Tourist attraction binding recommendation method based on strategy perception

Abstract

The invention provides a tourist attraction binding recommendation method and system based on strategy perception, a storage medium and electronic equipment, and relates to the technical field of recommendation systems. The method comprises the following steps: firstly, acquiring set information of users and scenic spots, a user-scenic spot interaction matrix and a predefined scenic spot binding set; secondly, obtaining feature representations of the user and the scenic spot; thirdly, an HGT model with a self-attention mechanism is introduced, user preferences and scenic spot features are deeply fused, and strategy perception binding representation and strategy perception user representation are obtained respectively; and finally, generating binding recommendation based on policy awareness. According to the method, when a tourist attraction recommendation task is processed, the method is not limited by a predefined scenic spot binding set any more, more personalized and strategic recommendation is generated by comprehensively utilizing the multi-dimensional preference information of the target user and the attribute characteristics of the scenic spots, the accuracy of a binding recommendation result is ensured, and thus the user experience and satisfaction are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of recommendation systems, and in particular to a tourist attraction bundling recommendation method, system, storage medium and electronic device based on strategy perception. Background Art

[0002] With the development of the tourism industry, tourism recommendation systems have gradually become an important tool to help users choose suitable attractions and optimize their travel experience.

[0003] In the related technologies, tourism recommendation systems mostly use traditional content-based or collaborative filtering recommendation methods, but these methods still have certain limitations when facing complex and changing user needs. Among them, many existing bundled recommendation systems use predefined bundle combinations, which are usually determined based on the general trend of big data or operation strategies, and cannot meet the highly personalized needs of users. For example, some systems will recommend fixed classic travel routes, but these routes may not necessarily meet the interests and needs of each user. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the deficiencies in the prior art, the present invention provides a tourist attraction bundle recommendation method, system, storage medium and electronic device based on strategy perception, which solves the technical problem that the predefined attraction bundle combination cannot meet the highly personalized needs of users.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0008] A strategy-aware tourist attraction bundling recommendation method, comprising:

[0009] Obtain a user set and a user information set of each user, a scenic spot set and a scenic spot information set of each scenic spot, a user-scenic spot interaction matrix recording the target user's visit history, and a predefined scenic spot bundle set to construct a bundle-scenic spot interaction graph; wherein any scenic spot bundle is a set consisting of several scenic spots in the scenic spot set;

[0010] Based on the user information set of the target user and the scenic spot information set of each scenic spot, respectively obtaining initial feature embedding vectors of the target user and each scenic spot;

[0011] Based on the initial feature embedding vectors of the target user and each scenic spot, combined with the user-scenic spot interaction matrix, respectively obtain the user token representation and the scenic spot token representation, and obtain the bundle representation of each scenic spot bundle;

[0012] Based on the bundle-attraction interaction graph, the user token representation and the attraction token representation, an HGT model with a self-attention mechanism is introduced to obtain a strategy-aware bundle representation and a strategy-aware user representation respectively;

[0013] Based on the strategy-aware bundle representation and the strategy-aware user representation, obtaining an interaction probability between the target user and each scenic spot bundle to obtain a preference value of the target user for each scenic spot in the corresponding scenic spot bundle;

[0014] The target user's preference value for each attraction in all attraction bundles is counted, and several attractions with the highest ranking are selected after sorting. The attractions are combined to generate an attraction set as a bundle for recommendation to the target user.

[0015] Preferably, the process of constructing the bundle-attraction interaction diagram is as follows:

[0016] The attractions and attraction bundles are regarded as two different types of nodes, and the subordinate relationship between the two is regarded as an edge between the attraction node and the bundle node, so as to construct the bundle-attraction interaction graph.

[0017] Preferably, an adaptive graph convolutional network is used to obtain user token representation and scenic spot token representation respectively, which can be expressed as:

[0018]

[0019] in, Represent the target user u m User token representation, attractions n Token representation of scenic spots; AGCN is an adaptive graph convolutional network; x m ,y n Represent the target user u m and attractions n The initial feature embedding vector of

[0020] Based on the attraction token representations of all attractions in each attraction bundle, the corresponding bundle representation is obtained; it is expressed as:

[0021]

[0022] in, Indicates attraction bundle b k The bundling representation, Indicates attraction bundle b k No. o k Attractions Token representation, O k Bundle for attractions k The number of attractions.

[0023] Preferably, the process of acquiring the policy-aware bundling representation includes:

[0024] The HGT model with L layers stacked is introduced. For the lth layer of the HGT model, the attraction bundle b is calculated based on the bundle-attraction interaction graph G. k No. o k Attractions Positional encoding in graphs It is expressed as:

[0025]

[0026] in, represents the scenic spot bundle b learned from the previous layer k The latent vector of and They represent the attractions on the bundle-attraction interaction graph G. Node and Attraction Bundle b k The node's neighbor set;

[0027] Bundle attractions k Position encoding and bundle representation of all scenic spots in Combined to obtain the enhanced bundle representation of the lth layer It is expressed as:

[0028]

[0029] in, Represents matrix addition operation;

[0030] Simplify the self-attention mechanism, retain only the core structure, and calculate the scenic spot bundle b k Self-attention matrix between inliers It is expressed as:

[0031]

[0032] in, and is a trainable matrix, representing the Token representation of scenic spots The key and value mapping of ; d represents the embedding dimension; the superscript T represents transposition;

[0033] Combined with the self-attention matrix, obtain the scenic spot bundle b in the lth layer k Characterization of bundling strategies It is expressed as:

[0034]

[0035] Among them, softmax(·) represents the softmax function;

[0036] Perform an average pooling operation on the representation containing the bundling strategy to obtain the bundle b of each scenic spot k Strategy-aware bundling representation It is expressed as:

[0037]

[0038] Among them, mean_pooling is the average pooling operation.

[0039] Preferably, the process of acquiring the policy-aware user representation includes:

[0040] In the lth layer, for the target user u m , using the representation containing the bundling strategy to represent the user's historical bundling representation, recorded as:

[0041]

[0042] in, The target user u m The rth bundle that has been interacted with, R is the total number of bundles that the target user has interacted with;

[0043] Calculate the target user u m The correlation matrix between its user history bundle representation It is expressed as:

[0044]

[0045] in, and Respectively represent user Token representation The key and value mapping matrix of ;

[0046] Correlation Matrix After extracting features from the user's historical bundle representation, combine it with the user's Token representation Get the target user u m Policy-aware user representation It is expressed as:

[0047]

[0048] Among them, Attention(·) represents the user-bundle association matrix and user history bundle representation Perform interaction modeling; Aggregate(·) represents the feature aggregation function.

[0049] Preferably, based on the policy-aware bundling representation and the policy-aware user representation, the target user u is obtained by an inner product operation. m Bundled with attractionsk The probability of interaction between them is expressed as:

[0050]

[0051] Among them, σ(·) is the sigmoid activation function.

[0052] Preferably, the interaction probability is mapped to the target user u by a uniform distribution method. m Bundle b for attractions k The preference value of each attraction in It is expressed as:

[0053]

[0054] A strategy-aware tourist attraction bundling recommendation system, including:

[0055] An information acquisition module is used to acquire a user set and a user information set of each user, a scenic spot set and a scenic spot information set of each scenic spot, a user-scenic spot interaction matrix recording the target user's visit history, and a predefined scenic spot bundle set, so as to construct a bundle-scenic spot interaction graph; wherein any scenic spot bundle is a set consisting of several scenic spots in the scenic spot set;

[0056] A vector embedding module, used to obtain initial feature embedding vectors of the target user and each scenic spot based on the user information set of the target user and the scenic spot information set of each scenic spot;

[0057] A representation acquisition module, used to acquire a user token representation and a scenic spot token representation based on the initial feature embedding vectors of the target user and each scenic spot, combined with the user-scenic spot interaction matrix, and acquire a bundle representation of each scenic spot bundle;

[0058] A strategy-aware module, for introducing an HGT model with a self-attention mechanism based on the bundle-attraction interaction graph, the user token representation and the attraction token representation, to obtain a strategy-aware bundle representation and a strategy-aware user representation respectively;

[0059] A preference acquisition module, used to acquire the interaction probability between the target user and each scenic spot bundle based on the policy-aware bundle representation and the policy-aware user representation, so as to acquire the target user's preference value for each scenic spot in the corresponding scenic spot bundle;

[0060] The bundle recommendation module is used to count the target user's preference values ​​for each attraction in all attraction bundles, select several attractions with the highest ranking after sorting, and combine them to generate an attraction set as a bundle to recommend to the target user.

[0061] A storage medium stores a computer program for policy-aware tourist attraction bundle recommendation, wherein the computer program enables a computer to execute the tourist attraction bundle recommendation method as described above.

[0062] An electronic device, comprising:

[0063] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include a method for executing the tourist attraction bundling recommendation method as described above.

[0064] (III) Beneficial effects

[0065] The present invention provides a method, system, storage medium and electronic device for recommending tourist attractions bundled based on strategy perception. Compared with the prior art, the present invention has the following beneficial effects:

[0066] In the present invention, first, the collection information of users and attractions, the user-attraction interaction matrix and the predefined attraction bundle set are obtained; secondly, the characteristic representation of users and attractions is obtained; thirdly, the HGT model with self-attention mechanism is introduced to deeply integrate user preferences with attraction features, and obtain strategy-aware bundle representation and strategy-aware user representation respectively; finally, bundle recommendation is generated based on strategy perception. When processing the task of recommending tourist attractions, this method is no longer confined to the predefined attraction bundle set, but comprehensively utilizes the multi-dimensional preference information of target users and the attribute characteristics of attractions to generate more personalized and strategic recommendations, ensure the accuracy of bundle recommendation results, and thus significantly improve user experience and satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only 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.

[0068] Figure 1 A block diagram of a tourist attraction bundling recommendation method based on strategy perception provided by an embodiment of the present invention;

[0069] Figure 2 A flow chart based on a policy-aware model is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. 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.

[0071] The embodiments of the present application provide a method, system, storage medium and electronic device for recommending tourist attractions bundles based on strategy perception, thereby solving the technical problem that predefined tourist attractions bundle combinations cannot meet the highly personalized needs of users. It is intended that when processing the task of recommending tourist attractions bundles in tourism scenarios, user preferences, tourist attractions characteristics and strategy perception factors can be comprehensively considered, thereby generating more targeted and highly satisfactory tourist recommendation plans, and improving the accuracy of recommendation results and user experience.

[0072] The technical solution in the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:

[0073] In the embodiment of the present invention, first, the collection information of users and attractions, the user-attraction interaction matrix and the predefined attraction bundle set are obtained; secondly, the characteristic representation of users and attractions is obtained; thirdly, the HGT model with self-attention mechanism is introduced to deeply integrate user preferences with attraction features, and obtain strategy-aware bundle representation and strategy-aware user representation respectively; finally, bundle recommendation is generated based on strategy perception. When processing the task of recommending tourist attractions, this method is no longer confined to the predefined attraction bundle set, but comprehensively utilizes the multi-dimensional preference information of the target user and the attribute characteristics of the attraction to generate more personalized and strategic recommendations, ensure the accuracy of the bundle recommendation results, and thus significantly improve the user experience and satisfaction.

[0074] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0075] Embodiment 1:

[0076] like Figure 1 As shown, the embodiment of the present invention provides a tourist attraction bundling recommendation method based on strategy perception, including:

[0077] S1, obtaining a user set and a user information set of each user, a scenic spot set and a scenic spot information set of each scenic spot, a user-scenic spot interaction matrix recording the target user's visit history, and a predefined scenic spot bundle set to construct a bundle-scenic spot interaction graph; wherein any scenic spot bundle is a set consisting of several scenic spots in the scenic spot set;

[0078] S2, based on the user information set of the target user and the scenic spot information set of each scenic spot, respectively obtain the initial feature embedding vector of the target user and each scenic spot;

[0079] S3, based on the initial feature embedding vectors of the target user and each scenic spot, combined with the user-scenic spot interaction matrix, respectively obtain the user token representation and the scenic spot token representation, and obtain the bundle representation of each scenic spot bundle;

[0080] S4. Based on the bundle-attraction interaction graph, the user token representation and the attraction token representation, an HGT model with a self-attention mechanism is introduced to obtain a strategy-aware bundle representation and a strategy-aware user representation, respectively;

[0081] S5. Based on the strategy-aware bundle representation and the strategy-aware user representation, obtain the interaction probability between the target user and each scenic spot bundle to obtain the target user's preference value for each scenic spot in the corresponding scenic spot bundle;

[0082] S6. Count the target user's preference value for each scenic spot in all scenic spot bundles, select several scenic spots with the highest ranking after sorting, and combine them to generate a scenic spot set as a bundle to recommend to the target user.

[0083] When processing tourist attraction recommendation tasks, the embodiments of the present invention are no longer limited to predefined attraction bundle sets, but instead comprehensively utilize the multi-dimensional preference information of target users and the attribute characteristics of attractions to generate more personalized and strategic recommendations, ensuring the accuracy of bundled recommendation results, thereby significantly improving user experience and satisfaction.

[0084] Next, we will introduce the steps of the above scheme in detail:

[0085] In step S1, a user set and a user information set of each user, a scenic spot set and a scenic spot information set of each scenic spot, a user-scenic spot interaction matrix recording the target user's visit history, and a predefined scenic spot bundle set are obtained to construct a bundle-scenic spot interaction graph; wherein any scenic spot bundle is a set consisting of several scenic spots in the scenic spot set.

[0086] Specifically, this step obtains the following information:

[0087] (1) User set, denoted as U = {u1,u2,…,u m ,…,u M}, where u m represents the mth user, and M is the total number of users.

[0088] In addition, for the target user u m(Assume that the mth user is the target user), collect their user ID and preference information such as travel style, budget, and attraction reviews, and build a user information set in Represents user u m The value of the i-th information, I is the total number of user information.

[0089] (2) The set of scenic spots, denoted by S = {s1, s2, …, s n ,…,s N}, where s n Represents the nth scenic spot, and N is the total number of scenic spots.

[0090] In addition, for attractions n , collect the scenic spot ID and geographical location, type, rating and other attribute information, and build a scenic spot information collection in Indicates attractions n The value of the jth information, J is the total number of scenic spot information.

[0091] (3) The user-attraction interaction matrix that records the target user’s visit history is denoted as Where X mn =1 indicates the target user u m Attractions visited n , otherwise X mn =0.

[0092] (4) A predefined set of scenic spot bundles, denoted as B = {b1, b2, …, b k ,…,b K}, where b k represents the kth bundle, K is the total number of bundles. k is a set consisting of several scenic spots in the scenic spot set S, denoted as in Indicates attraction bundle b k o k Attractions, O k Bundle for attractions k The number of attractions.

[0093] On this basis, this step constructs the bundle-attraction interaction diagram:

[0094] The attractions and attraction bundles are regarded as two different types of nodes, and the subordinate relationship between them is regarded as an edge between the attraction node and the bundle node, and the bundle-attraction interaction graph G is constructed.

[0095] In step S2, based on the user information set of the target user and the scenic spot information set of each scenic spot, initial feature embedding vectors of the target user and each scenic spot are obtained respectively.

[0096] In this step, the user and scenic spot information are converted into trainable vectors. Using equations (1) and (2), we can obtain the target user u m and attractions n The initial feature embedding vector and

[0097] x m =f embed (P m ) (1)

[0098] y n =f embed (Q n ) (2)

[0099] Among them, f embed (·) represents the embedding mapping function, and d represents the embedding dimension.

[0100] In step S3, based on the initial feature embedding vectors of the target user and each scenic spot, combined with the user-scenic spot interaction matrix, the user Token representation and the scenic spot Token representation are respectively obtained, and the bundle representation of each scenic spot bundle is obtained.

[0101] In this step, the attractions in the bundle are regarded as tokens and embedded into the model, so as to obtain the representation of the attraction bundle for subsequent policy-aware processing, which includes:

[0102] S31, combined with the user-attraction interaction matrix, using the adaptive graph convolutional network to obtain the optimized user u m and attractions n Token representation, expressed as:

[0103]

[0104] in, Represent the target user u m User token representation, attractions n Token representation of scenic spots; AGCN is an adaptive graph convolutional network.

[0105] S32, based on the attraction Token representations of all attractions in each attraction bundle, obtain the corresponding bundle representation; expressed as:

[0106]

[0107] in, Indicates attraction bundle b k The bundling representation, Indicates attraction bundle b kNo. o k Attractions Token representation.

[0108] In step S4, based on the bundle-attraction interaction graph, the user token representation and the attraction token representation, an HGT model with a self-attention mechanism is introduced to obtain a strategy-aware bundle representation and a strategy-aware user representation, respectively.

[0109] The present invention constructs an HGT model with L layers stacked, models bundles and users in a graph structure, captures the complex relationship between them, and obtains the strategic perception representation of bundles and users.

[0110] like Figure 2 As shown, Figure 2 A flow chart based on a policy-aware model provided by an embodiment of the present invention is given.

[0111] Reference Figure 2 ,The process of acquiring the policy-aware bundling representation in this step specifically includes:

[0112] S41. Each scenic spot bundle implies the strategy information of selecting a certain scenic spot. By introducing the statistical relationship between scenic spots in the bundle, a graph-based position encoder is used to reflect the prior knowledge of the predefined bundling strategy. Specifically, for the lth (1≤l≤L) layer of the HGT model, based on the bundle-scenic spot interaction graph G, the scenic spot bundle b is calculated. k No. o k Attractions Positional encoding in graphs It is expressed as:

[0113]

[0114] in, represents the scenic spot bundle b learned from the previous layer k The latent vector of , when l = 1, the vector is obtained by random initialization; and They represent the attractions on the bundle-attraction interaction graph G. Node and Attraction Bundle b k The set of neighbors of a node.

[0115] S42, Bundle the attractions k Position encoding and bundle representation of all scenic spots in Combined to obtain the enhanced bundle representation of the lth layer It is expressed as:

[0116]

[0117] in, Represents a matrix addition operation.

[0118] S43. After obtaining the bundle representation, the self-attention mechanism is used to learn the association between attractions in the bundle for bundling strategy modeling. Considering issues such as sparsity, the self-attention mechanism is simplified by removing the feedforward network (FFN), multi-head attention scoring, and hierarchical normalization, retaining only the core structure, and calculating the attraction bundle b. k Self-attention matrix between inliers It is expressed as:

[0119]

[0120] in, and is a trainable matrix, representing the Token representation of scenic spots A mapping of keys and values ​​of ; the superscript T indicates transpose.

[0121] S44. Combine the self-attention matrix to obtain the scenic spot bundle b in the lth layer k Characterization of bundling strategies It is expressed as:

[0122]

[0123] Among them, softmax(·) represents the softmax function.

[0124] S45, perform an average pooling operation on the representation containing the bundling strategy to obtain the bundle b of each scenic spot k Strategy-aware bundling representation It is expressed as:

[0125]

[0126] Among them, mean_pooling is the average pooling operation.

[0127] Also refer to Figure 2 ,The process of acquiring the policy aware user representation in this step specifically includes:

[0128] S46. In the first layer, for the target user u m , using the representation containing the bundling strategy to represent the user's historical bundling representation, recorded as:

[0129]

[0130] in, The target user u m The rth bundle that has been interacted with, and R is the total number of bundles interacted with by the target user.

[0131] S47. Calculate target user um The correlation matrix between its user history bundle representation It is expressed as:

[0132]

[0133] in, and Respectively represent user Token representation The key and value mapping matrix of ;

[0134] S48, correlation matrix After extracting features from the user's historical bundle representation, combine it with the user's Token representation Get the target user u m Policy-aware user representation It is expressed as:

[0135]

[0136] Among them, Attention(·) represents the user-bundle association matrix and user history bundle representation Perform interaction modeling; Aggregate(·) represents the feature aggregation function.

[0137] In step S5, based on the policy-aware bundle representation and the policy-aware user representation, the interaction probability between the target user and each scenic spot bundle is obtained to obtain the target user's preference value for each scenic spot in the corresponding scenic spot bundle.

[0138] In this step, if Figure 2 As shown, based on the policy-aware bundling representation and policy-aware user representation, the target user u is obtained through the inner product operation m Bundled with attractions k The probability of interaction between them is expressed as:

[0139]

[0140] Among them, σ(·) is the sigmoid activation function.

[0141] On this basis, this step then maps the interaction probability to the target user um's preference value for each attraction in the attraction bundle bk through a uniform distribution method. It is expressed as:

[0142]

[0143] In step S6, the target user's preference value for each scenic spot in all scenic spot bundles is counted, and several scenic spots with high rankings are selected after sorting, and a scenic spot set is generated as a bundle for recommendation to the target user.

[0144] In this step, for the target user u m , sort the preference values ​​of all attractions in the attraction bundle in non-ascending order, select the attractions corresponding to the top preference values, and combine them to generate an attraction set as the bundle to the target user u m Make a recommendation.

[0145] So far, the embodiment of the present invention has completed the entire process of the strategy-awareness-based tourist attraction bundling recommendation method.

[0146] Embodiment 2:

[0147] The embodiment of the present invention provides a tourist attraction bundling recommendation system based on strategy perception, including:

[0148] An information acquisition module is used to acquire a user set and a user information set of each user, a scenic spot set and a scenic spot information set of each scenic spot, a user-scenic spot interaction matrix recording the target user's visit history, and a predefined scenic spot bundle set, so as to construct a bundle-scenic spot interaction graph; wherein any scenic spot bundle is a set consisting of several scenic spots in the scenic spot set;

[0149] A vector embedding module, used to obtain initial feature embedding vectors of the target user and each scenic spot based on the user information set of the target user and the scenic spot information set of each scenic spot;

[0150] A representation acquisition module, used to acquire a user token representation and a scenic spot token representation based on the initial feature embedding vectors of the target user and each scenic spot, combined with the user-scenic spot interaction matrix, and acquire a bundle representation of each scenic spot bundle;

[0151] A strategy-aware module, for introducing an HGT model with a self-attention mechanism based on the bundle-attraction interaction graph, the user token representation and the attraction token representation, to obtain a strategy-aware bundle representation and a strategy-aware user representation respectively;

[0152] A preference acquisition module, used to acquire the interaction probability between the target user and each scenic spot bundle based on the policy-aware bundle representation and the policy-aware user representation, so as to acquire the target user's preference value for each scenic spot in the corresponding scenic spot bundle;

[0153] The bundle recommendation module is used to count the target user's preference values ​​for each attraction in all attraction bundles, select several attractions with the highest ranking after sorting, and combine them to generate an attraction set as a bundle to recommend to the target user.

[0154] Embodiment 3:

[0155] An embodiment of the present invention provides a storage medium storing a computer program for policy-aware tourist attraction bundle recommendation, wherein the computer program enables a computer to execute the tourist attraction bundle recommendation method as described in Example 1.

[0156] Embodiment 4:

[0157] An embodiment of the present invention provides an electronic device, including:

[0158] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include a method for executing the tourist attraction bundling recommendation method as described in Example 1.

[0159] It can be understood that the policy-aware tourist attraction bundling recommendation system, storage medium and electronic device provided in the embodiments of the present invention correspond to the policy-aware tourist attraction bundling recommendation method provided in the embodiments of the present invention, and the explanations, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the tourist attraction bundling recommendation method, which will not be repeated here.

[0160] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0161] 1. Strategy perception capability: By introducing adaptive graph convolutional networks and self-attention mechanisms, the present invention can dynamically adjust the bundling strategy to better capture the complex correlation between scenic spot bundles and achieve more accurate recommendation combinations. This strategy perception capability helps the recommendation system to flexibly adapt and optimize the recommendation effect in complex tourism scenarios.

[0162] 2. Efficient computing model: By combining the hierarchical graph transformer (HGT) and the self-attention mechanism, the present invention can efficiently process large-scale tourist attractions and user data, ensuring the real-time and response speed of the system while generating recommendations.

[0163] 3. Highly personalized recommendations: The user's bundling preferences are accurately modeled, and can generate bundled recommendations of tourist attractions that meet the user's interests and preferences, thereby significantly improving user experience and satisfaction.

[0164] 4. Flexible scalability: It is not only suitable for the recommendation task of tourist attractions, but can also be flexibly extended to other types of bundled recommendation tasks, such as product recommendation, video recommendation, etc., thus having broad application potential and market adaptability.

[0165] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0166] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. 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 tourist attraction bundling recommendation method based on strategy perception, characterized in that: include: Obtain a user set and a user information set of each user, a scenic spot set and a scenic spot information set of each scenic spot, a user-scenic spot interaction matrix recording the target user's visit history, and a predefined scenic spot bundle set to construct a bundle-scenic spot interaction graph; wherein any scenic spot bundle is a set consisting of several scenic spots in the scenic spot set; Based on the user information set of the target user and the scenic spot information set of each scenic spot, respectively obtaining initial feature embedding vectors of the target user and each scenic spot; Based on the initial feature embedding vectors of the target user and each scenic spot, combined with the user-scenic spot interaction matrix, respectively obtain the user token representation and the scenic spot token representation, and obtain the bundle representation of each scenic spot bundle; Based on the bundle-attraction interaction graph, the user token representation and the attraction token representation, an HGT model with a self-attention mechanism is introduced to obtain a strategy-aware bundle representation and a strategy-aware user representation respectively; Based on the strategy-aware bundle representation and the strategy-aware user representation, obtaining an interaction probability between the target user and each scenic spot bundle to obtain a preference value of the target user for each scenic spot in the corresponding scenic spot bundle; The target user's preference value for each attraction in all attraction bundles is counted, and several attractions with the highest ranking are selected after sorting. The attractions are combined to generate an attraction set as a bundle for recommendation to the target user.

2. The tourist attraction bundle recommendation method according to claim 1, characterized in that: The construction process of the bundle-attraction interaction diagram is as follows: The attractions and attraction bundles are regarded as two different types of nodes, and the subordinate relationship between the two is regarded as an edge between the attraction node and the bundle node, so as to construct the bundle-attraction interaction graph.

3. The tourist attraction bundle recommendation method according to claim 1, characterized in that: Adaptive graph convolutional network is used to obtain user token representation and attraction token representation respectively, which can be expressed as: in, Represent the target user u m User token representation, attractions n Token representation of scenic spots; AGCN is an adaptive graph convolutional network; x m ,y n Represent the target user u m and attractions n The initial feature embedding vector of Based on the attraction token representations of all attractions in each attraction bundle, the corresponding bundle representation is obtained; it is expressed as: in, Indicates attraction bundle b k The bundling representation, Indicates attraction bundle b k No. o k Attractions Token representation, O k Bundle for attractions k The number of attractions.

4. The method for bundling tourist attractions as claimed in claim 3, characterized in that: The process of acquiring the policy-aware bundling representation includes: The HGT model with L layers stacked is introduced. For the lth layer of the HGT model, the attraction bundle b is calculated based on the bundle-attraction interaction graph G. k No. o k Attractions Positional encoding in graphs It is expressed as: in, represents the scenic spot bundle b learned from the previous layer k The latent vector of and They represent the attractions on the bundle-attraction interaction graph G. Node and Attraction Bundle b k The node's neighbor set; Bundle attractions k Position encoding and bundle representation of all scenic spots in Combined to obtain the enhanced bundle representation of the lth layer It is expressed as: in, Represents matrix addition operation; Simplify the self-attention mechanism, retain only the core structure, and calculate the scenic spot bundle b k Self-attention matrix between inliers It is expressed as: in, and is a trainable matrix, representing the Token representation of scenic spots The key and value mapping of ; d represents the embedding dimension; the superscript T represents transposition; Combined with the self-attention matrix, obtain the scenic spot bundle b in the lth layer k Characterization of bundling strategies It is expressed as: Among them, softmax(·) represents the softmax function; Perform an average pooling operation on the representation containing the bundling strategy to obtain the bundle b of each scenic spot k Strategy-aware bundling representation It is expressed as: Among them, mean_pooling is the average pooling operation.

5. The method for bundling tourist attractions as claimed in claim 4, characterized in that: The process of acquiring the policy-aware user representation includes: In the lth layer, for the target user u m , using the representation containing the bundling strategy to represent the user's historical bundling representation, recorded as: in, The target user u m The rth bundle that has been interacted with, R is the total number of bundles that the target user has interacted with; Calculate the target user u m The correlation matrix between its user history bundle representation It is expressed as: in, and Respectively represent user Token representation The key and value mapping matrix of ; Correlation Matrix After extracting features from the user's historical bundle representation, combine it with the user's Token representation Get the target user u m Policy-aware user representation It is expressed as: Among them, Attention(·) represents the user-bundle association matrix and user history bundle representation Perform interaction modeling; Aggregate(·) represents the feature aggregation function.

6. The method for bundling tourist attractions as claimed in claim 5, characterized in that: Based on the policy-aware bundle representation and policy-aware user representation, the target user u is obtained through the inner product operation. m Bundled with attractions k The probability of interaction between them is expressed as: Among them, σ(·) is the sigmoid activation function.

7. The method for bundling tourist attractions as claimed in claim 6, characterized in that: The interaction probability is mapped to the target user u through the uniform distribution method. m Bundle b for attractions k The preference value of each attraction in It is expressed as:

8. A strategy-aware tourist attraction bundling recommendation system, characterized in that: include: An information acquisition module is used to acquire a user set and a user information set of each user, a scenic spot set and a scenic spot information set of each scenic spot, a user-scenic spot interaction matrix recording the target user's visit history, and a predefined scenic spot bundle set, so as to construct a bundle-scenic spot interaction graph; wherein any scenic spot bundle is a set consisting of several scenic spots in the scenic spot set; A vector embedding module, used to obtain initial feature embedding vectors of the target user and each scenic spot based on the user information set of the target user and the scenic spot information set of each scenic spot; A representation acquisition module, used to acquire a user token representation and a scenic spot token representation based on the initial feature embedding vectors of the target user and each scenic spot, combined with the user-scenic spot interaction matrix, and acquire a bundle representation of each scenic spot bundle; A strategy-aware module, for introducing an HGT model with a self-attention mechanism based on the bundle-attraction interaction graph, the user token representation and the attraction token representation, to obtain a strategy-aware bundle representation and a strategy-aware user representation respectively; A preference acquisition module, used to acquire the interaction probability between the target user and each scenic spot bundle based on the policy-aware bundle representation and the policy-aware user representation, so as to acquire the target user's preference value for each scenic spot in the corresponding scenic spot bundle; The bundle recommendation module is used to count the target user's preference values ​​for each attraction in all attraction bundles, select several attractions with the highest ranking after sorting, and combine them to generate an attraction set as a bundle to recommend to the target user.

9. A storage medium, characterized in that: The computer program for recommending tourist attractions bundles based on strategy awareness is stored therein, wherein the computer program enables a computer to execute the tourist attractions bundle recommendation method as claimed in any one of claims 1 to 7.

10. An electronic device, characterized in that: include: one or more processors; Memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, the programs including a program for executing the tourist attraction bundling recommendation method as described in any one of claims 1 to 7.

Images