A method and system for recommending points of interest

Abstract

The present invention provides a method and system for recommending points of interest, wherein the method includes: obtaining user information of a target user and item information searched by the target user as user-item interaction data; inputting the user-item interaction data into a point of interest recommendation model to obtain the point of interest item data recommended for the target user output by the point of interest recommendation model; wherein the point of interest recommendation model is trained based on the similarity between users and items obtained from user-item interaction data samples using an attention mechanism, and the similarity between users and items obtained from the user-item interaction data samples according to preset rules. The method is used to solve the defect of low recommendation accuracy due to data sparsity in the prior art of similarity measurement based on user-item interaction information, and achieve accurate recommendation of point of interest items for target users.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and system for recommending points of interest. Background Art

[0002] In the field of recommendation systems, as location-based services become increasingly popular, point of interest recommendations have attracted more and more attention. Traditional methods basically use a variety of contextual features to improve the recommendation effect, including temporal information, geographic location, or social relationships. These auxiliary information are used to model the behavioral similarities between users and items. However, these methods only implicitly measure similarity by using user-item interaction information. Therefore, they are plagued by data sparsity problems. When there is no interaction between users and items, the dependency relationship between users and items cannot be obtained.

[0003] Based on the defect of implicitly measuring similarity by utilizing user-item interaction information, there are currently methods that explicitly model the similarity between users and items. However, these methods either use descriptive features or use heuristic designs to manually define similarity measures, so they are also affected by insufficient auxiliary information and lack generalization capabilities. Summary of the invention

[0004] The present invention provides a method and system for recommending points of interest, which are used to solve the defect in the prior art of similarity measurement based on user-item interaction information, resulting in low recommendation accuracy due to data sparsity, and achieve accurate recommendation of points of interest items for target users.

[0005] The present invention also provides a method for recommending points of interest, comprising:

[0006] Acquire user information of a target user and information of items searched by the target user as user item interaction data;

[0007] Inputting the user entry interaction data into an interest point recommendation model to obtain interest point entry data recommended for the target user output by the interest point recommendation model;

[0008] Among them, the point of interest recommendation model is trained based on the similarities between users and items obtained from user-item interaction data samples using the attention mechanism, and the similarities between users and items obtained from the user-item interaction data samples according to preset rules.

[0009] According to the interest point recommendation method described in the present invention, the loss function of the interest point recommendation model is determined based on the degree of association between a user and an item associated with the user obtained from the user-item interaction data sample using a link prediction mechanism, and the similarity between users and users, and between items obtained from the user-item interaction data sample using an attention mechanism, and the degree of difference between the similarity between users and users, and between items obtained from the user-item interaction data sample according to preset rules.

[0010] According to the method for recommending points of interest of the present invention, the degree of association between the user and the items associated with the user includes:

[0011] The distance difference between the user and the entries associated with the user.

[0012] According to the interest point recommendation method of the present invention, the loss function of the interest point recommendation model is:

[0013]

[0014] in, is a function of the degree of association; is a function of the degree of difference; α is a preset hyperparameter representing the weight;

[0015] The function of the degree of association is:

[0016]

[0017] The function of the degree of difference is:

[0018]

[0019] Wherein, u and v represent users and items respectively; (u, v) and (u′, v′) are real user-item interaction pairs and fake user-item interaction pairs of the user-item interaction data sample respectively; and are the set of real user-item interaction pairs and the set of fake user-item interaction pairs respectively; γ is a preset hyperparameter representing the calculation boundary; and are the sum of the similarity differences between the interactions between user i and user j and all items, and between item i and item j and all users, respectively, obtained according to the preset rules; m and n are the total number of users and the total number of items in the user-item interaction data sample, respectively; and They are the similarities between user i and user j, and item i and item j obtained using the attention mechanism.

[0020] According to the method for recommending points of interest of the present invention, the sum of the similarity differences between the interactions between user i and user j and all items, and between item i and item j and all users obtained according to the preset rules are respectively:

[0021]

[0022]

[0023] Among them, R∈{0,1} represents the interaction between users and items; d is the degree, which represents the total number of items interacting with the user, or the total number of users interacting with the items.

[0024] According to the method for recommending points of interest of the present invention, inputting the user item interaction data into a point of interest recommendation model to obtain the point of interest item data recommended for the target user output by the point of interest recommendation model specifically includes:

[0025] Inputting the user item interaction data into the interaction relationship analysis layer of the POI recommendation model to obtain all item information interactively associated with the target user and all user information interactively associated with the item information searched by the target user;

[0026] Inputting all the item information and all the user information into the similarity analysis layer of the POI recommendation model to obtain a similarity matrix between items in all the item information and between users in all the user information;

[0027] All the item information and all the user information, as well as the similarity matrix between items in all the item information and between users in all the user information, are input into the result output layer of the point of interest recommendation model, and the result output layer constructs a recommendation interaction matrix based on all the item information and all the user information, as well as the similarity matrix between items in all the item information and between users in all the user information, to obtain the point of interest item data recommended for the target user.

[0028] According to the method for recommending points of interest of the present invention, the step of inputting the user item interaction data into the interaction relationship analysis layer of the point of interest recommendation model to obtain all item information interactively associated with the target user and all user information interactively associated with the item information searched by the target user specifically includes:

[0029] Inputting the user-item interaction data into the first-order interaction relationship analysis layer of the interaction relationship analysis layer, and processing the user-item interaction data by the first-order interaction relationship layer based on the graph attention mechanism to obtain all item information directly interactively associated with the target user and all user information directly interactively associated with the item information searched by the target user;

[0030] Inputting all the item information with direct interaction association and all the user information with direct interaction association into the second-order interaction relationship analysis layer of the interaction relationship analysis layer, and processing all the item information with direct interaction association and all the user information with direct interaction association based on the cross-attention mechanism by the second-order interaction relationship analysis layer to obtain all the item information with potential interaction association with the target user and all the user information with potential interaction association with the item information searched by the target user;

[0031] All the item information with direct interaction associations, all the user information with direct interaction associations, all the item information with potential interaction associations, and all the user information with potential interaction associations are input into the relationship construction layer of the interaction relationship analysis layer to obtain all the item information with direct interaction associations and all the item information with potential interaction associations that are constructed by splicing together all the item information with direct interaction associations and all the item information with potential interaction associations, and all the user information with interaction associations with the item information searched by the target user that is constructed by splicing together all the user information with direct interaction associations and all the user information with potential interaction associations.

[0032] According to the method for recommending points of interest of the present invention, the recommendation interaction matrix constructed by the result output layer according to all the item information and all the user information, and the similarity matrix between items in all the item information and between users in all the user information, obtains the item data of points of interest recommended for the target user, specifically including:

[0033] The result output layer obtains the POI item data recommended for the target user based on a preset formula according to an original interaction matrix constructed by the user item interaction data and the recommendation interaction matrix, wherein the original interaction matrix is ​​constructed according to all item information and all user information;

[0034] The preset formula is:

[0035]

[0036] Wherein, R* is the recommended interaction matrix, and R is the original interaction matrix; and are the matching differences between the similarity matrix between users and between items and the recommendation interaction matrix, respectively. λ1, λ2, and λ3 are preset hyperparameters representing weights.

[0037] According to the method for recommending points of interest of the present invention, the matching degree differences between the similarity matrix between users and between items and the recommendation interaction matrix are:

[0038]

[0039]

[0040] Among them, T is the transpose of the matrix; : represents all values ​​of the rows or columns of the matrix; L is the Laplace regularization of the similarity matrix; tr represents the matrix trace operation.

[0041] The present invention also provides a point of interest recommendation system, comprising:

[0042] A data acquisition module, used to acquire user information of a target user and item information searched by the target user as user item interaction data;

[0043] A data processing module, used for inputting the user entry interaction data into an interest point recommendation model, and obtaining interest point entry data recommended for the target user output by the interest point recommendation model;

[0044] Among them, the point of interest recommendation model is trained based on the similarities between users and items obtained from user-item interaction data samples using the attention mechanism, and the similarities between users and items obtained from the user-item interaction data samples according to preset rules.

[0045] The present invention provides a method and system for recommending points of interest. By inputting the acquired user information of a target user and the item information searched by the target user as user-item interaction data into a point of interest recommendation model, a point of interest recommendation model is obtained by training the similarities between users and users, and between items, obtained from user-item interaction data samples using an attention mechanism, and the similarities between users and users, and between items, obtained from the user-item interaction data samples according to preset rules. Thus, the point of interest item data recommended for the target user can be obtained, thereby achieving the purpose of making more accurate point of interest recommendations for the target user by using the point of interest recommendation model that fully considers the fine-grained personalized information between users and users, and between items during the training process. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 It is a flowchart of a method for recommending points of interest provided by the present invention;

[0048] Figure 2 It is a structural schematic diagram of a point of interest recommendation system provided by the present invention;

[0049] Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] Combine the following Figure 1 A method for recommending points of interest of the present invention is described, and the method is executed by a computer or a combination of software and / or hardware therein, such as Figure 1 As shown, the method comprises the following steps:

[0052] 101. Obtain user information of a target user and item information searched by the target user as user-item interaction data.

[0053] Among them, each computer on the Internet has its own unique IP address. When users use search engines to search for web pages, they generally register their personal information. By crawling data on the Internet, the user information of the target user can be obtained through IP addresses, registration information, etc., and the item information searched on the Internet can be obtained.

[0054] 102. Input the user item interaction data into an interest point recommendation model to obtain interest point item data recommended for the target user output by the interest point recommendation model.

[0055] As location-based services become increasingly popular, POI recommendations can help users search their surroundings and improve their quality of life. However, the data information of users in location social networks is highly sparse, so this problem brings great difficulties to the accurate recommendation of users' POIs. However, there are certain similarities between users and between items. When the similarities between users and between items are counted, the amount of item information data used to analyze users and user searches is greatly increased, effectively overcoming the problem of data sparsity.

[0056] At the same time, the most accurate way to recommend points of interest items to target users based on user-item interaction data is manual recommendation. However, it is obviously impossible to manually obtain user data on the Internet and then recommend points of interest items based on the user's search content on the Internet due to the huge amount of users and search content data involved. Therefore, based on the learning-based similarity measurement, the similarities between users and users, and between items and items obtained according to manually set rules are used to train the similarities between users and users, and between items and items obtained through learning. This can make the learning results conform to the manually set rules and realize the distinction of fine-grained personalized information between users and users, and between items and items.

[0057] Specifically, by inputting user-item interaction data into the similarities between users and items obtained from the user-item interaction data samples according to preset rules, and training the interest point recommendation model obtained by using the attention mechanism to obtain the similarities between users and items from the user-item interaction data samples, it is possible to obtain interest point items that are accurately recommended for target users.

[0058] As an embodiment of the present invention, the loss function of the interest point recommendation model is determined based on the degree of association between a user and an item associated with the user obtained from the user-item interaction data sample using a link prediction mechanism, and the similarity between users and users, and between items obtained from the user-item interaction data sample using an attention mechanism, and the degree of difference between the similarity between users and users, and between items obtained from the user-item interaction data sample according to preset rules.

[0059] Specifically, link prediction is one of the applications of knowledge graph embedding, which maps the contents of entities and relationships in the knowledge graph into a continuous vector space and predicts entities or relationships in the knowledge graph. Therefore, by utilizing the link prediction mechanism, it is possible to optimize the degree of association between users and items associated with the users in the user-item interaction data sample, that is, the degree of interaction closeness between users and items, and the degree of interaction remoteness between users and unassociated items, so that the degree of interaction closeness for each group of users and items in the learned point of interest recommendation model is guaranteed.

[0060] The attention mechanism in deep learning is essentially similar to the selective visual attention mechanism of humans. Its core goal is to select the information that is more critical to the current task goal from a large amount of information. Therefore, by using the attention mechanism, the relationship between users and associated items, as well as items and associated users, can be obtained from the user-item interaction data sample, that is, the aggregate representation for each user and the aggregate representation for each item. Then, based on the aggregate representation for each user and the aggregate representation for each item, the similarity between users and users, and between items can be obtained. When the difference between the similarity between users and users, and between items and items obtained through learning and the similarity between users and users, and between items and items obtained from the user-item interaction data sample according to preset rules becomes smaller and smaller, the similarity between each user and each item in the learned point of interest recommendation model becomes closer to the artificially defined similarity.

[0061] That is, by making the loss function based on the degree of association between the user and the items associated with the user obtained from the user-item interaction data sample using the link prediction mechanism, and the similarity between users and users, and between items obtained from the user-item interaction data sample using the attention mechanism, and the degree of difference between the similarity between users and users, and between items obtained from the user-item interaction data sample according to preset rules, the training of the interest point recommendation model comprehensively considers the two optimization objectives of connection prediction and similarity matching, namely, the learning of local interaction information and global context information.

[0062] As an embodiment of the present invention, the degree of association between the user and the item associated with the user includes:

[0063] The distance difference between the user and the entries associated with the user.

[0064] Specifically, the type of the distance difference is not specifically limited. Taking the calculation of the distance difference of the L2 distance between the user and the item associated with the user as an example, the calculation is performed by the following formula 1:

[0065] d(u,v)=‖h u -h v ‖twenty one)

[0066] Among them, ‖·‖2 represents the L2 distance, h u and h v Represents the aggregate representation of users and items respectively.

[0067] As an embodiment of the present invention, the loss function of the interest point recommendation model is:

[0068]

[0069] in, is a function of the degree of association; is a function of the degree of difference; α is a preset hyperparameter representing the weight;

[0070] The function of the degree of association is:

[0071]

[0072] The function of the degree of difference is:

[0073]

[0074] Wherein, u and v represent users and items respectively; (u, v) and (u′, v′) are real user-item interaction pairs and fake user-item interaction pairs of the user-item interaction data sample respectively; and are the set of real user-item interaction pairs and the set of fake user-item interaction pairs respectively; γ is a preset hyperparameter representing the calculation boundary; and are the sum of the similarity differences between the interactions between user i and user j and all items, and between item i and item j and all users, respectively, obtained according to the preset rules; m and n are the total number of users and the total number of items in the user-item interaction data sample, respectively; and They are the similarities between user i and user j, and item i and item j obtained using the attention mechanism.

[0075] Specifically, the similarity between users and between items obtained by the attention mechanism is calculated by the following formula:

[0076]

[0077]

[0078] in, and represent the aggregate representation of user i and item i respectively; and represent the aggregate representation of user j and item j respectively.

[0079] More specifically, in Formula 4, log represents logarithmic calculation, which is the calculation process of KL divergence (an asymmetric measure of the distance between distributions). Here, log(a) / log(b) is used to replace log(a / b), ensuring that the results calculated for each user and entry are positive.

[0080] As an embodiment of the present invention, the sums of the similarity differences between the interactions between user i and user j and all items, and between item i and item j and all users obtained according to the preset rules are respectively:

[0081]

[0082]

[0083] Among them, R∈{0,1} represents the interaction between users and items; d is the degree, which represents the total number of items interacting with the user, or the total number of users interacting with the items.

[0084] Specifically, R ik ∈{0,1}, represents the interaction between user i and item k, when R ik When R is 0, it means that user i and item k have no interaction. ik When it is 1, it means that there is interaction between user i and item k; d represents degree, which is the number of all edges connected to an entity; R represents the original user-item interaction matrix, which is 1 if there is interaction between the user and the item, otherwise it is 0; k is the index index. When calculating user similarity, for each group of users i and j, it is necessary to traverse all items k to calculate, that is, to obtain the sum of the interaction differences between user i and all items k, and between user j and all items k.

[0085] Similarly, Formula 8 can be used to obtain the sum of the interaction differences between item i and the user k, and between item j and all users k.

[0086] As an embodiment of the present invention, inputting the user item interaction data into an interest point recommendation model to obtain interest point item data output by the interest point recommendation model and recommended for the target user specifically includes:

[0087] Inputting the user item interaction data into the interaction relationship analysis layer of the POI recommendation model to obtain all item information interactively associated with the target user and all user information interactively associated with the item information searched by the target user;

[0088] Inputting all the item information and all the user information into the similarity analysis layer of the POI recommendation model to obtain a similarity matrix between items in all the item information and between users in all the user information;

[0089] All the item information and all the user information, as well as the similarity matrix between items in all the item information and between users in all the user information, are input into the result output layer of the point of interest recommendation model, and the result output layer constructs a recommendation interaction matrix based on all the item information and all the user information, as well as the similarity matrix between items in all the item information and between users in all the user information, to obtain the point of interest item data recommended for the target user.

[0090] Specifically, in the interest point recommendation model, through the interactive relationship analysis layer, the item information and user information that are interactively associated with the target user and item information contained in the user item interaction data are first obtained from the sample user item interaction data when training the interest point recommendation model, that is, the original user item interaction data. Then, based on the obtained item information and user information, through the similarity analysis layer, the similarity matrix between items and between users is obtained. Finally, based on the similarity matrix, the recommendation interaction matrix is ​​constructed through the result output layer, and the interest point item data recommended for the target user is obtained through the recommendation interaction matrix.

[0091] As an embodiment of the present invention, the inputting of the user item interaction data into the interaction relationship analysis layer of the POI recommendation model to obtain all item information interactively associated with the target user and all user information interactively associated with the item information searched by the target user specifically includes:

[0092] Inputting the user-item interaction data into the first-order interaction relationship analysis layer of the interaction relationship analysis layer, and processing the user-item interaction data by the first-order interaction relationship layer based on the graph attention mechanism to obtain all item information directly interactively associated with the target user and all user information directly interactively associated with the item information searched by the target user;

[0093] Inputting all the item information with direct interaction association and all the user information with direct interaction association into the second-order interaction relationship analysis layer of the interaction relationship analysis layer, and processing all the item information with direct interaction association and all the user information with direct interaction association based on the cross-attention mechanism by the second-order interaction relationship analysis layer to obtain all the item information with potential interaction association with the target user and all the user information with potential interaction association with the item information searched by the target user;

[0094] All the item information with direct interaction associations, all the user information with direct interaction associations, all the item information with potential interaction associations, and all the user information with potential interaction associations are input into the relationship construction layer of the interaction relationship analysis layer to obtain all the item information with direct interaction associations and all the item information with potential interaction associations that are constructed by splicing together all the item information with direct interaction associations and all the item information with potential interaction associations, and all the user information with interaction associations with the item information searched by the target user that is constructed by splicing together all the user information with direct interaction associations and all the user information with potential interaction associations.

[0095] Specifically, the user-item interaction data is input into the POI recommendation model in the form of a bipartite graph of target user-item interaction. The first-order interaction relationship analysis layer of the interaction relationship analysis layer processes the user-item interaction data based on the graph attention mechanism, that is, firstly, the edge between the target user and the item is established through Formula 9:

[0096]

[0097] in is the initial embedding, W u , W v is the transformation matrix of the target user and item, and the initial embedding refers to the representation of the target user and item directly obtained through the mapping matrix.

[0098] Then, the attention weight of the message passing through each target user-item edge is calculated by Equation 10-13, and the calculation is performed through the target user and item embedding representation corresponding to the edge, and all edges connected to each user and each item are normalized:

[0099]

[0100]

[0101]

[0102]

[0103] The arrow indicates the direction of data transmission; W is the change matrix; N is other items or users that directly interact with the target user or item; c uv is the unnormalized attention weight transferred from the target user u to the item v data; α uv is the normalized attention weight transferred from the target user u to the item v data, that is, the attention weight of each edge is calculated by the target user and item connected to the edge, and is normalized so that the total attention of the target user to all associated items is 1, and the total attention of the item to all associated target users is 1, thereby obtaining the representation of all item information that has a direct interaction association with the target user, and the representation of all user information that has a direct interaction association with the item information searched by the target user, as shown in Equations 14 and 15, respectively:

[0104]

[0105]

[0106] in, and are the first-order representations of the target user and item respectively; σ is a nonlinear activation function.

[0107] Furthermore, the second-order interaction relationship analysis layer processes all the item information with direct interaction associations and all the user information with direct interaction associations based on the cross-attention mechanism, specifically including: calculating the second-order representations of the target user and the item through the first-order representations of the target user and the item, that is, the representation of all item information with potential interaction associations with the target user, and the representation of all user information with potential interaction associations with the item information searched by the target user.

[0108] First, based on natural language processing, the cross attention of the first-order representation is calculated by equations 16-19:

[0109]

[0110]

[0111]

[0112]

[0113] in, represents the projection matrix; Q (query) refers to the query; K (key) refers to the keyword; V (value) refers to the value; Q, K, V are the first-order aggregate representations of the target user and the item converted by the mapping matrix. In the calculation process of cross-attention, Q should represent different types of entities with K and V, such as: Q represents the target user, then K and V represent the item; Q represents the item, then K and V represent the target user. Q and K are matrix-producted to get the cross-attention value, and then the same as V, the second-order aggregate representation after the cross-attention weighting can be calculated; A represents the adjacency matrix of the target user-item; T represents the matrix transpose.

[0114] Then, based on the cross attention of the first-order representation, the second-order representation of the target user and item is obtained, as shown in Equations 20 and 21:

[0115]

[0116]

[0117] in, and are the second-order representations of the target user and item respectively; : represents all values ​​of the rows or columns of the matrix.

[0118] Furthermore, the relationship construction layer of the interactive relationship analysis layer obtains all item information that is interactively associated with the target user, which is composed of the first-order representation and the second-order representation of the target user, and all user information that is interactively associated with the item information searched by the target user, which is composed of the first-order representation and the second-order representation of the item.

[0119] In summary, by adopting graph attention message passing and cross-attention information aggregation to explicitly learn the similarity metric, the interaction correlation between the target users and items is more comprehensive.

[0120] It can be understood that based on this function of the interactive relationship analysis layer of the POI recommendation model, during the model training process, when calculating the distance between a user and an item associated with the user, the user and item used are also concatenated representations of the first-order representation and the second-order representation, that is, Formula 1 is transformed into:

[0121]

[0122] As an embodiment of the present invention, the recommendation interaction matrix constructed by the result output layer according to all the item information and all the user information, and the similarity matrix between items in all the item information and between users in all the user information, obtains the POI item data recommended for the target user, specifically including:

[0123] The result output layer obtains the POI item data recommended for the target user based on a preset formula according to an original interaction matrix constructed by the user item interaction data and the recommendation interaction matrix, wherein the original interaction matrix is ​​constructed according to all item information and all user information;

[0124] The preset formula is:

[0125]

[0126] Wherein, R* is the recommended interaction matrix, and R is the original interaction matrix; and are the matching differences between the similarity matrix between users and between items and the recommendation interaction matrix, respectively. λ1, λ2, and λ3 are preset hyperparameters representing weights.

[0127] Specifically, according to the recommended interaction matrix, when using the interest point recommendation model to provide interest point recommendations for the target user, the interest point recommendation model can optimize the constructed original interaction matrix through the recommended interaction matrix by using linear extrapolation, that is, the original interaction matrix is ​​more in line with the recommended interaction matrix. The original interaction matrix is ​​constructed by obtaining all item information that has an interactive association with the target user through the user item interaction data input into the interest point recommendation model, as well as all user information that has an interactive association with the item information searched by the target user. Therefore, the interest points recommended for the target user can be obtained through the interest point recommendation model according to the input user item interaction data.

[0128] As an embodiment of the present invention, the matching degree differences between the similarity matrix between users and between items and the recommendation interaction matrix are:

[0129]

[0130]

[0131] Among them, T is the transpose of the matrix; : represents all values ​​of the rows or columns of the matrix; L is the Laplace regularization of the similarity matrix; tr represents the matrix trace operation.

[0132] Specifically, the formula for similarity matching calculation is as follows:

[0133] L=ID -1 / 2 SD -1 / 2 (26)

[0134] Among them, L is the Laplace regularization of the learned similarity matrix, I represents the identity matrix, D represents the degree matrix, and S represents the learned similarity matrix or

[0135] Substituting Formula 26 into Formulas 24 and 25, we can obtain the matching degree differences between the similarity matrix between users and between items and the recommendation interaction matrix.

[0136] The point of interest recommendation method described in the present invention adopts graph attention message passing and cross-attention information aggregation to explicitly learn similarity metrics; by using a combined objective function, namely local interaction prediction and global similarity matching, the learned similarity metric reflects the fine-grained personalized information of each user and item, which is ignored in previous heuristic methods. Compared with other graph representation learning recommendation methods, the method described in the present invention can better utilize the characteristics of learned similarity, thereby obtaining more robust performance, which is of great significance to the development of the recommendation system field.

[0137] Combine the following Figure 2 A point of interest recommendation system provided by the present invention is described. The point of interest recommendation system described below and the point of interest recommendation method described above can correspond to each other.

[0138] The present invention provides a point of interest recommendation system, comprising a data acquisition module 210 and a data processing module 220; wherein:

[0139] The data acquisition module 210 is used to acquire user information of a target user and item information searched by the target user as user item interaction data;

[0140] The data processing module 220 is used to input the user entry interaction data into the interest point recommendation model to obtain the interest point entry data recommended for the target user output by the interest point recommendation model;

[0141] Among them, the point of interest recommendation model is trained based on the similarities between users and items obtained from user-item interaction data samples using the attention mechanism, and the similarities between users and items obtained from the user-item interaction data samples according to preset rules.

[0142] Specifically, the interest point recommendation system described in the present invention can obtain the interest point item data recommended for the target user by inputting the acquired user information of the target user and the item information searched by the target user as user-item interaction data into the interest point recommendation model, and using the interest point recommendation model trained based on the similarities between users and users, and between items, obtained from the user-item interaction data samples using the attention mechanism, and the similarities between users and users, and between items, obtained from the user-item interaction data samples according to preset rules, thereby achieving the purpose of making more accurate interest point recommendations for the target user by using the interest point recommendation model that fully considers the fine-grained personalized information between users and users, and between items during the training process.

[0143] The interest point recommendation system provided in the embodiment of the present invention is used to sign an interest point recommendation method of each embodiment. The specific methods and processes of each module included in the interest point recommendation system to implement corresponding functions are detailed in the embodiment of the interest point recommendation method above, and will not be repeated here.

[0144] The interest point recommendation system of the present invention is used for an interest point recommendation method in the above-mentioned embodiments. Therefore, the description and definition in an interest point recommendation method in the above-mentioned embodiments can be used for understanding each execution module in the embodiments of the present invention.

[0145] Figure 3 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 3 As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330 and a communication bus 340, wherein the processor 310, the communication interface 320 and the memory 330 communicate with each other through the communication bus 340. The processor 310 may call the logic instructions in the memory 330 to execute the interest point recommendation method, which includes: obtaining user information of a target user and item information searched by the target user as user item interaction data; inputting the user item interaction data into an interest point recommendation model to obtain the interest point item data recommended for the target user output by the interest point recommendation model; wherein the interest point recommendation model is trained based on the similarity between users and users and items obtained from the user item interaction data samples using the attention mechanism, and the similarity between users and users and items obtained from the user item interaction data samples according to preset rules.

[0146] In addition, the logic instructions in the above-mentioned memory 330 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0147] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the interest point recommendation method provided by the above-mentioned methods, which method includes: obtaining user information of a target user, and item information searched by the target user as user-item interaction data; inputting the user-item interaction data into an interest point recommendation model, and obtaining interest point item data recommended for the target user output by the interest point recommendation model; wherein the interest point recommendation model is trained based on the similarities between users and items obtained from user-item interaction data samples using an attention mechanism, and the similarities between users and items obtained from the user-item interaction data samples according to preset rules.

[0148] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the interest point recommendation method provided by the above-mentioned methods, the method comprising: obtaining user information of a target user, and item information searched by the target user as user-item interaction data; inputting the user-item interaction data into an interest point recommendation model, and obtaining interest point item data recommended for the target user output by the interest point recommendation model; wherein the interest point recommendation model is trained based on the similarities between users and users, and between items, obtained from user-item interaction data samples using an attention mechanism, and the similarities between users and users, and between items, obtained from the user-item interaction data samples according to preset rules.

[0149] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0150] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for recommending points of interest, characterized in that: include: Acquire user information of a target user and information of items searched by the target user as user item interaction data; Inputting the user entry interaction data into an interest point recommendation model to obtain interest point entry data recommended for the target user output by the interest point recommendation model; The POI recommendation model is trained based on the similarities between users and items obtained from the user-item interaction data samples using the attention mechanism, and the similarities between users and items obtained from the user-item interaction data samples according to preset rules; The loss function of the POI recommendation model is determined based on the degree of association between a user and an item associated with the user obtained from the user-item interaction data sample using a link prediction mechanism, and the degree of difference between the similarity between users and users, and between items and items obtained from the user-item interaction data sample using an attention mechanism, and the degree of difference between the similarity between users and users, and between items and items obtained from the user-item interaction data sample according to a preset rule; The loss function of the POI recommendation model is: in is a function of the degree of association; is a function of the degree of difference; α is a preset hyperparameter representing the weight; The function of the degree of association is: The function of the degree of difference is: Where u and v represent users and entries respectively; (u,v) and (u ′ ,v ′ ) are respectively real user-item interaction pairs and false user-item interaction pairs of the user-item interaction data sample; and are the set of real user-item interaction pairs and the set of fake user-item interaction pairs respectively; γ is a preset hyperparameter representing the calculation boundary; and are the sum of the similarity differences between the interactions between user i and user j and all items, and between item i and item j and all users, respectively, obtained according to the preset rules; m and n are the total number of users and the total number of items in the user-item interaction data sample, respectively; and are the similarities between user i and user j, and item i and item j obtained using the attention mechanism; The sum of the similarity differences between the interactions between user i and user j and all items, and between item i and item j and all users obtained according to the preset rules are: Among them, R∈{0,1} represents the interaction between users and items; d is the degree, which represents the total number of items interacting with the user, or the total number of users interacting with the items.

2. The method for recommending points of interest according to claim 1, characterized in that: The degree of association between the user and the items associated with the user includes: The distance difference between the user and the entries associated with the user.

3. The method for recommending points of interest according to any one of claims 1 to 2, characterized in that: The step of inputting the user item interaction data into an interest point recommendation model to obtain interest point item data output by the interest point recommendation model and recommended for the target user specifically includes: Inputting the user item interaction data into the interaction relationship analysis layer of the POI recommendation model to obtain all item information interactively associated with the target user and all user information interactively associated with the item information searched by the target user; Inputting all the item information and all the user information into the similarity analysis layer of the POI recommendation model to obtain a similarity matrix between items in all the item information and between users in all the user information; All the item information and all the user information, as well as the similarity matrix between items in all the item information and between users in all the user information, are input into the result output layer of the point of interest recommendation model, and the result output layer constructs a recommendation interaction matrix based on all the item information and all the user information, as well as the similarity matrix between items in all the item information and between users in all the user information, to obtain the point of interest item data recommended for the target user.

4. The method for recommending points of interest according to claim 3, characterized in that: The step of inputting the user item interaction data into the interaction relationship analysis layer of the POI recommendation model to obtain all item information interactively associated with the target user and all user information interactively associated with the item information searched by the target user specifically includes: Inputting the user-item interaction data into the first-order interaction relationship analysis layer of the interaction relationship analysis layer, and processing the user-item interaction data by the first-order interaction relationship layer based on the graph attention mechanism to obtain all item information directly interactively associated with the target user and all user information directly interactively associated with the item information searched by the target user; Inputting all the item information with direct interaction association and all the user information with direct interaction association into the second-order interaction relationship analysis layer of the interaction relationship analysis layer, and processing all the item information with direct interaction association and all the user information with direct interaction association based on the cross-attention mechanism by the second-order interaction relationship analysis layer to obtain all the item information with potential interaction association with the target user and all the user information with potential interaction association with the item information searched by the target user; All the item information with direct interaction associations, all the user information with direct interaction associations, all the item information with potential interaction associations, and all the user information with potential interaction associations are input into the relationship construction layer of the interaction relationship analysis layer to obtain all the item information with direct interaction associations and all the item information with potential interaction associations that are constructed by splicing together all the item information with direct interaction associations and all the item information with potential interaction associations, and all the user information with interaction associations with the item information searched by the target user that is constructed by splicing together all the user information with direct interaction associations and all the user information with potential interaction associations.

5. The method for recommending points of interest according to claim 4, characterized in that: The recommendation interaction matrix constructed by the result output layer according to all the item information and all the user information, and the similarity matrix between items in all the item information and between users in all the user information, obtains the POI item data recommended for the target user, specifically including: The result output layer obtains the POI item data recommended for the target user based on a preset formula according to the original interaction matrix constructed by the user item interaction data and the recommendation interaction matrix; The preset formula is: Wherein, R* is the recommended interaction matrix, and R is the original interaction matrix; and are the matching differences between the similarity matrix between users and between items and the recommendation interaction matrix respectively, and λ1, λ2, and λ3 are the preset hyperparameters representing the weights.

6. The method for recommending points of interest according to claim 5, characterized in that: The matching differences between the similarity matrix between users and between items and the recommendation interaction matrix are: Among them, T is the transpose of the matrix; : represents all values ​​of the rows or columns of the matrix; L is the Laplace regularization of the similarity matrix; tr represents the matrix trace operation.

7. A point of interest recommendation system, characterized in that: include: A data acquisition module, used to acquire user information of a target user and item information searched by the target user as user item interaction data; A data processing module, used for inputting the user entry interaction data into an interest point recommendation model, and obtaining interest point entry data recommended for the target user output by the interest point recommendation model; The POI recommendation model is trained based on the similarities between users and items obtained from the user-item interaction data samples using the attention mechanism, and the similarities between users and items obtained from the user-item interaction data samples according to preset rules; The loss function of the POI recommendation model is determined based on the degree of association between a user and an item associated with the user obtained from the user-item interaction data sample using a link prediction mechanism, and the degree of difference between the similarity between users and users, and between items and items obtained from the user-item interaction data sample using an attention mechanism, and the degree of difference between the similarity between users and users, and between items and items obtained from the user-item interaction data sample according to a preset rule; The loss function of the POI recommendation model is: in is a function of the degree of association; is a function of the degree of difference; α is a preset hyperparameter representing the weight; The function of the degree of association is: The function of the degree of difference is: Where u and v represent users and entries respectively; (u,v) and (u ′ ,v ′ ) are respectively real user-item interaction pairs and false user-item interaction pairs of the user-item interaction data sample; and are the set of real user-item interaction pairs and the set of fake user-item interaction pairs respectively; γ is a preset hyperparameter representing the calculation boundary; and are the sum of the similarity differences between the interactions between user i and user j and all items, and between item i and item j and all users, respectively, obtained according to the preset rules; m and n are the total number of users and the total number of items in the user-item interaction data sample, respectively; and are the similarities between user i and user j, and item i and item j obtained using the attention mechanism; The sum of the similarity differences between the interactions between user i and user j and all items, and between item i and item j and all users obtained according to the preset rules are: Among them, R∈{0,1} represents the interaction between users and items; d is the degree, which represents the total number of items interacting with the user, or the total number of users interacting with the items.

Images