Tourism commodity recommendation method, device and equipment, readable storage medium and product

Abstract

The invention discloses a text travel commodity recommendation method and device, equipment, a readable storage medium and a product, and relates to the technical field of big data, and the method comprises the steps: obtaining user data and text travel commodity data, and the user data comprises online behavior data and offline behavior data; performing vectorization processing on the online behavior data to obtain an online behavior embedded vector, performing vectorization processing on the offline behavior data to obtain an offline behavior embedded vector, and performing vectorization processing on the document travel commodity data to obtain a commodity node embedded vector; and inputting the user node embedding vector and the commodity node embedding vector into a pre-trained travel commodity recommendation model to obtain a travel commodity recommendation result. According to the method and the device, the technical problem of low precision of the text travel commodity recommendation result is solved, and the effect of improving the precision of the text travel commodity recommendation result is achieved.

Description

Technical Field

[0001] The present application relates to the field of big data technology, and in particular to a method, device, equipment, readable storage medium and product for recommending cultural and tourism products. Background Art

[0002] Traditional cultural and tourism product recommendation services mine user preference characteristics (i.e., user portraits) based on online data such as user gender, age, purchase history, product ratings, usage habits, and product attributes, and then recommend cultural and tourism products with similar characteristics. However, this traditional cultural and tourism product recommendation method only relies on online data generated by users to create user portraits, and cannot fully characterize and learn user portraits, resulting in low accuracy in cultural and tourism product recommendation results.

[0003] Therefore, how to improve the accuracy of cultural and tourism product recommendation results is a technical problem that needs to be urgently solved in this technical field. Summary of the invention

[0004] The main purpose of this application is to provide a method, device, equipment, readable storage medium and product for recommending cultural and tourism products, aiming to solve the technical problem of how to improve the accuracy of cultural and tourism product recommendation results.

[0005] To achieve the above objectives, the present application provides a method for recommending cultural and tourism products, which includes the following steps:

[0006] Obtain user data and cultural tourism product data, wherein the user data includes online behavior data and offline behavior data;

[0007] Performing vectorization processing on the online behavior data to obtain an online behavior embedding vector, performing vectorization processing on the offline behavior data to obtain an offline behavior embedding vector, and performing vectorization processing on the cultural and tourism product data to obtain a product node embedding vector, wherein the vector dimension of the online behavior embedding vector is consistent with that of the offline behavior embedding vector;

[0008] Fusion of the online behavior embedding vector and the offline behavior embedding vector to obtain a user node embedding vector;

[0009] The user node embedding vector and the product node embedding vector are input into a pre-trained cultural and tourism product recommendation model to obtain a cultural and tourism product recommendation result, and cultural and tourism product recommendations are made based on the cultural and tourism product recommendation result.

[0010] In one embodiment, before the step of inputting the user node embedding vector and the product node embedding vector into a pre-trained cultural and tourism product recommendation model to obtain a cultural and tourism product recommendation result, the method further includes:

[0011] Obtain the historical user node representation corresponding to the user data, and obtain the historical commodity node representation corresponding to the cultural and tourism commodity data, where the historical user node representation is the historical value state space representation of the user data, and the historical commodity node representation is the historical value state space representation of the cultural and tourism commodity data;

[0012] Update the user node embedding vector based on the historical user node representation to obtain a target user node embedding vector, and update the commodity node embedding vector based on the historical commodity node representation to obtain a target commodity node embedding vector;

[0013] Use the target user node embedding vector as the new user node embedding vector, use the target commodity node embedding vector as the new commodity node embedding vector, and based on the new user node embedding vector and the new commodity node embedding vector, perform the step of inputting the user node embedding vector and the commodity node embedding vector into a pre-trained cultural and tourism commodity recommendation model to obtain a cultural and tourism commodity recommendation result.

[0014] In one embodiment, the step of updating the user node embedding vector based on the historical user node representation to obtain a target user node embedding vector and updating the commodity node embedding vector based on the historical commodity node representation to obtain a target commodity node embedding vector includes:

[0015] Perform weighted summation on the historical user node representation and the user node embedding vector to obtain a target user node embedding vector;

[0016] Perform weighted summation on the historical commodity node representation and the commodity node embedding vector to obtain a target commodity node embedding vector.

[0017] In one embodiment, the cultural and tourism commodity recommendation model includes a dynamic graph network and a prediction network. The step of inputting the user node embedding vector and the commodity node embedding vector into a pre-trained cultural and tourism commodity recommendation model to obtain a cultural and tourism commodity recommendation result includes:

[0018] Input the user node embedding vector and the commodity node embedding vector into the dynamic graph network to obtain a current user representation and a current commodity representation, where the dynamic graph network is used to update the node representations of the nodes in the user-commodity interaction graph, where the node representation is the value state space representation of the node, the user-commodity interaction graph includes user nodes and commodity nodes, the user node is a node representing a user, and the commodity node is a node representing a cultural and tourism commodity;

[0019] Input the current user representation and the current commodity representation into the prediction network to obtain a cultural and tourism commodity recommendation result.

[0020] In one embodiment, the step of inputting the user node embedding vector and the product node embedding vector into the dynamic graph network to obtain the current user representation and the current product representation includes:

[0021] Input the user node embedding vector and the product node embedding vector into the dynamic graph network, and obtain or construct a user-product interaction graph through the dynamic graph network;

[0022] Determine the node corresponding to the user data in the user-product interaction graph as the target user node, and determine the node corresponding to the cultural and tourism product data in the product interaction graph as the target product node;

[0023] Update the node representation of the target user node based on the user node embedding vector to obtain the current node representation of the target user node;

[0024] Update the node representation of the target product node based on the product node embedding vector to obtain the node representation of the target product node;

[0025] Determine the current node representation of the target user node as the current user representation, and determine the current node representation of the target product node as the current product representation.

[0026] In one embodiment, the step of updating the node representation of the target user node based on the user node embedding vector to obtain the current node representation of the target user node includes:

[0027] Determine the neighbor nodes adjacent to the target user node in the user-product interaction graph, and obtain the neighbor node embedding vectors of the neighbor nodes;

[0028] Perform message aggregation on the user node embedding vector and the neighbor node embedding vectors based on the graph attention mechanism to obtain the current node representation of the target user node.

[0029] In one embodiment, the offline behavior data includes at least one trajectory point, and the step of vectorizing the offline behavior data to obtain an offline behavior embedding vector includes:

[0030] Input each of the trajectory points into a pre-trained spatio-temporal large model to obtain an output vector corresponding to each of the trajectory points, where the spatio-temporal large model includes at least one layer of encoding units, and each layer of the encoding units includes an attention layer, a normalization layer, a feed-forward neural network, and a normalization layer connected in sequence;

[0031] Calculate the average vector of the output vectors, and determine the average vector as the offline behavior embedding vector.

[0032] In addition, to achieve the above object, the present application further provides a cultural and tourism commodity recommendation device, which includes:

[0033] An acquisition module, configured to acquire user data and cultural and tourism commodity data, wherein the user data includes online behavior data and offline behavior data;

[0034] A vectorization module, configured to perform vectorization processing on the online behavior data to obtain an online behavior embedding vector, perform vectorization processing on the offline behavior data to obtain an offline behavior embedding vector, and perform vectorization processing on the cultural and tourism commodity data to obtain a commodity node embedding vector, wherein the vector dimensions of the online behavior embedding vector and the offline behavior embedding vector are the same;

[0035] A fusion module, configured to fuse the online behavior embedding vector and the offline behavior embedding vector to obtain a user node embedding vector;

[0036] A recommendation module, configured to input the user node embedding vector and the commodity node embedding vector into a pre-trained cultural and tourism commodity recommendation model to obtain a cultural and tourism commodity recommendation result, and perform cultural and tourism commodity recommendation according to the cultural and tourism commodity recommendation result.

[0037] In addition, to achieve the above object, the present application further provides a cultural and tourism commodity recommendation device, which includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the above-mentioned cultural and tourism commodity recommendation method.

[0038] In addition, to achieve the above object, the present application further provides a readable storage medium, which is a computer-readable storage medium, and a computer program is stored on the computer-readable storage medium, and the computer program is executed by a processor to implement the steps of the above-mentioned cultural and tourism commodity recommendation method.

[0039] The present application also provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the steps of the above-mentioned cultural and tourism commodity recommendation method.

[0040] One or more technical solutions proposed by the present application have at least the following technical effects:

[0041] (1) Extract user node embedding vectors based on the user's online behavior data and offline behavior data, so as to represent the user portrait through the user node embedding vectors, and then perform subsequent cultural and tourism product recommendations based on the user portrait. In this way, compared with the method of constructing a user portrait only based on online behavior data, by integrating the user's online and offline behavior data, a more comprehensive user portrait can be constructed, and the online behavior data and offline behavior data are integrated for cultural and tourism product recommendations, thereby improving the accuracy of the recommendation results.

[0042] (2) Considering that the user's online behavior data and offline behavior data may be multi-source heterogeneous, such as the extremely large difference in the number of behavior trajectories generated by different users during tourism, and the online behavior data may be in various forms such as text, pictures, and tables. Based on this, the embodiments of this application perform independent vectorization processing on the online behavior data and offline behavior data respectively, and convert the online behavior data and offline behavior data into vectors of the same dimension. In this way, the multi-dimensional integration and dimension unification of the user's online and offline behavior data are realized, and the input method of the cultural and tourism product recommendation model is unified, thereby greatly reducing the complexity of different cultural and tourism product recommendation modeling, that is, reducing the complexity of constructing the cultural and tourism product recommendation model.

[0043] (3) Performing independent vectorization processing on the online behavior data and offline behavior data respectively can also better capture the special properties and subtle differences of each feature in the online data and offline behavior data, reduce the mutual interference between different features, and reduce the situation where some features are suppressed during the unified vectorization of data, so as to extract deeper features in the online behavior data and offline behavior data, thereby realizing the deep integration of the online behavior data and offline behavior data, and further improving the accuracy of the cultural and tourism product recommendation results. Brief Description of the Drawings

[0044] The accompanying drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with this application, and are used together with the specification to explain the principles of this application.

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0046] Figure 1 It is a schematic flowchart of the first embodiment of the cultural and tourism product recommendation method of this application;

[0047] Figure 2 It is a schematic flowchart of the embedding processing involved in an embodiment of the cultural and tourism product recommendation method of this application;

[0048] Figure 3 This is a schematic diagram of the node representation update process involved in a specific embodiment of the cultural and tourism commodity recommendation method of the present application;

[0049] Figure 4 This is a schematic diagram of the large model structure at a certain moment involved in an embodiment of the cultural and tourism commodity recommendation method of the present application;

[0050] Figure 5 This is a schematic diagram of the overall process of cultural and tourism commodity recommendation involved in a specific embodiment of the cultural and tourism commodity recommendation method of the present application;

[0051] Figure 6 This is a schematic diagram of the device structure of the cultural and tourism commodity recommendation device of the present application;

[0052] Figure 7 This is a schematic diagram of the device structure of the hardware operating environment involved in the cultural and tourism commodity recommendation method device in the embodiment of the present application.

[0053] The realization of the purpose, functional characteristics and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. Detailed implementation manners

[0054] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0055] Recommendation algorithms mainly recommend the required information or commodities to users according to the user's interests and behavioral characteristics, helping users quickly discover the truly required commodities in the vast amount of information to improve user stickiness. The inputs of recommendation algorithms mainly include hot information or commodities, user Profile information such as gender, age, occupation, income, and the city where the user is located, the user's historical browsing or behavioral records, social relationships, etc. Common recommendation algorithms include three categories: (1) Content-based recommendation, which mines the user's preference characteristics based on the user's past preferences or behaviors, and then recommends items with similar characteristics. (2) Collaborative filtering, which predicts the items that the user may be interested in by analyzing the behavioral patterns among users. It is divided into user-based collaborative filtering and item-based collaborative filtering. (3) Hybrid recommendation, which combines the advantages of multiple recommendation algorithms to improve the accuracy of recommendation.

[0056] The traditional cultural and tourism commodity recommendation methods mainly include the following two categories:

[0057] The first category is the tourism and cultural commodity recommendation method based on clustering. This type of method first uses clustering algorithms to classify the tourism demand intentions of multiple historical users, obtains a target classification model composed of multiple user clusters, and determines the user cluster to which the target user belongs in the classification model according to the tourism demand intention of the target user. Then, the various historical users and corresponding cross-border tourism and cultural commodity recommendation plans in the user cluster are used as a training data set to construct a classification decision tree. Finally, each validation data set is used to prune the classification decision tree, obtaining multiple pruned classification decision trees, and the optimal pruned classification decision tree is used to analyze the tourism demand intention of the target user to determine the cross-border tourism and cultural commodity recommendation plan for the target user. This type of method does not incorporate the online and offline behavior characteristics of users to assist in judging the potential preferences of users. Moreover, there is a lack of in-depth exploration and matching of user behavior and tourism intentions, and it cannot be applied to the recommendation of various different types of tourism and cultural products, resulting in poor applicability.

[0058] The second category is the tourism and cultural commodity recommendation method based on matrices. This type of method first constructs the portrait information and interest information of users, and determines the user characteristic information according to the portrait information and interest information. Then, a preset number of knowledge chain groups are extracted from the knowledge graph and transformed into a multi-dimensional feature matrix, or a user-scenic spot rating multi-dimensional matrix is constructed. Finally, the multi-dimensional feature matrix decomposition is used to train a recommendation model that matches the user characteristic information, and the user characteristic information is input into the trained recommendation model to obtain the output target tourism service information. This type of method considers the complex relationship between the user's own attributes and the scenic spot attributes and can achieve personalized recommendations for tourism scenic spots. However, this type of method lacks the comprehensive integration of the online and offline behaviors of users, and the construction of user characteristics depends on subjective division and cannot be automatically learned and generated. The matrix decomposition algorithm cannot capture the complex dynamic interaction relationship between user characteristics and tourism scenarios, resulting in the inability of this method to be effectively extended to the recommendation of other tourism and cultural commodities.

[0059] However, traditional tourism and cultural recommendation methods have at least the following problems:

[0060] (1) The high correlation between the user's offline spatio-temporal trajectory and the purchase of tourism and cultural commodities is not fully considered. For example, tourists who have visited the same scenic spot may have the same preference for purchasing tourism and cultural commodities. Existing methods only rely on a small number of static user attributes, such as age, gender, tourism willingness, etc., and cannot achieve a comprehensive description and learning of the user portrait, resulting in deviations in the results of tourism and cultural commodity recommendations.

[0061] (2) The online and offline behavior data of users are not converted to the same dimension, and the input methods for modeling the personalized cultural and tourism product recommendation model are not unified. The online and offline attributes of users are multi-source heterogeneous. For example, the number of spatio-temporal trajectories generated by different users' travels varies greatly, and the online attributes of users may be in various forms such as text, pictures, and tables. Traditional recommendation methods cannot handle multi-source heterogeneous user behavior data, resulting in the need to separately select appropriate input preprocessing and modeling methods for different cultural and tourism products, with relatively high complexity.

[0062] Based on this, the main solution of this application is: obtaining user data and cultural and tourism product data, where the user data includes online behavior data and offline behavior data; performing vectorization processing on the online behavior data to obtain an online behavior embedding vector, performing vectorization processing on the offline behavior data to obtain an offline behavior embedding vector, and performing vectorization processing on the cultural and tourism product data to obtain a product node embedding vector, where the vector dimensions of the online behavior embedding vector and the offline behavior embedding vector are the same; fusing the online behavior embedding vector and the offline behavior embedding vector to obtain a user node embedding vector; and inputting the user node embedding vector and the product node embedding vector into a pre-trained cultural and tourism product recommendation model to obtain a cultural and tourism product recommendation result.

[0063] This application extracts a user node embedding vector based on the online behavior data and offline behavior data of users, uses the user node embedding vector to represent the user profile, and then makes subsequent cultural and tourism product recommendations based on the user profile. In this way, compared with the method of constructing a user profile only based on online behavior data, by fusing the online and offline behavior data of users, a more comprehensive user profile can be constructed, and the online behavior data and offline behavior data are fused for cultural and tourism product recommendations, thereby improving the accuracy of the recommendation results.

[0064] Furthermore, considering that the online behavior data and offline behavior data of users may be multi-source heterogeneous, such as the number of behavior trajectories generated by different users' travels varying greatly, and the online behavior data may be in various forms such as text, pictures, and tables, etc., based on this, the embodiments of this application separately perform independent vectorization processing on the online behavior data and the offline behavior data, and convert the online behavior data and the offline behavior data into vectors with the same dimension. In this way, the multi-dimensional fusion and dimension unification of the online and offline behavior data of users are realized, the input methods for the cultural and tourism product recommendation model are unified, and thus the complexity of recommending different cultural and tourism products for modeling is greatly reduced, that is, the complexity of constructing the cultural and tourism product recommendation model is reduced.

[0065] Furthermore, independent vectorization processes are performed on the online behavior data and the offline behavior data respectively, which can better capture the special properties and subtle differences of each feature in the online data and the offline behavior data, reduce the mutual interference between different features, and reduce the situation where some features are suppressed during the unified vectorization of the data, so as to enable the extraction of deeper features in the online behavior data and the offline behavior data, thereby realizing the deep integration of the online behavior data and the offline behavior data, and further improving the accuracy of the recommended results of cultural and tourism products.

[0066] It should be noted that the execution subject of each embodiment of the cultural and tourism product recommendation method of this application can be a computing service device with data processing, network communication, and program running functions, such as a server, a tablet computer, a personal computer, a mobile phone, etc., or a cultural and tourism product recommendation device capable of implementing the above functions. Each embodiment of the cultural and tourism product recommendation method of this application does not make specific restrictions on this.

[0067] Based on this, the first embodiment of the cultural and tourism product recommendation method of this application is proposed. Please refer to Figure 1 As shown, the cultural and tourism product recommendation method includes the following steps S10 to S40:

[0068] Step S10, obtain user data and cultural and tourism product data, where the user data includes online behavior data and offline behavior data;

[0069] The online behavior data is the relevant data generated by the user online, such as including but not limited to one or more of user attributes, consumption records, and social browsing records; the offline behavior data is the relevant data generated by the user offline, such as including but not limited to behavior trajectory records; the cultural and tourism product data is the data related to cultural and tourism products, such as including but not limited to one or more of product types, product prices, and user anonymous ID (Identification). Relevant personnel can set the data to be obtained based on actual needs. This embodiment does not make specific restrictions on this. Hereinafter, by way of example, the embodiments of this application will be described and illustrated with the online behavior data including user attributes, consumption records, and social browsing records, the offline behavior data including behavior trajectory records, and the cultural and tourism product data including product types, product prices, and user anonymous ID.

[0070] Among them, the user attributes can specifically be the personal attributes of the user, including but not limited to information such as age, gender, income, etc.; the consumption records can specifically be the online consumption records of the user, including but not limited to the types of goods consumed (such as clothing, catering, travel, etc.) and prices. The consumption records do not include the records of the user purchasing cultural and tourism goods; the social browsing records can specifically be the records of the user browsing web pages and / or apps (Application) on the Internet, including but not limited to the browsing frequency and the types of web pages and / or apps, such as historical site types, natural scenery types, overseas scenic spot types, etc.

[0071] The behavioral trajectory data can specifically be sourced from the user trajectories recorded by Internet maps and / or location service apps, such as the location information generated when the user initiates navigation or location services, and / or from operator signaling trajectory data. Signaling trajectory data refers to the location information (such as the location of the communication base station) generated when the user uses a mobile device to send text messages, make phone calls, or use Internet services and interacts with the communication base station. The behavioral trajectory data includes but is not limited to the user anonymous ID, longitude, latitude, and time, etc.

[0072] Furthermore, all the behavioral trajectory records of the same user can be sorted in chronological order to obtain the user's behavioral trajectory data.

[0073] The cultural and tourism goods data can specifically be sourced from the records of the user purchasing cultural and tourism goods on the cultural and tourism platform. The records include attributes such as the user anonymous ID, the types of goods (such as tickets, souvenirs, cultural and creative products, etc.), and the prices of the goods. Among them, the user anonymous ID can specifically be the ID of the user who purchased the cultural and tourism goods.

[0074] It should be noted that the user data can specifically be sourced from one or more users, that is, the user data can specifically include the online and offline behavioral data corresponding to one or more users respectively. Similarly, the cultural and tourism goods data can specifically be sourced from one or more cultural and tourism goods. For example, in a specific application scenario, the online and offline behavioral data of all users who purchased cultural and tourism goods within a certain historical period can be obtained to obtain the user data, and the types of goods, the prices of the goods, and the user anonymous ID of all the cultural and tourism goods purchased within a certain historical period can be obtained to obtain the cultural and tourism goods data.

[0075] Further, the triggering condition of the cultural and tourism product recommendation request can be set. When it is detected that the cultural and tourism product recommendation request is triggered, user data and cultural and tourism product data are obtained for cultural and tourism product recommendation. For example, in a specific application scenario, after it is detected that a user purchases a cultural and tourism product (assuming this is the t + 1 moment), it is determined that the cultural and tourism product recommendation request is triggered. Then, the online behavior data and offline behavior data of all users who have had cultural and tourism product purchase behaviors at the t + 1 moment or in the previous period including the t + 1 moment can be obtained to get user data. Similarly, the product types, product prices, and user anonymized IDs of all cultural and tourism products purchased at the t + 1 moment or in the previous period including the t + 1 moment can be obtained to get cultural and tourism product data. For example, taking the data at the t + 1 moment as an example, assuming that at the t + 1 moment, user 1 purchases cultural and tourism product A and cultural and tourism product B, and user 2 purchases cultural and tourism product C, then the relevant data of user 1 and user 2 at the t + 1 moment can be obtained to form user data, and the relevant data of cultural and tourism product A, cultural and tourism product B, and cultural and tourism product C at the t + 1 moment can be obtained to form cultural and tourism product data.

[0076] Step S20: Perform vectorization processing on the online behavior data to obtain an online behavior embedding vector, perform vectorization processing on the offline behavior data to obtain an offline behavior embedding vector, and perform vectorization processing on the cultural and tourism product data to obtain a product node embedding vector, where the vector dimensions of the online behavior embedding vector and the offline behavior embedding vector are the same;

[0077] Vectorization refers to the process of converting data into a numerical vector. The specific processing method of vectorization processing can be preset. For example, the vectorization processing of data can be completed by embedding (such as word embedding). Embedding Processing generally refers to converting data from its original form into a low-dimensional, continuous vector representation, which can capture the internal characteristics and structure of the data. Exemplarily, in the following, the vectorization processing is taken as embedding processing to elaborate and illustrate the embodiments of the present application.

[0078] Further, the embedding processing methods of the online behavior data, the offline behavior data, and the cultural and tourism product data can be the same or different, and this embodiment does not make specific limitations on this. For example, in a specific embodiment, refer to Figure 2As shown, online behavior data such as user attributes, consumption records, and social browsing records are respectively input into a multi-layer perceptron. The multi-layer perceptron embeds the online behavior data. The hidden layer dimension of the multi-layer perceptron is 2*dim, and the output dimension is dim. Dim is the dimension of the finally embedded online behavior embedding vector, such as 512 dimensions. Add the user attribute embedding vector, consumption record embedding vector, and social browsing record embedding vector to obtain the online behavior embedding vector. Input the offline behavior data, which is the offline behavior trajectory of the user, into a pre-trained spatio-temporal large model, and the spatio-temporal large model embeds the offline behavior data.

[0079] It should be noted that the user data can specifically come from one or more users. Then, it is easy to understand that the online behavior data and the offline behavior data can also come from one or more users respectively. Based on this, the online behavior data and the offline behavior data belonging to different users can be embedded separately. Similarly, the cultural and tourism product data belonging to different cultural and tourism products can be embedded separately.

[0080] Exemplarily, assume that the user data includes online behavior data 1 and offline behavior data 1 from user 1 and online behavior data 2 and offline behavior data 2 from user 2. Then, embed online behavior data 1 and online behavior data 2 separately to obtain the corresponding online behavior embedding vector 1 and online behavior embedding vector 2. Embed offline behavior data 1 and offline behavior data 2 separately to obtain the corresponding offline behavior embedding vector 1 and offline behavior embedding vector 2. The cultural and tourism product data includes cultural and tourism product data A from cultural and tourism product A, cultural and tourism product data B from cultural and tourism product B, and cultural and tourism product data C from cultural and tourism product C. Then, embed cultural and tourism product data A, cultural and tourism product data B, and cultural and tourism product data C separately to obtain the corresponding node embedding vector A, node embedding vector B, and node embedding vector C.

[0081] Step S30: Fuse the online behavior embedding vector and the offline behavior embedding vector to obtain a user node embedding vector;

[0082] After obtaining the online behavior embedding vector and the offline behavior embedding vector, fuse the online behavior embedding vector and the offline behavior embedding vector to obtain a user node embedding vector. Specifically, relevant personnel can set a specific vector fusion method. For example, in a specific embodiment, the sum vector of the online behavior embedding vector and the offline behavior embedding vector can be calculated, and the two vectors are fused by summing, and the sum vector is determined as the user node embedding vector.

[0083] It should be noted that the online behavior embedding vectors and offline behavior embedding vectors corresponding to each user are respectively fused to obtain the node embedding vector corresponding to the user. That is, the user node embedding vector includes one or more node embedding vectors corresponding to the user. Exemplarily, taking the online behavior embedding vector 1, online behavior embedding vector 2, offline behavior embedding vector 1, and offline behavior embedding vector 2 obtained in the above example, the node embedding vector 1 is obtained by fusing the online behavior embedding vector 1 and the offline behavior embedding vector 1, and the node embedding vector 2 is obtained by fusing the online behavior embedding vector 2 and the offline behavior embedding vector 2. Among them, the node embedding vector 1 is the node embedding vector corresponding to user 1, and the node embedding vector 2 is the node embedding vector corresponding to user 2. The user node embedding vector can be expressed as [node embedding vector 1, node embedding vector 2].

[0084] Similarly, the obtained commodity node embedding vector specifically includes the node embedding vector corresponding to each cultural and tourism commodity. Exemplarily, taking the node embedding vector A, node embedding vector B, and node embedding vector C obtained in the above example, where the node embedding vector A is the node embedding vector corresponding to the cultural and tourism commodity A, the node embedding vector B is the node embedding vector corresponding to the cultural and tourism commodity B, and the node embedding vector C is the node embedding vector corresponding to the cultural and tourism commodity C. The commodity node embedding vector can be expressed as [node embedding vector A, node embedding vector B, node embedding vector C].

[0085] Step S40, input the user node embedding vector and the commodity node embedding vector into a pre-trained cultural and tourism commodity recommendation model to obtain a cultural and tourism commodity recommendation result, and perform cultural and tourism commodity recommendation according to the cultural and tourism commodity recommendation result.

[0086] After obtaining the user node embedding vector and the commodity node embedding vector, input the user node embedding vector and the commodity node embedding vector into a pre-trained cultural and tourism commodity recommendation model to obtain a cultural and tourism commodity recommendation result.

[0087] The cultural and tourism commodity recommendation model can specifically be a model pre-constructed by relevant personnel, such as a deep learning model, etc. This embodiment does not make specific limitations on this.

[0088] The cultural tourism product recommendation result can specifically be a result that characterizes whether a certain user will purchase a certain cultural tourism product, and further, it can also be a result that characterizes the probability of a certain user purchasing each cultural tourism product. Exemplarily, assuming that the input of the model is the user node embedding vector corresponding to user 1 and user 2, and the product node embedding vector corresponding to cultural tourism product A, cultural tourism product B, and cultural tourism product C, then the recommendation result can specifically be whether user 1 will purchase cultural tourism product A, whether user 1 will purchase cultural tourism product B, whether user 1 will purchase cultural tourism product C, whether user 2 will purchase cultural tourism product A, whether user 2 will purchase cultural tourism product B, whether user 2 will purchase cultural tourism product C, or the probability of user 1 purchasing cultural tourism product A, the probability of user 1 purchasing cultural tourism product B, the probability of user 1 purchasing cultural tourism product C, the probability of user 2 purchasing cultural tourism product A, the probability of user 2 purchasing cultural tourism product B, and the probability of user 2 purchasing cultural tourism product C.

[0089] After obtaining the cultural and tourism recommendation results, cultural and tourism recommendations are made according to the cultural and tourism product recommendation results. Specifically, the recommended cultural and tourism products indicated by the cultural and tourism product recommendation results can be output to the user to complete the cultural and tourism product recommendation. The cultural and tourism product recommendation results of different users can be recommended to the corresponding users respectively to achieve personalized recommendations. For example, assuming that the cultural and tourism recommendation result is that user 1 will purchase cultural and tourism product C, and user 2 will purchase cultural and tourism product A, then cultural and tourism product C will be displayed on the display interface of user 1 to complete the cultural and tourism product recommendation to user 1, and cultural and tourism product A will be displayed on the display interface of user 2 to complete the cultural and tourism product recommendation to user 2; for another example, assuming that the cultural and tourism recommendation result is that the probability of user 1 purchasing cultural and tourism product C is 90%, the probability of purchasing cultural and tourism product A is 10%, and the probability of purchasing cultural and tourism product B is 10%, and the probability of user 2 purchasing cultural and tourism product C is 10%, the probability of purchasing cultural and tourism product A is 90%, and the probability of purchasing cultural and tourism product B is 95%, then cultural and tourism product C, cultural and tourism product A (or cultural and tourism product B), cultural and tourism product B (or cultural and tourism product A) can be displayed in order on the display interface of user 1, and cultural and tourism product B, cultural and tourism product A and cultural and tourism product C can be displayed in order on the display interface of user 2.

[0090] This embodiment extracts user node embedding vectors based on the user's online behavior data and offline behavior data, so as to represent the user portrait through the user node embedding vector, and then make subsequent cultural and tourism product recommendations based on the user portrait. In this way, compared with the method of making user portraits based only on online behavior data, the fusion of online and offline behavior data of users can build a more comprehensive user portrait, and the fusion of online and offline behavior data can be used to recommend cultural and tourism products, thereby improving the accuracy of the recommendation results.

[0091] Considering that the online behavior data and offline behavior data of users may be multi-source heterogeneous, such as the significant differences in the number of behavioral trajectories generated by different users during travel, and the online behavior data may be in various forms such as text, images, tables, etc. Based on this, in this embodiment, the online behavior data and offline behavior data are respectively subjected to independent vectorization processing, and the online behavior data and offline behavior data are converted into vectors of the same dimension. In this way, the multi-dimensional fusion and dimension unification of the user's online and offline behavior data are realized, the input method of the cultural and tourism commodity recommendation model is unified, thereby greatly reducing the complexity of different cultural and tourism commodity recommendation modeling, that is, reducing the complexity of the construction of the cultural and tourism commodity recommendation model.

[0092] In this embodiment, the online behavior data and offline behavior data are respectively subjected to independent vectorization processing, which can also better capture the special properties and subtle differences of each feature in the online data and offline behavior data, reduce the mutual interference between different features, and reduce the situation where some features are suppressed during the unified vectorization of data, so as to extract deeper features in the online behavior data and offline behavior data, thereby realizing the deep fusion of the online behavior data and offline behavior data, and further improving the accuracy of the cultural and tourism commodity recommendation results.

[0093] Based on the first embodiment of the present application, in the second embodiment of the present application, the content that is the same as or similar to the above-mentioned first embodiment can be referred to the above introduction and will not be elaborated hereinafter. On this basis, before the step of inputting the user node embedding vector and the commodity node embedding vector into the pre-trained cultural and tourism commodity recommendation model to obtain the cultural and tourism commodity recommendation result, the method further includes:

[0094] Step A10, obtaining the historical user node representation corresponding to the user data, and obtaining the historical commodity node representation corresponding to the cultural and tourism commodity data, where the historical user node representation is the historical value state space representation of the user data, and the historical commodity node representation is the historical value state space representation of the cultural and tourism commodity data;

[0095] It should be noted that the node representations of each user and each cultural and tourism commodity can be stored. Initially, the node representations can be initialized by setting default parameters or by random parameters, where the node representations are used to represent the value state space. For example, the node representations of the stored users can be referred to Table 1, and the node representations of the stored cultural and tourism commodities can be referred to Table 2.

[0096] User ID Node representation 1 (......) 2 (......) 3 (......) ....... .......

[0097] Table 1

[0098] Product ID Node representation A (......) B (......) C (......) ....... .......

[0099] Table 2

[0100] Further, the stored node representations may specifically be the node representations of all users and all cultural and tourism products that can be obtained. For example, in a specific scenario, if a certain APP uses this cultural and tourism product recommendation method for cultural and tourism recommendations, it can store the node representations of the users who register for this APP and the node representations of all the cultural and tourism products that can be sold on this APP. When a new user registers or a new cultural and tourism product is launched, the node representation of this user or cultural and tourism product can be initialized and stored.

[0101] The historical user node representation corresponding to the user data may specifically be the node representation of the source users of the user data (that is, the user data comes from these users). At this time, since the node representation has not been updated based on the latest obtained user data or cultural and tourism product data, for the convenience of distinction, the node representation obtained at this time is denoted as the historical node representation. For example, assume that the user data comes from User 1 and User 2. Then the historical user node representation corresponding to the user data includes the historical node representation of User 1 and the historical node representation of User 2, and the historical user node representation can be expressed as [Historical Node Representation 1, Historical Node Representation 2]. Similarly, the historical product node representation corresponding to the cultural and tourism product data may specifically be the node representation of the cultural and tourism products from which the cultural and tourism product data comes. For example, assume that the cultural and tourism product data comes from Cultural and Tourism Product A, Cultural and Tourism Product B, and Cultural and Tourism Product C. Then the historical product node representation corresponding to the cultural and tourism product data includes the historical node representation of Cultural and Tourism Product A, the historical node representation of Cultural and Tourism Product B, and the historical node representation of Cultural and Tourism Product C, and the historical product node representation can be expressed as [Historical Node Representation A, Historical Node Representation B, Historical Node Representation C].

[0102] Step A20: Update the user node embedding vector based on the historical user node representation to obtain a target user node embedding vector, and update the product node embedding vector based on the historical product node representation to obtain a target product node embedding vector;

[0103] It can be understood that the user node embedding vectors include the node embedding vectors corresponding to one or more users, and the historical user node representations also include the historical node representations corresponding to one or more of the same users. Thus, the node embedding vectors of each user are updated respectively based on the historical node representations corresponding to each user to obtain the target node embedding vectors corresponding to each user, and the target node embedding vectors corresponding to all users form the target user node embedding vectors. For example, assuming that the historical user node representations are [historical node representation 1, historical node representation 2], and the user node embedding vectors are [node embedding vector 1, node embedding vector 2], then the node embedding vector 1 is updated based on the historical node representation 1 to obtain the target node embedding vector 1, and the node embedding vector 2 is updated based on the historical node representation 2 to obtain the target node embedding vector 2. The target user node embedding vector can be expressed as [target node embedding vector 1, target node embedding vector 2].

[0104] The commodity node embedding vectors can be updated in a manner similar to the update of the user node embedding vectors, which will not be elaborated here again.

[0105] Furthermore, the update method can be set in advance. For example, in a possible implementation manner, step A20 includes:

[0106] Step A201: perform a weighted sum of the historical user node representation and the user node embedding vector to obtain the target user node embedding vector;

[0107] For the weighted sum of the historical user node representation and the user node embedding vector, specifically, the weights corresponding to the historical user node representation and the user node embedding vector can be set in advance, and the historical user node representation and the user node embedding vector are weighted and summed with their respective corresponding weights.

[0108] Furthermore, after the weighted sum, the sum result can be activated by an activation function, and the output of the activation function is determined as the target node embedding vector. Specifically, it can be expressed by the formula U i (t + 1) = σ(W1u i (t + 1) + W2e i (t)), where σ(·) represents the sigmoid activation function, W1 and W2 respectively represent weight matrices, u i (t + 1) represents the node embedding vector of user i at time t + 1 (which can be understood as the time when user data and cultural and tourism commodity data are obtained), U i (t + 1) represents the target node embedding vector of user i at time t + 1, and e i (t) is the node representation of user i at time t, that is, the historical node representation of user i.

[0109] Step A202: Perform weighted summation on the historical commodity node representation and the commodity node embedding vector to obtain the target commodity node embedding vector;

[0110] The target commodity node embedding vector can be obtained in a similar manner to obtaining the target user node vector, which will not be elaborated here again. Similarly, it can also be expressed by the formula G j (t + 1) = σ(W1g j (t + 1) + W2ejt, gjt+1 represents the node embedding vector of cultural and tourism commodity j at time t + 1, Gj(t + 1) represents the target node embedding vector of cultural and tourism commodity j at time t + 1, and e j (t) is the node representation of cultural and tourism commodity j at time t, that is, the historical node representation of cultural and tourism commodity j.

[0111] Step A30: Use the target user node embedding vector as the new user node embedding vector, and use the target commodity node embedding vector as the new commodity node embedding vector. Based on the new user node embedding vector and the new commodity node embedding vector, perform the step of inputting the user node embedding vector and the commodity node embedding vector into the pre-trained cultural and tourism commodity recommendation model to obtain the cultural and tourism commodity recommendation result.

[0112] In this embodiment, the target node embedding vector is used as the new user node embedding vector, and the target commodity node embedding vector is used as the new commodity node embedding vector and input into the pre-trained cultural and tourism commodity recommendation model to perform cultural and tourism commodity recommendation based on the new user node embedding vector and the new commodity node embedding vector. In this way, the user node embedding vector is updated based on the historical user node representation, and the commodity node embedding vector is updated based on the historical commodity node representation. It can be understood that the historical data and the latest data of users and cultural and tourism commodities are fused, so as to perform cultural and tourism commodity recommendation based on richer information, which can further improve the accuracy of the cultural and tourism commodity recommendation result.

[0113] Furthermore, after updating the node embedding vectors of users and cultural and tourism commodities, the node representations of users and cultural and tourism commodities can also be updated. For example, the updated node embedding vectors of users and cultural and tourism commodities, that is, the target node embedding vectors, are determined as the new node representations of users and cultural and tourism commodities, and the node representations of users and cultural and tourism commodities stored are correspondingly updated. Relevant personnel can also set other node representation update methods based on actual needs, and this embodiment does not make specific restrictions on this.

[0114] Considering that traditional recommendation methods are still unable to dynamically and intelligently learn the potential correlation between users and cultural and tourism products. Specifically, traditional recommendation methods rely on clustering and matrix methods, which rely on a large number of manually set parameters. When facing different cultural and tourism products, they rely on manual experience and are unable to intelligently learn the complex interaction relationship between user features and cultural and tourism product features. Moreover, the existing methods lack attention to the dynamic relationship between users and cultural and tourism products and are unable to capture new changes in users' purchases of cultural and tourism products in a timely manner, resulting in poor timeliness of recommendation results.

[0115] Based on this and the first and / or second embodiments of the present application, in the third embodiment of the present application, the same or similar content as in the first and second embodiments above can be referred to the above introduction and will not be repeated hereinafter. On this basis, the cultural and tourism product recommendation model includes a dynamic graph network and a prediction network. The step of inputting the user node embedding vector and the product node embedding vector into the pre-trained cultural and tourism product recommendation model to obtain the cultural and tourism product recommendation result includes:

[0116] Step B10, input the user node embedding vector and the product node embedding vector into the dynamic graph network to obtain the current user representation and the current product representation, where the dynamic graph network is used to update the node representations of the nodes in the user-product interaction graph, where the node representation is the value state space representation of the node, the user-product interaction graph includes user nodes and product nodes, the user node is the node representing the user, and the product node is the node representing the cultural and tourism product;

[0117] It should be noted that the current user representation and the current product representation may specifically be the latest value state space representations of the user and the cultural and tourism product.

[0118] Step B20, input the current user representation and the current product representation into the prediction network to obtain the cultural and tourism product recommendation result.

[0119] The cultural and tourism product recommendation result may specifically be the output of the prediction network. Relevant personnel can set the network structure of the prediction network based on actual needs, and this embodiment does not make specific limitations on this. For example, in a specific implementation manner, the prediction network is a fully connected layer.

[0120] In a possible implementation manner, the step of inputting the user node embedding vector and the product node embedding vector into the dynamic graph network to obtain the current user representation and the current product representation includes:

[0121] Step C10, input the user node embedding vector and the product node embedding vector into the dynamic graph network, and obtain or construct a user-product interaction graph through the dynamic graph network;

[0122] The user-commodity interaction diagram can specifically be a schematic diagram indicating the interaction relationship between a user and a commodity. The interaction relationship can specifically be relationships such as purchase, browsing, adding to cart, and favoriting. Exemplarily, the following will illustrate with the interaction relationship being purchase.

[0123] A user-commodity interaction diagram can be constructed based on the obtained user data and the user-commodity interaction relationship corresponding to the cultural and tourism commodity data. The constructed user-commodity interaction diagram includes user nodes and commodity nodes. Each user node represents a user, and each commodity node represents a cultural and tourism commodity. If a user purchases a cultural and tourism commodity, there is a connection edge between the user node corresponding to this user and the commodity node corresponding to this cultural and tourism commodity. Conversely, if a user does not purchase a cultural and tourism commodity, there is no connection edge between the user node corresponding to this user and the commodity node corresponding to this cultural and tourism commodity. For example, assume that the obtained user data comes from User 1 and User 2, and the cultural and tourism commodity data comes from Cultural and Tourism Commodity A, Cultural and Tourism Commodity B, and Cultural and Tourism Commodity C. The corresponding user-commodity interaction relationship is that User 1 purchases Cultural and Tourism Commodity A, and User 2 purchases Cultural and Tourism Commodity B and Cultural and Tourism Commodity C. Then the constructed user-commodity interaction diagram includes Node 1 representing User 1, Node 2 representing User 2, Node A representing Cultural and Tourism Commodity A, Node B representing Cultural and Tourism Commodity B, and Node C representing Cultural and Tourism Commodity C, and there is a connection edge between Node 1 and Node A, a connection edge between Node 2 and Node B, and a connection edge between Node 2 and Node C.

[0124] Step C20, determine the node corresponding to the user data in the user-commodity interaction diagram as the target user node, and determine the node corresponding to the cultural and tourism commodity data in the commodity interaction diagram as the target commodity node;

[0125] The node corresponding to the user data can specifically be the node corresponding to the source user of the user data. For example, assume that the user data comes from User 1 and User 2. Then the nodes corresponding to the user data are the node of User 1 in the commodity interaction diagram (denoted as Node 1) and the node of User 2 in the commodity interaction diagram (denoted as Node 2). The target user node can be represented as [Node 1, Node 2].

[0126] The method for determining the target commodity node is similar to that of the target user node, and will not be repeated here.

[0127] Step C30, update the node representation of the target user node based on the user node embedding vector to obtain the current node representation of the target user node;

[0128] Relevant personnel can pre-set a specific node representation update method, such as directly updating the node representation of the target user node to the user node embedding vector, or updating the node representation of the target user node based on the graph attention mechanism.

[0129] Update the node representation of the target user node based on the user node embedding vectors. Specifically, update the node representation of the node corresponding to each user based on the node embedding vectors of each user. For example, assume that the user node embedding vectors include [node embedding vector 1, node embedding vector 2], and the target user node includes [node 1, node 2]. Then, update the node representation of node 1 based on node embedding vector 1, and update the node representation of node 2 based on node embedding vector 2.

[0130] Step C40: Update the node representation of the target commodity node based on the commodity node embedding vectors to obtain the node representation of the target commodity node.

[0131] The way to update the node representation of the target commodity node is similar to the way to update the node representation of the target user node, and will not be repeated here.

[0132] Step C50: Determine the current node representation of the target user node as the current user representation, and determine the current node representation of the target commodity node as the current commodity representation.

[0133] It can be understood that the current user representation includes the current node representations of one or more users, and the current commodity representation includes the current node representations of one or more cultural and tourism commodities, so as to perform cultural and tourism commodity recommendations based on the current node representations of each user and the current node representations of each cultural and tourism commodity.

[0134] In a possible implementation manner, the step of updating the node representation of the target user node based on the user node embedding vectors to obtain the current node representation of the target user node includes:

[0135] Step D10: Determine the neighbor nodes adjacent to the target user node in the user commodity interaction graph, and obtain the neighbor node embedding vectors of the neighbor nodes.

[0136] It should be noted that the neighbor nodes adjacent to the target user node can specifically be the nodes that have connection edges with the target user node in the commodity interaction graph. Further, taking each node corresponding to each user in the target user node as an atom, determine the neighbor nodes adjacent to this node, and continue to execute the subsequent steps to update the node representation of each node in the target user node. After the node representations of all nodes in the target user node are updated, that is, the update of the node representation of the target user node is completed, and the updated node representations of all users constitute the current node representation.

[0137] Step D20: Aggregate the user node embedding vectors and the neighbor node embedding vectors based on the graph attention mechanism to obtain the current node representation of the target user node.

[0138] The Graph Attention Mechanism is an attention mechanism applied to graph-structured data. It draws on the idea of the attention mechanism in natural language processing and is used to process node relationships and features in graph neural networks. The core idea of the graph attention mechanism is to assign different attention weights between nodes in the graph, enabling the model to pay more attention to important nodes or edges, thereby improving the model's expressive ability and performance.

[0139] In a specific embodiment, the specific steps of graph attention mechanism message aggregation are as follows:

[0140] (1) As shown in (Formula 1), use shared parameters W1 and W2 to increase the dimension of the node features. Then, for the source node (user node) U i and one of its corresponding neighbor nodes (cultural and tourism commodity node) G j concatenate (||) the transformed features. Finally, use the feed-forward neural network A to map the concatenated high-dimensional features to a real number to obtain the similarity coefficient e ij .

[0141] e ij = A(W1U i ||W2G j )(Formula 1).

[0142] (2) As shown in (Formula 2), normalize the similarity coefficient. a ij is the normalized similarity coefficient between nodes i and j, LeakyReLu is the activation function, e in is the similarity coefficient between the source node i and its respective neighbor nodes such as j, k, m, and exp is the exponential function.

[0143]

[0144] (3) As shown in (Formula 3) and (Formula 4), use the multi-head attention mechanism to weight-aggregate the neighbor features of node i to node i to obtain the new representation of node i. In (Formula 3), R n is the representation of node i after single-head attention mechanism aggregation, a ij is the normalized similarity coefficient between nodes i and j, W i is the shared weight parameter of node i, and n is the number of attention heads. In (Formula 4), A ′ is the representation of node i after multi-head attention mechanism aggregation, that is, the new node representation of node i, N is the number of attention heads, σ(·) represents the sigmoid activation function, and R is the representation of node i after single-head attention mechanism aggregation.

[0145]

[0146] The node representation of the target commodity node can be updated based on the graph attention mechanism in a manner similar to the node representation of the updated target user node, which will not be repeated here.

[0147] Exemplarily, to facilitate understanding of the update process of the node representation in this embodiment, a specific embodiment is now enumerated. In this specific embodiment, refer to Figure 3 as shown. Assume that at time t+1, the user node interacts with the cultural and tourism commodity node, that is, the user purchases a certain cultural and tourism commodity. At this time, cultural and tourism commodity recommendation is required.

[0148] First, obtain the latest user embedding (i.e., the user node embedding vector) and the cultural and tourism commodity embedding (i.e., the commodity node embedding vector). The embedding of user i input into the cultural and tourism recommendation model at time t+1 (i.e., the target user node embedding) is obtained by weighting the latest user embedding and the user representation at time t.

[0149] Secondly, for the cultural and tourism commodity nodes at each moment, the data comes from the records of users purchasing cultural and tourism commodities on the cultural and tourism platform. This record includes attributes such as the user's unique anonymous ID, the type of cultural and tourism commodity, and the price. Input the data attributes of the cultural and tourism commodity into the multi-layer perceptron. The input dimension of the multi-layer perceptron is the same as the dimension of the input cultural and tourism commodity attribute values. The hidden layer dimension is 2*dim, 2*dim, and the output dimension is dim, such as 512 dimensions. Thus, the commodity node embedding is obtained. The embedding of cultural and tourism commodity j input into the cultural and tourism recommendation model at time t+1 (i.e., the target commodity node embedding) is obtained by weighting the latest embedding of cultural and tourism commodity j (i.e., the commodity node embedding) and the cultural and tourism commodity representation at time t.

[0150] Next, use the graph attention mechanism ( Figure 3 the graph attention layer shown in) to perform message aggregation on the user node and the cultural and tourism commodity node (abbreviated as the commodity node) at time t+1, respectively, to obtain the user representation (i.e., the current user representation) and the cultural and tourism commodity representation (i.e., the current commodity node representation) at time t+1.

[0151] After aggregation by the graph attention mechanism, new representations of the user node and the cultural and tourism commodity node at time t+1 are obtained. Then, use a fully connected layer model to predict the connection relationship between the user node and the cultural and tourism commodity node, that is, predict whether the user is likely to purchase a certain cultural and tourism commodity. Specifically, f UG = Sigmoid(W(ReLU(U i ||G j )) is used for prediction. Among them, f UG is the prediction result, Sigmoid is the activation function, ReLU is the activation function, W is the fully connected layer function, U i ||G jis the result of concatenating the node representation of user i and the node representation of cultural and tourism product j.

[0152] It should be noted that the above examples are only used to assist in understanding this embodiment and do not constitute a limitation on the way of updating the node representation in this embodiment. Based on this technical concept, more simple transformations in various forms are within the protection scope of this application.

[0153] Based on the first embodiment, the second embodiment, and / or the third embodiment of this application, in the fourth embodiment of this application, the content that is the same as or similar to the above-mentioned first embodiment, second embodiment, and third embodiment can be referred to the above introduction and will not be elaborated hereinafter. On this basis, the offline behavior data includes at least one trajectory point, and the step of vectorizing the offline behavior data to obtain an offline behavior embedding vector includes:

[0154] Step E10, input each of the trajectory points into a pre-trained spatio-temporal large model to obtain an output vector corresponding to each of the trajectory points, where the spatio-temporal large model includes at least one layer of encoding units, and each layer of the encoding units includes an attention layer, a normalization layer, a feed-forward neural network, and a normalization layer connected in sequence;

[0155] Each trajectory point can specifically be a trajectory point composed of longitude and latitude, that is, each trajectory point can be expressed by coordinates. Input each of the trajectory points into a pre-trained spatio-temporal large model to perform embedding processing on each of the trajectory points through the spatio-temporal large model.

[0156] The spatio-temporal large model refers to a complex model that can process and analyze spatio-temporal data. Relevant personnel can preset the specific structure of the spatio-temporal large model. For example, in a specific implementation manner, refer to Figure 4As shown, the spatio-temporal large model includes a multi-layer perceptron, a position vectorization module, and a Decoder-Only Transformers structure connected in sequence. The encoding unit of a Transformer in the Decoder-Only Transformers structure is generated by stacking a multi-head attention layer (Multi-head-Attention), layer normalization (Layer Normalization), a feedforward neural network (Feedforword), and layer normalization (Layer Normalization). Among them, layer normalization performs addition and normalization (Add&Norm) processing, and each layer in the Decoder-Only Transformers is composed of such an encoding unit. In a relatively large spatio-temporal large model, there are 24 encoding layers, with 16 attention units in each layer, and the input and output vector dimensions are 1024. In a relatively small Bert model, there are 12 encoding layers, with 12 attention units in each layer, and the input and output vector dimensions are 512. Input each trajectory point [(x1, y1), (x2, y2), (x3, y3)......(xn, yn)] into the multi-layer perceptron. The multi-layer perceptron performs mapping processing on each trajectory point to obtain a trajectory point embedding. Input the trajectory point embedding into the position encoding model. The position encoding model performs position encoding processing (i.e., position embedding) on the trajectory point embedding to obtain an input vector. The input vector is input into the Decoder-Only Transformers to output an output vector.

[0157] It should be noted that when training the spatio-temporal large model, in order to facilitate the calculation of the loss value to optimize the model parameters, a mapping conversion layer is connected after the Decoder-Only Transformers. The output vector of the Decoder-Only Transformers is input into this mapping conversion layer. The mapping conversion layer adopts the mapping method of a fully connected layer neural network and the softmax function to realize the transformation of the output vector to the predicted position, and obtain the predicted positions corresponding to each trajectory point [(x2’, y2’), (x3’, y3’), (x4’, y4’)......(xn+1’, yn+1’)]. Among them, in the fully connected layer neural network, each neuron in each layer is connected to all neurons in the previous layer and all neurons in the next layer, forming a "fully connected" structure, that is, a feedforward neural network (Feedforward Neural Network, FFNN). The input layer dimension of the fully connected layer neural network is 2, the hidden layer dimension is 2*dim, 2*dim, and the output dimension is dim, such as 512 dimensions. The Softmax function maps the elements in the input vector to longitude and latitude normalization values between 0 and 1, and the calculation method is shown in (Formula 5):

[0158]

[0159] In the formula, σ is the softmax function, z is the vector of the input function, is the exponential function, K is the number of categories, i represents the i-th trajectory point, and n represents the total number of trajectory points. This Decoder-Only Transformers structure can predict the position of the next point according to the input trajectory point sequence. Finally, the position probability value output by Softmax is restored to the actual longitude and latitude value by (Formula 6) and (Formula 7), as follows:

[0160] x’ = x min +(x max -x min )*σ x (Formula 6);

[0161] y’ = y min +(y max -y min )*σ y (Formula 7).

[0162] Among them, x’, y’ are the longitude and latitude values of the predicted position, x min 、x max 、y min 、y max are the minimum and maximum values of the longitude and latitude (such as the longitude and latitude of a certain city), σ xand σ y are the longitude and latitude normalized values output by the Softmax function.

[0163] Step E20, calculate the average vector of each of the output vectors, and determine the average vector as the offline behavior embedding vector.

[0164] Considering the high correlation between the user's offline spatio-temporal trajectory and the cultural and tourism commodity purchase activities, in this embodiment, a spatio-temporal large model is used to model and learn the behavior preferences reflected by the user's spatio-temporal trajectory, so as to realize the extraction of the user's general offline behavior representation and deeply integrate it into the complete cultural and tourism commodity recommendation, solving the problem that the traditional cultural and tourism commodity recommendation method insufficiently considers the user's offline behavior and is not fully integrated.

[0165] Exemplarily, in order to help understand the technical concept or technical principle of the cultural and tourism commodity recommendation method after combining this embodiment with the first embodiment, the second embodiment, and the third embodiment, a specific embodiment is now listed. In this specific embodiment, first, the spatio-temporal large model and the dynamic graph model are trained. The dynamic graph model is specifically a model composed of a multi-layer perceptron connected before the input of the cultural and tourism recommendation model. It can be understood that the dynamic graph model includes a multi-layer perceptron model and a cultural and tourism recommendation model connected in sequence.

[0166] Spatio-temporal large model pre-training: Input the signaling trajectory points or GPS trajectory points of a single user within a period of time one by one, and then the model predicts the trajectory points at the next moment based on the input trajectory points and calculates the loss function. The loss function is the distance between the longitude and latitude of the predicted point and the actual point, as shown in (Formula 8):

[0167]

[0168] where, e ij is the distance between the predicted point i and the actual point j, and x i , x j are the longitudes of the predicted point i and the actual point j respectively, and y i , y j are the latitudes of the predicted point i and the actual point j respectively. After obtaining the prediction error value e ij , use the Adam optimization algorithm to update the neural network parameters in the spatio-temporal large model until the neural network parameters converge.

[0169] (3) Dynamic graph model pre-training: Input the offline behavior embedding vector, online behavior data, and cultural and tourism commodity data of the user respectively, and then the model calculates the representations of the user node and the cultural and tourism commodity node and predicts whether the user will purchase a certain type of cultural and tourism commodity. This task belongs to a binary classification task, and the cross entropy is used as the loss function. The function formula is as shown in (Formula 9):

[0170]

[0171] In the formula, θ is the parameter of the encoding part of the dynamic graph model, θ2 is the parameter of the classifier connected to the encoding part for the prediction task, N is the logical judgment value 0 or 1, and 0 and 1 represent that the user will not / will purchase a certain type of cultural and tourism product respectively. After obtaining the prediction error value L(θ,θ2), the neural network parameters in the dynamic graph model are updated using the Adam optimization algorithm until the neural network parameters converge. Based on the trained spatio-temporal large model and dynamic graph model, please refer to Figure 5 As shown, the cultural and tourism product recommendation method includes:

[0172] Obtain cultural and tourism product data, as well as the offline behavior data (the offline trajectory data shown in Figure 5 ) and online behavior data of the user. Input the cultural and tourism product data and the online behavior data into a multi-layer perceptron model, and through the multi-layer perceptron, perform embedding processing on the cultural and tourism product data and the online behavior data respectively to obtain a product node embedding vector (the cultural and tourism product embedding shown in Figure 5 ) and an online behavior embedding vector. Input the offline behavior data into the spatio-temporal large model to obtain an offline behavior embedding vector, and fuse the online behavior embedding vector and the offline behavior embedding vector to obtain a user node embedding vector (the user node embedding output in Figure 5 ). Input the user node embedding vector and the product node embedding vector into the dynamic graph network of the cultural and tourism recommendation model (the graph neural network model shown in Figure 5 ), and output to obtain the current user representation (the user node representation shown in Figure 5 ) and the current product representation (the cultural and tourism product node representation shown in Figure 5 ). Input the current user representation and the current product representation into the prediction network of the cultural and tourism recommendation model (the fully connected layer model shown in Figure 5 ), and output to obtain the cultural and tourism recommendation result, and obtain a recommendation list based on the cultural and tourism recommendation result.

[0173] It should be noted that the above examples are only used to assist in understanding the present application and do not constitute a limitation on the cultural and tourism product recommendation method of the present application. Based on this technical concept, more forms of simple transformations are within the protection scope of the present application.

[0174] In addition, an embodiment of the present application also proposes a cultural and tourism product recommendation device. Referring to Figure 6 As shown, the cultural and tourism product recommendation device includes:

[0175] An acquisition module 10, configured to acquire user data and cultural and tourism product data, wherein the user data includes online behavior data and offline behavior data;

[0176] A vectorization module 20 is configured to perform vectorization processing on the online behavior data to obtain an online behavior embedding vector, perform vectorization processing on the offline behavior data to obtain an offline behavior embedding vector, and perform vectorization processing on the cultural and tourism commodity data to obtain a commodity node embedding vector, wherein the vector dimensions of the online behavior embedding vector and the offline behavior embedding vector are the same;

[0177] A fusion module 30 is configured to fuse the online behavior embedding vector and the offline behavior embedding vector to obtain a user node embedding vector;

[0178] A recommendation module 40 is configured to input the user node embedding vector and the commodity node embedding vector into a pre-trained cultural and tourism commodity recommendation model to obtain a cultural and tourism commodity recommendation result, and perform cultural and tourism commodity recommendation according to the cultural and tourism commodity recommendation result.

[0179] In addition, an embodiment of the present application further provides a cultural and tourism commodity recommendation device, where the cultural and tourism commodity recommendation device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the cultural and tourism commodity recommendation method as described above.

[0180] Reference Figure 7 , which shows a schematic structural diagram of a cultural and tourism commodity recommendation device suitable for implementing an embodiment of the present application. The cultural and tourism commodity recommendation device in the embodiment of the present application may further include, but is not limited to, mobile terminals such as servers, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 7 The shown cultural and tourism commodity recommendation device is merely an example and should not impose any limitation on the functions and usage scope of the embodiment of the present application.

[0181] As Figure 7As shown, the cultural and tourism product recommendation device may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM: Read Only Memory) 1002 or the program loaded from the storage device 1003 into the random access memory (RAM: Random Access Memory) 1004. In the RAM 1004, various programs and data required for the operation of the cultural and tourism product recommendation device are also stored. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. The input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems may be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the cultural and tourism product recommendation device to communicate with other devices wirelessly or wiredly to exchange data. Although the figure shows a cultural and tourism product recommendation device with various systems, it should be understood that it is not required to implement or have all the shown systems. More or fewer systems can be alternatively implemented or had.

[0182] Particularly, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device, or installed from the storage device 1003, or installed from the ROM 1002. When the computer program is executed by the processing device 1001, the above functions defined in the methods of the embodiments disclosed in the present application are executed.

[0183] The cultural and tourism product recommendation device provided by the embodiments of the present application adopts the cultural and tourism product recommendation method in the above embodiments, and can solve the technical problem of how to improve the accuracy of the cultural and tourism product recommendation results. Compared with the prior art, the beneficial effects of the cultural and tourism product recommendation device provided by the present application are the same as those of the cultural and tourism product recommendation method provided by the above embodiments, and the other technical features in the cultural and tourism product recommendation device are the same as the features disclosed in the method of the previous embodiment, and will not be elaborated here.

[0184] It should be understood that each part disclosed in this application can be implemented by hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0185] As described above, the above are only specific embodiments of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application, and all of them should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0186] In addition, to achieve the above object, an embodiment of this application also provides a readable storage medium, having computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the cultural and tourism commodity recommendation method in the above embodiments.

[0187] The computer-readable storage medium provided by the embodiment of this application can be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this embodiment, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, device, or device. The program code contained on the computer-readable storage medium can be transmitted by any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0188] The above computer-readable storage medium can be included in the cultural and tourism commodity recommendation device; it can also exist separately and not be assembled into the cultural and tourism commodity recommendation device.

[0189] The above computer-readable storage medium carries one or more programs, which, when executed by the cultural and tourism commodity recommendation device, cause the cultural and tourism commodity recommendation device to: obtain user data and cultural and tourism commodity data, wherein the user data includes online behavior data and offline behavior data; perform vectorization processing on the online behavior data to obtain an online behavior embedding vector, perform vectorization processing on the offline behavior data to obtain an offline behavior embedding vector, and perform vectorization processing on the cultural and tourism commodity data to obtain a commodity node embedding vector, wherein the vector dimensions of the online behavior embedding vector and the offline behavior embedding vector are the same; fuse the online behavior embedding vector and the offline behavior embedding vector to obtain a user node embedding vector; input the user node embedding vector and the commodity node embedding vector into a pre-trained cultural and tourism commodity recommendation model to obtain a cultural and tourism commodity recommendation result, and perform cultural and tourism commodity recommendation according to the cultural and tourism commodity recommendation result.

[0190] Computer program code for performing the operations of this application can be written in one or more programming languages or combinations thereof. The above programming languages include object-oriented programming languages - such as Java, Smalltalk, C++, and also include conventional procedural programming languages - such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network - including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0191] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a part of code, which contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0192] The modules described in the embodiments of the present application can be implemented in software or in hardware. In this regard, the name of a module does not constitute a limitation on the module itself in some cases.

[0193] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned cultural and tourism commodity recommendation method, and can solve the technical problem of how to improve the accuracy of the cultural and tourism commodity recommendation result. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as those of the cultural and tourism commodity recommendation method provided by the above embodiments, and will not be elaborated here.

[0194] In addition, an embodiment of the present application also proposes a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the steps of the cultural and tourism commodity recommendation method as described above.

[0195] The specific implementation manners of the computer program product of the present application are basically the same as those of the above embodiments of the cultural and tourism commodity recommendation and / or voice enhancement method, and will not be elaborated here.

[0196] It should be noted that in this article, the term "including", "comprising", or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article, or system including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article, or system. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article, or system including the element.

[0197] The serial numbers of the embodiments of the present application above are only for description and do not represent the advantages or disadvantages of the embodiments.

[0198] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software sensor. This computer software sensor is stored in a storage medium as described above (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to enable a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in various embodiments of the present application.

[0199] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.

Claims

1. A method for recommending cultural and tourism products, characterized in that, The cultural and tourism product recommendation method includes the following steps: Obtain user data and cultural and tourism product data, where the user data includes online behavior data and offline behavior data; Perform vectorization processing on the online behavior data to obtain an online behavior embedding vector, perform vectorization processing on the offline behavior data to obtain an offline behavior embedding vector, and perform vectorization processing on the cultural and tourism product data to obtain a product node embedding vector, where the vector dimensions of the online behavior embedding vector and the offline behavior embedding vector are the same; Fuse the online behavior embedding vector and the offline behavior embedding vector to obtain a user node embedding vector; Input the user node embedding vector and the product node embedding vector into a pre-trained cultural and tourism product recommendation model to obtain a cultural and tourism product recommendation result, and perform cultural and tourism product recommendation according to the cultural and tourism product recommendation result.

2. The cultural and tourism product recommendation method according to claim 1, wherein Before the step of inputting the user node embedding vector and the product node embedding vector into a pre-trained cultural and tourism product recommendation model to obtain a cultural and tourism product recommendation result, the method further includes: Obtain the historical user node representation corresponding to the user data, and obtain the historical product node representation corresponding to the cultural and tourism product data, where the historical user node representation is the historical value state space representation of the user data, and the historical product node representation is the historical value state space representation of the cultural and tourism product data; Update the user node embedding vector based on the historical user node representation to obtain a target user node embedding vector, and update the product node embedding vector based on the historical product node representation to obtain a target product node embedding vector; Use the target user node embedding vector as the new user node embedding vector, use the target product node embedding vector as the new product node embedding vector, and perform the step of inputting the user node embedding vector and the product node embedding vector into a pre-trained cultural and tourism product recommendation model to obtain a cultural and tourism product recommendation result based on the new user node embedding vector and the new product node embedding vector.

3. The cultural and tourism product recommendation method according to claim 2, wherein The step of updating the user node embedding vector based on the historical user node representation to obtain a target user node embedding vector and updating the product node embedding vector based on the historical product node representation to obtain a target product node embedding vector includes: Perform weighted summation on the historical user node representation and the user node embedding vector to obtain a target user node embedding vector; Perform weighted summation on the historical product node representation and the product node embedding vector to obtain a target product node embedding vector.

4. The cultural and tourism product recommendation method according to any one of claims 1 to 3, characterized in that The cultural and tourism product recommendation model includes a dynamic graph network and a prediction network. The step of inputting the user node embedding vector and the product node embedding vector into a pre-trained cultural and tourism product recommendation model to obtain a cultural and tourism product recommendation result includes: Input the user node embedding vector and the product node embedding vector into the dynamic graph network to obtain the current user representation and the current product representation, where the dynamic graph network is used to update the node representations of the nodes in the user-product interaction graph, where the node representation is the value state space representation of the node, the user-product interaction graph includes user nodes and product nodes, the user node is a node representing a user, and the product node is a node representing a cultural and tourism product; Input the current user representation and the current product representation into the prediction network to obtain the cultural and tourism product recommendation result.

5. The cultural and tourism product recommendation method according to claim 4, wherein The step of inputting the user node embedding vector and the product node embedding vector into the dynamic graph network to obtain the current user representation and the current product representation includes: Input the user node embedding vector and the product node embedding vector into the dynamic graph network, and obtain or construct a user-product interaction graph through the dynamic graph network; Determine the node corresponding to the user data in the user-product interaction graph as the target user node, and determine the node corresponding to the cultural and tourism product data in the product interaction graph as the target product node; Update the node representation of the target user node based on the user node embedding vector to obtain the current node representation of the target user node; Update the node representation of the target product node based on the product node embedding vector to obtain the node representation of the target product node; Determine the current node representation of the target user node as the current user representation, and determine the current node representation of the target product node as the current product representation.

6. The cultural and tourism product recommendation method according to claim 5, wherein, The step of updating the node representation of the target user node based on the user node embedding vector to obtain the current node representation of the target user node includes: Determine the neighbor nodes adjacent to the target user node in the user-product interaction graph, and obtain the neighbor node embedding vectors of the neighbor nodes; Aggregate the user node embedding vector and the neighbor node embedding vectors based on the graph attention mechanism to obtain the current node representation of the target user node.

7. The cultural and tourism product recommendation method according to claim 1, wherein The offline behavior data includes at least one trajectory point. The step of vectorizing the offline behavior data to obtain an offline behavior embedding vector includes: Input each of the trajectory points into a pre-trained spatio-temporal large model to obtain an output vector corresponding to each of the trajectory points, where the spatio-temporal large model includes at least one layer of encoding units, and each layer of the encoding units includes an attention layer, a normalization layer, a feed-forward neural network, and a normalization layer connected in sequence; Calculate the average vector of each of the output vectors, and determine the average vector as the offline behavior embedding vector.

8. A cultural and tourism commodity recommendation device, characterized in that, The cultural and tourism product recommendation device includes: An acquisition module, configured to acquire user data and cultural and tourism product data, where the user data includes online behavior data and offline behavior data; A vectorization module, configured to perform vectorization processing on the online behavior data to obtain an online behavior embedding vector, perform vectorization processing on the offline behavior data to obtain an offline behavior embedding vector, and perform vectorization processing on the cultural and tourism commodity data to obtain a commodity node embedding vector, wherein the vector dimensions of the online behavior embedding vector and the offline behavior embedding vector are the same; A fusion module, configured to fuse the online behavior embedding vector and the offline behavior embedding vector to obtain a user node embedding vector; A recommendation module, configured to input the user node embedding vector and the commodity node embedding vector into a pre-trained cultural and tourism commodity recommendation model to obtain a cultural and tourism commodity recommendation result.

9. A cultural and tourism commodity recommendation device, characterized in that, The device includes: a memory, a processor, and a cultural and tourism commodity recommendation program stored on the memory and executable on the processor, where the cultural and tourism commodity recommendation program is configured to implement the steps of the cultural and tourism commodity recommendation method according to any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium includes a computer-readable storage medium, on which a cultural and tourism commodity recommendation program is stored, and when the cultural and tourism commodity recommendation program is executed by a processor, it implements the steps of the cultural and tourism commodity recommendation method according to any one of claims 1 to 7.

11. A computer program product, characterized in that, The computer program product includes a cultural and tourism commodity recommendation program, and when the cultural and tourism commodity recommendation program is executed by a processor, it implements the steps of the cultural and tourism commodity recommendation method according to any one of claims 1 to 7.

Images