generate precise strategies, and finally relying on online incremental learning for continuous optimization. For users, personalized recommendation lists and dynamic pricing accurately adapt to changes in interests, from single sightseeing to combined recommendations of "light outdoor activities + intangible cultural heritage handicrafts," significantly improving demand satisfaction. For the platform, online incremental learning avoids the high cost of full retraining, real-time model optimization makes strategy adjustments more efficient, and the response speed to accurately capture sudden hot demand is significantly improved. For partner merchants, intelligent traffic diversion and resource pre-allocation reduce the idleness of homestays and experience projects, and the revenue growth and redemption rate driven by bundled packages are improved, maximizing resource value. At the same time, the supply-demand gaps and interest patterns discovered provide a clear direction for subsequent product innovation, achieving a synergistic improvement in user experience, platform efficiency, and merchant revenue.

[0113] Next, referring to the accompanying drawings, an artificial intelligence-based big data mining system for tourism e-commerce is described according to an embodiment of this application.

[0114] Figure 5 This is a schematic diagram of the structure of a tourism e-commerce big data mining system based on artificial intelligence, according to an embodiment of this application.

[0115] like Figure 5 As shown, the tourism e-commerce big data mining system 10 based on artificial intelligence includes: an acquisition module 100, a construction module 200, a prediction module 300, and a generation module 400.

[0116] The system comprises four modules: Acquisition Module 100, which acquires user behavior data, cross-platform product supply data, historical transaction characteristic data, and social media sentiment data; Construction Module 200, which constructs a dynamic evolution map of tourism interests based on historical transaction characteristic data and social media sentiment data, quantifies the temporal drift and community diffusion patterns of user interests through a time-series graph neural network model, and identifies interest shifts triggered by sudden hot events using an anomaly detection algorithm, dynamically updating the weight parameters of the evolution map; Prediction Module 300, which constructs a supply-demand matching prediction model based on cross-platform product supply data and user behavior data, combines the dynamically updated tourism interest evolution map with a generative adversarial network to mine potential tourism demand and predict regional product supply-demand gaps, generating personalized tourism product recommendation lists and dynamic pricing strategies; and Generation Module 400, which performs intelligent traffic redirection and resource pre-allocation based on personalized tourism product recommendation lists and dynamic pricing strategies, simultaneously collecting user interaction feedback data and market conversion efficiency data, and optimizing the time-series graph neural network model through an online incremental learning algorithm based on the interaction feedback data and conversion efficiency data, generating a tourism demand mining and market response assessment report.

[0117] It should be noted that the foregoing explanation of an embodiment of a tourism e-commerce big data mining method based on artificial intelligence also applies to this embodiment of a tourism e-commerce big data mining system based on artificial intelligence, and will not be repeated here.

[0118] According to an embodiment of this application, a tourism e-commerce big data mining system based on artificial intelligence integrates user behavior data, cross-platform product supply data, historical transaction feature data, and social media sentiment data to construct a dynamic evolution graph of tourism interests that comprehensively covers user interest dimensions. Combined with a time-series graph neural network model, it accurately quantifies the temporal drift patterns and community diffusion characteristics of user interests. Simultaneously, relying on anomaly detection algorithms, it captures interest shifts triggered by sudden hot events in real time and dynamically updates the graph weights, avoiding the problem of recommended content being out of touch with real-time user needs and market trends. Furthermore, based on a supply-demand matching prediction model constructed from cross-platform product supply and user behavior data, combined with generative adversarial networks, it can not only accurately predict regional tourism product supply-demand gaps but also deeply mine potential needs not explicitly expressed by users. This results in the generation of personalized tourism product recommendation lists tailored to user needs and flexible dynamic pricing strategies, achieving efficient matching between supply and demand. Furthermore, by optimizing the efficiency of tourism resource allocation through intelligent traffic redirection and resource pre-allocation, and simultaneously leveraging online incremental learning algorithms to continuously optimize the time-series graph neural network model based on user interaction feedback and market conversion efficiency data, the efficiency of user decision-making and the platform's market conversion capabilities are improved. This provides tourism e-commerce platforms with operational support that dynamically adapts to market changes and significantly reduces ineffective resource consumption. Thus, it solves the problems of insufficient capture of dynamic user interest features and weak user demand mining capabilities in existing technologies.

[0119] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0120] The memory 601, the processor 602, and the computer program stored on the memory 601 and capable of running on the processor 602.

[0121] When the processor 602 executes the program, it implements the big data mining method for tourism e-commerce based on artificial intelligence provided in the above embodiments.

[0122] Furthermore, electronic devices also include:

[0123] Communication interface 603 is used for communication between memory 601 and processor 602.

[0124] The memory 601 is used to store computer programs that can run on the processor 602.

[0125] The memory 601 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0126] If the memory 601, processor 602, and communication interface 603 are implemented independently, then the communication interface 603, memory 601, and processor 602 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0127] Optionally, in a specific implementation, if the memory 601, processor 602, and communication interface 603 are integrated on a single chip, then the memory 601, processor 602, and communication interface 603 can communicate with each other through an internal interface.

[0128] The processor 602 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0129] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described artificial intelligence-based big data mining method for tourism e-commerce.

[0130] Furthermore, this application also provides a computer program product, including a computer program or instructions, which, when executed, implement the aforementioned artificial intelligence-based tourism e-commerce big data mining method.

[0131] In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0132] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0133] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0134] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0135] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium. When executed, the program includes one or a combination of the steps of the method embodiments.

[0136] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for big data mining in tourism e-commerce based on artificial intelligence, characterized in that, include: Acquire user behavior data, cross-platform product supply data, historical transaction characteristic data, and social sentiment data; Based on the historical transaction feature data and the social sentiment data, a dynamic evolution map of tourism interests is constructed. The temporal drift and community diffusion patterns of user interests are quantified through a time-series graph neural network model. At the same time, an anomaly detection algorithm is used to identify interest shifts caused by sudden hot events, and the weight parameters of the evolution map are dynamically updated. Based on the cross-platform product supply data and the user behavior data, a supply and demand matching prediction model is constructed. Combined with the dynamically updated tourism interest evolution map, a generative adversarial network is used to mine potential tourism demand and predict regional product supply and demand gaps, generating a personalized tourism product recommendation list and dynamic pricing strategy. Based on the personalized tourism product recommendation list and dynamic pricing strategy, intelligent traffic redirection and resource pre-allocation are carried out, and user interaction feedback data and market conversion efficiency data are collected simultaneously. Based on the interaction feedback data and conversion efficiency data, the time-series neural network model is optimized through an online incremental learning algorithm to generate a tourism demand mining and market response evaluation report.

2. The method for big data mining of tourism e-commerce based on artificial intelligence according to claim 1, characterized in that, Based on the historical transaction characteristic data and the social sentiment data, a dynamic evolution map of tourism interest is constructed, including: Construct a multi-source heterogeneous tourism interest dataset that integrates historical transactions and social media sentiment; Based on the aforementioned multi-source heterogeneous tourism interest dataset, structural features are extracted through graph neural networks, and cross-modal attention mechanisms are used for feature fusion to generate a user-interest feature matrix. Based on the user-interest feature matrix, a dynamically weighted interest network is constructed using a graph attention mechanism, and a time series analysis algorithm is combined to extract the spatiotemporal evolution path and community propagation pattern of user interests, generating a dynamic evolution map of tourism interests.

3. The method for big data mining of tourism e-commerce based on artificial intelligence according to claim 1, characterized in that, The temporal drift and community diffusion patterns of user interests are quantified using a temporal graph neural network model, including: Construct a time-series graph neural network model; The dynamic evolution graph of tourism interests is input into a temporal graph neural network model. The model uses a combination of temporal attention and graph diffusion mechanisms to generate the temporal drift coefficient and community diffusion intensity of user interests.

4. The method for big data mining of tourism e-commerce based on artificial intelligence according to claim 1, characterized in that, Anomaly detection algorithms are used to identify interest transitions triggered by sudden hotspot events, and the weight parameters of the evolutionary graph are dynamically updated, including: Construct an anomaly detection algorithm; Based on the aforementioned anomaly detection algorithm, the social media sentiment data is monitored in real time to identify sudden changes in the popularity of topics of interest, and these sudden changes are identified as sudden hot events. In the dynamic evolution graph of tourism interests, the users and interest nodes associated with the sudden hot events are located, and the user interest subgraph composed of these nodes and their connecting edges is extracted. The structural metric changes of the user interest subgraph before and after the occurrence of the sudden hotspot event are calculated using a graph structure time series comparison algorithm. The structural metric includes changes in node centrality and edge weights. Based on the changes in the structural metric, adjust the influence weight of the corresponding node and the propagation probability parameter of the edge in the dynamic evolution graph of tourism interest.

5. The method for big data mining of tourism e-commerce based on artificial intelligence according to claim 1, characterized in that, Based on the cross-platform product supply data and the user behavior data, a supply-demand matching degree prediction model is constructed, including: Acquire cross-platform product supply data and user behavior data; The cross-platform product supply data is analyzed and vectorized in multiple dimensions to generate standardized product supply feature vectors. At the same time, the user behavior data is sequence modeled and feature extracted to generate user behavior feature vectors. The product supply feature vector and user behavior feature vector are concatenated or cross-calculated using a deep matching learning algorithm, and then deeply fused through a multi-layer nonlinear network. A training sample set is constructed based on historical user-product interaction records. The neural network is iteratively optimized using the backpropagation algorithm with the prediction of interaction probability as the objective function, thus completing the construction of the supply and demand matching degree prediction model.

6. The method for big data mining of tourism e-commerce based on artificial intelligence according to claim 1, characterized in that, Based on the aforementioned interactive feedback data and conversion efficiency data, the time-series graph neural network model is optimized using an online incremental learning algorithm to generate a tourism demand mining and market response assessment report, including: Constructing online incremental learning algorithms; Based on the aforementioned online incremental learning algorithm, interactive feedback data and conversion efficiency data are used as real-time monitoring signals to incrementally optimize the time-series graph neural network model; Based on the optimized time-series graph neural network model, a tourism demand mining and market response assessment report is generated.

7. A tourism e-commerce big data mining system based on artificial intelligence, characterized in that, include: The acquisition module is used to acquire user behavior data, cross-platform product supply data, historical transaction feature data, and social sentiment data. The construction module is used to construct a dynamic evolution map of tourism interests based on the historical transaction feature data and the social sentiment data. It quantifies the temporal drift and community diffusion patterns of user interests through a time-series graph neural network model. At the same time, it uses an anomaly detection algorithm to identify interest leaps caused by sudden hot events and dynamically updates the weight parameters of the evolution map. The prediction module is used to construct a supply and demand matching prediction model based on the cross-platform product supply data and the user behavior data. Combined with the dynamically updated tourism interest dynamic evolution map, it uses a generative adversarial network to mine potential tourism demand and predict regional product supply and demand gaps, and generate a personalized tourism product recommendation list and dynamic pricing strategy. The generation module is used to intelligently guide traffic and pre-allocate resources based on the personalized tourism product recommendation list and dynamic pricing strategy, simultaneously collect user interaction feedback data and market conversion efficiency data, and optimize the time-series graph neural network model based on the interaction feedback data and conversion efficiency data through an online incremental learning algorithm to generate a tourism demand mining and market response evaluation report.

8. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the artificial intelligence-based big data mining method for tourism e-commerce as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When a computer program or instruction is executed, it implements the big data mining method for tourism e-commerce based on artificial intelligence as described in any one of claims 1-6.

10. A computer program product, comprising a computer program or instructions, characterized in that, When a computer program or instruction is executed, it implements the big data mining method for tourism e-commerce based on artificial intelligence as described in any one of claims 1-6.

Images