Product recommendation method and device, equipment, medium and product

Abstract

The invention discloses a product recommendation method and device, equipment, a medium and a product. Relates to the field of product recommendation, can be applied to the technical field of finance, and comprises the following steps: obtaining historical user transaction data and a product description feature set, and determining a two-part interaction graph set according to the historical user transaction data and the product description feature set; and inputting the two-part interaction graph set into a combined comparative learning recommendation model to obtain a target product recommendation list. Through the technical scheme of the invention, the corresponding two-part interaction graph can be constructed according to the transaction data of the user and the description characteristics of the product, a combined comparative learning recommendation model is used, the preference representation of the user is modeled and learned by using a graph comparative learning network, the preference and demand of the user on the product are identified, and the user experience is improved. And products which the users may be interested in are directionally recommended to the users, so that personalized recommendation service is realized.

Description

Technical Field

[0001] Embodiments of the present invention relate to the technical field of product recommendation, and in particular to a product recommendation method, apparatus, device, medium, and product. Background Art

[0002] With the advancement of information technology, the internet is flooded with a rich and diverse array of electronic resources, satisfying people's daily needs. However, this has also led to the problem of information overload. As recipients of online resources, users often spend a considerable amount of time sifting through resources that meet their needs. Search engines, as an effective means of addressing information overload, allow users to quickly filter through vast amounts of online resources using simple descriptions. However, this method requires users to clearly define their needs and provide accurate keywords.

[0003] When user needs are unclear or difficult to accurately describe, recommendation systems become another effective tool for alleviating information overload. Unlike search engines, recommendation systems only require users to provide vague descriptions of their needs. They can then mine useful information from vast resources and help users discover content of potential interest.

[0004] However, current deep learning-based recommendation algorithms have low transparency in their computational process and are unable to make convincing recommendations based on user needs. Therefore, a better recommendation method is urgently needed. Summary of the Invention

[0005] The embodiments of the present invention provide a product recommendation method, apparatus, device, medium, and product to identify user preferences and needs for products, and then recommend products that may be of interest to users, thereby realizing personalized recommendation services.

[0006] According to one aspect of the present invention, a product recommendation method is provided, comprising:

[0007] Obtaining historical user transaction data and a product description feature set, and determining a bipartite interaction graph set based on the historical user transaction data and the product description feature set;

[0008] The set of bipartite interaction graphs is input into a joint contrastive learning recommendation model to obtain a target product recommendation list.

[0009] According to another aspect of the present invention, a product recommendation device is provided, the device comprising:

[0010] an acquisition and determination module, configured to acquire historical user transaction data and a product description feature set, and determine a set of bipartite interaction graphs based on the historical user transaction data and the product description feature set;

[0011] An input module is used to input the set of bipartite interaction graphs into a joint contrastive learning recommendation model to obtain a target product recommendation list.

[0012] According to another aspect of the present invention, an electronic device is provided, comprising:

[0013] at least one processor; and

[0014] a memory communicatively connected to the at least one processor; wherein,

[0015] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the product recommendation method described in any embodiment of the present invention.

[0016] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the product recommendation method described in any embodiment of the present invention when executed.

[0017] According to another aspect of the present invention, an embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the product recommendation method described in any embodiment of the present invention.

[0018] The present invention obtains a list of target product recommendations by acquiring historical user transaction data and a set of product description features, determining a bipartite interaction graph based on these data, and then inputting the bipartite interaction graph into a joint contrastive learning recommendation model. The present invention's technical solution constructs a corresponding bipartite interaction graph based on user transaction data and product description features. Using the joint contrastive learning recommendation model, a graph contrastive learning network is used to model and learn user preference representations, identify user preferences and needs for products, and then provide targeted recommendations of products that may be of interest to users, achieving personalized recommendation services.

[0019] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 is a flow chart of a product recommendation method in an embodiment of the present invention;

[0022] Figure 2 Schematic diagram of the structure of a joint contrastive learning recommendation model in an embodiment of the present invention;

[0023] Figure 3 is a structural diagram of a product recommendation device in an embodiment of the present invention;

[0024] Figure 4 3 is a schematic diagram of the structure of an electronic device for implementing the product recommendation method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only 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 making creative efforts should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and their inclusion are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] It is understandable that before using the technical solutions disclosed in the various embodiments of this disclosure, the type, scope of use, usage scenarios, etc. of the personal information involved in this disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0028] The information collected is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with the relevant laws, regulations and standards of relevant countries and regions, take necessary confidentiality measures, do not violate public order and good customs, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0029] Provide users with corresponding operation entrances for them to choose to agree or reject the automated decision-making results; if the user chooses to reject, they will enter the expert decision-making process.

[0030] Example 1

[0031] Figure 1 This is a flow chart of a product recommendation method in an embodiment of the present invention. This embodiment is applicable to product recommendation situations. This method can be executed by a product recommendation device in an embodiment of the present invention. The device can be implemented in software and / or hardware. Figure 1 As shown, the method specifically includes the following steps:

[0032] S101. Obtain historical user transaction data and a product description feature set, and determine a bipartite interaction graph set based on the historical user transaction data and the product description feature set.

[0033] In this embodiment, historical user transaction data may be transaction data generated by users during historical interactions, such as purchases, fixed investments, refunds, and the like.

[0034] It should be noted that the product description feature set may be a set consisting of a plurality of descriptive feature information used to describe a financial product. This embodiment does not limit the specific product description features.

[0035] The bipartite interaction graph set may be a set of multiple bipartite interaction graphs constructed based on the user's historical transaction data and the product's description features to express different interaction relationship data between the user and the product.

[0036] In this embodiment, user-item interaction data can be represented using a bipartite interaction graph, where users and items are nodes, and interactions between different users and items are edges connecting the nodes. Therefore, the recommendation task can be transformed into a link prediction problem. A graph convolutional network can then be used to extract information from the bipartite interaction graph to perform the recommendation task. By leveraging the message passing and neighborhood aggregation operations of the graph convolutional network, node representations incorporate their own characteristics and different types of connection relationship information, resulting in a richer representation of user and item nodes.

[0037] Specifically, historical transaction data of users for whom product recommendations are to be made and a set of product description features are obtained, and a set of bipartite interaction graphs is constructed based on the historical user transaction data and the set of product description features.

[0038] S102: Input the bipartite interaction graph set into a joint contrastive learning recommendation model to obtain a target product recommendation list.

[0039] In this embodiment, the joint contrastive learning recommendation model can be an improved graph contrastive learning network, which is used to perform the recommended product list generation task and the contrastive learning task based on the bipartite interaction graph set, and finally generate a product recommendation list suitable for the user and recommend suitable products to the user.

[0040] Traditional recommendation algorithms are often limited by insufficient supervisory signals—scarce or missing user interaction data—making it difficult for models to provide accurate, personalized recommendations. The success of contrastive learning in computer vision offers a new approach to this problem. By generating positive and negative examples as self-supervisory signals for model learning, contrastive learning alleviates the problem of scarce supervisory signals and improves model robustness.

[0041] The Graph Contrastive Learning Network (GCLN) is a deep learning architecture that combines graph convolutional neural networks with contrastive learning. The network processes data through a graph structure and uses contrastive learning strategies to enhance the discriminability of node representations. This allows for more effective feature representations to be learned from graph data, alleviating the challenges of sparse or missing interaction data in recommendation tasks.

[0042] The target product recommendation list may be a recommendation list of products suitable for the user.

[0043] Specifically, a set of bipartite interaction graphs constructed based on historical user transaction data and product description feature sets is input into a joint contrastive learning recommendation model to ultimately obtain a list of target product recommendations.

[0044] The present invention obtains a list of target product recommendations by acquiring historical user transaction data and a set of product description features, determining a bipartite interaction graph based on these data, and then inputting the bipartite interaction graph into a joint contrastive learning recommendation model. The present invention's technical solution constructs a corresponding bipartite interaction graph based on user transaction data and product description features. Using the joint contrastive learning recommendation model, a graph contrastive learning network is used to model and learn user preference representations, identify user preferences and needs for products, and then provide targeted recommendations of products that may be of interest to users, achieving personalized recommendation services.

[0045] Optionally, the set of bipartite interaction graphs includes: a user-product bipartite interaction graph, a user-product feature graph, and an invisible user-user relationship graph.

[0046] In this embodiment, the user-product bipartite interaction graph can be a bipartite interaction graph used to describe a user's past transaction records. In this graph, users and products are represented by nodes, and undirected edges <user, product> represent different types of relationships between users and products, including viewing, adding to favorites, and purchasing.

[0047] In this embodiment, the user-product feature graph can be a bipartite interaction graph that describes the different attribute properties of products that a user has purchased in the past, such as product profitability and product management manager, from the perspective of product features. The undirected edge <user, product feature> identifies the connection between the user and each product feature.

[0048] In this embodiment, the hidden user-user relationship graph can be a bipartite interaction graph that describes users and users with whom they have hidden links. Users connected on this graph have similar purchase records, and from a statistical perspective, these users have hidden links that are not directly observable.

[0049] The set of bipartite interaction graphs is input into the joint contrastive learning recommendation model to obtain a list of target product recommendations, including:

[0050] The user-product bipartite interaction graph is input into the joint contrastive learning recommendation model to obtain local user nodes and local product nodes.

[0051] Specifically, the user-product bipartite interaction graph is input into the joint contrastive learning recommendation model, and graph convolution calculation is performed on the user-product bipartite interaction graph to obtain local user nodes and local product nodes.

[0052] The initial product recommendation list is determined based on the similarity scores between the local user node and the local product node.

[0053] The initial product recommendation list may be a list of products recommended to the user determined based on similarity scores between local user nodes and local product nodes.

[0054] Specifically, the inner product is used to calculate the similarity score between the local user node and the local product node to represent the priority weight of the recommendation, and the top K (the K value can be set by the user based on actual conditions or experience, and this embodiment does not limit this) products are recommended to the user as the initial product recommendation list.

[0055] The user-product feature graph and the hidden user-user relationship graph are input into the joint contrastive learning recommendation model to obtain user nodes described by product features and local hidden user nodes.

[0056] Specifically, the user-product feature graph and the implicit user-user relationship graph are input into the joint contrastive learning recommendation model, and graph convolution calculations are performed on the user-product feature graph and the implicit user-user relationship graph respectively, and the aggregate output is user nodes described by product features and local implicit user nodes.

[0057] The global user node is determined according to the local user node, the user node described by the product feature, and the local hidden user node.

[0058] The global user node may be a global user node that describes user interests from different perspectives.

[0059] Specifically, based on the spatial attention aggregation mechanism of the feedforward neural network, the global user node is determined according to the local user nodes, the user nodes described by product features, and the local hidden user nodes.

[0060] According to the distance between the local user node and the global user node in the vector space, the initial product recommendation list is updated to obtain the target product recommendation list.

[0061] Specifically, the distance between the two types of nodes, local user nodes and global user nodes, in the vector space is calculated. The size of the distance represents the priority weight of the recommendation. The initial product recommendation list is updated to obtain the target product recommendation list.

[0062] The embodiments of the present invention utilize descriptive labels generated during transactions, including investment preferences and user levels, from the user's perspective, and descriptive labels of financial products, such as product risk ratings, from the product's perspective, to construct a user's implicit social network. A graph convolutional neural network is then used to model and learn the user's implicit investment preference representation. A corresponding bipartite interaction graph is constructed based on the interaction information between the user and the product, and a graph convolutional neural network is used to model and learn the user's explicit investment preferences. A graph contrast learning network is used to align the user's explicit and implicit investment preferences, alleviating the problem of incomplete user investment preference representation and inaccurate recommendations caused by sparse transaction signals.

[0063] Optionally, the joint contrastive learning recommendation model includes: an initial embedding layer, a recommendation task main view channel, and a contrast task implicit user relationship graph channel.

[0064] The initial embedding layer includes: user nodes, product nodes, and description feature nodes.

[0065] In this embodiment, the randomly generated node representations of the initial embedding layer of the joint contrastive learning recommendation model include the following three types of nodes: user nodes, product nodes, and description feature nodes.

[0066] Input the user-product bipartite interaction graph into the joint contrastive learning recommendation model to obtain local user nodes and local product nodes, including:

[0067] The user-product bipartite interaction graph, user nodes, and product nodes are input into the recommendation task main view channel to obtain local user nodes and local product nodes.

[0068] Specifically, a specific task is performed in the recommendation task main view channel of the joint contrastive learning recommendation model: the user-product interaction graph, user nodes, and product nodes are input, and graph convolution calculation is performed on the user-product interaction graph to obtain the information aggregation results of the user nodes and product nodes after each layer of convolution propagation in the user-product two-part interaction graph. The output of each layer of convolution is aggregated to obtain local user node representation and local product node representation.

[0069] The embodiment of the present invention constructs a joint contrastive learning recommendation model. The financial product recommendation algorithm based on the improved graph contrastive learning network can comprehensively consider multiple aspects such as the user's investment preferences, implicit social network, interest preferences, etc., and provide each user with financial product services that meet their needs. This solves the problem that the recommendation effect of the existing recommendation algorithm is affected by the sparsity of the interaction data set, and there is a user cold start problem that makes it difficult for the algorithm model to accurately model the user's interest preferences.

[0070] Optionally, a joint contrastive learning recommendation model is input based on the user-product feature graph and the hidden user-user relationship graph to obtain user nodes and local hidden user nodes described by product features, including:

[0071] According to the user-product feature graph, the invisible user-user relationship graph, the user nodes and the description feature nodes input comparison task implicit user relationship graph channel, the user nodes described by product features and the local invisible user nodes are obtained.

[0072] Specifically, a specific task is performed in the implicit user relationship graph channel of the comparison task: input the user-product feature graph and the implicit user-user relationship graph, user nodes, and nodes describing product features, perform graph convolution calculations on the user-product feature graph and the implicit user-user relationship graph respectively, and aggregate the output of local product feature nodes, user nodes described by product features, and local implicit user nodes.

[0073] In actual operation, it is also necessary to construct a contrastive learning task, align user nodes in different channels, use contrastive learning functions, optimize local user nodes and global user nodes, store global user node representations and local product node representations, and calculate the distance between the two types of node representations in the vector space. The size of the distance represents the priority weight of the recommendation.

[0074] The embodiment of the present invention utilizes the clustering relationship of user groups and the static characteristics of financial products to expand the relationship between user nodes, so that the joint contrastive learning recommendation model has better performance in user cold start problem than the traditional recommendation model based on graph convolutional neural network collaborative filtering method; the recommendation algorithm of the present invention constructs a joint learning method to align user representations collected from recommendation tasks and contrastive learning tasks. Compared with the recommendation algorithm that does not apply the contrastive learning mode, the contrastive learning recommendation model performs better in the problem of item sparsity, alleviating the long-tail problem that is common in recommendation scenarios due to the uneven distribution of interaction data.

[0075] Optionally, a set of bipartite interaction graphs is determined based on historical user transaction data and a set of product description features, including:

[0076] Generate a user-product bipartite interaction graph based on historical user transaction data.

[0077] Specifically, a user-product bipartite interaction graph is extracted from raw user transaction data (i.e., historical user transaction data). Users and products are represented by nodes, and undirected edges (<user, product>) identify different types of relationships between users and products, including viewing, adding to favorites, and purchasing. This interaction graph is used to describe a user's past transaction history.

[0078] Generate a user-product feature graph based on historical user transaction data and product description feature sets.

[0079] Specifically, a user-product feature graph is constructed from raw user transaction data and product description feature sets. A hyperedge of <user, product attribute, product> is constructed, and undirected edges of <user, product feature> are extracted from this to construct a user-product feature interaction graph. Undirected edges <user, product feature> identify the connections between users and product features. This interaction graph describes the various attributes of products a user has previously purchased, such as product profitability and product manager, from the perspective of product features.

[0080] Generate an invisible user-user relationship graph based on the user-product bipartite interaction graph.

[0081] Specifically, we extract the third-order hyperedges <user, product, user> and the fifth-order hyperedges <user-item-user-item-user> from the <user, product> of the user-product bipartite interaction graph. Only the starting and ending points of each hyperedge are retained, resulting in undirected edges <user, user> that identify users and users with whom they have hidden links. From this, we construct a hidden user-user relationship graph. Users connected in this graph have similar purchase histories, and from a statistical perspective, these users have hidden links that are not directly observable.

[0082] A set of bipartite interaction graphs is determined based on the user-product bipartite interaction graph, the user-product feature graph, and the invisible user-user relationship graph.

[0083] In practice, we collect interaction data generated by users and descriptive information about financial products, and use this information to construct implicit social networks. Traditional recommendation algorithms based on graph convolutional neural networks, when learning and modeling user investment preference representations on bipartite interaction graphs converted from user-item interaction datasets, can only capture third- and fifth-order high-level user links, such as "user-item-user" and "user-item-user-item-user," to explore user groups that may have similar investment preferences to the target user. However, this approach is limited by the distribution frequency and interaction quality of interaction datasets.

[0084] This embodiment of the present invention expands existing high-level user relationships, such as "user-item (attribute A)-user," by using descriptive attributes of financial products, such as product risk ratings and return on investment. These expanded high-level user relationships are then cleaned and filtered to reveal a social network of users from the perspective of descriptive attributes, such as a social network of users with the same product risk rating. The algorithm learns and optimizes user preference representations on the network. Compared to existing methods, it provides connections between users from different perspectives and further shortens paths between users, alleviating smoothing issues associated with graph convolutional computations.

[0085] Optionally, after obtaining the target product recommendation list, the following is also included:

[0086] Push the target product recommendation list and obtain the interaction behavior and transaction status for the target product recommendation list.

[0087] For example, the interactive behavior may be that the user clicks to view a product in the target product recommendation list, and the transaction status may be, for example, purchase, fixed investment, etc.

[0088] In practice, when a user accesses the system implementing this financial product recommendation method, the search server searches for the corresponding user node based on the user ID and pushes a preset product list to the user. The server then obtains the user's interaction behavior with the current product recommendation list and the user's current transaction status.

[0089] A target product is determined from the target product recommendation list according to the interaction behavior and the transaction status, and the target product recommendation list is reordered based on the target product.

[0090] The target product refers to the product selected by the user in the target product recommendation list.

[0091] In actual operation, the currently pushed list is re-sorted based on the user's interaction with the current list. Specifically, based on the user's current transaction status and the previous operation in the system (including viewing, adding to favorites, and purchasing products with certain characteristics), qualified products are screened from the candidate product list and replaced with the current recommended product list. The user's transaction behavior in this recommendation system is recorded, including the order of interaction and the order of interaction with the re-sorted list, to optimize the user node representation in the recommendation server.

[0092] Traditional recommendation algorithms suffer from a "cold start problem" because users entering the system for the first time haven't interacted with any financial products, making it difficult for the algorithm to model their investment preferences. A common industry solution is to integrate the user's external social network to expand interactions between the user and financial products, but this can easily lead to privacy issues.

[0093] This embodiment of the present invention provides users with both offline and online recommendation functions. The offline recommendation process generates a set of K candidate items for recommendation, i.e., a target product recommendation list. Based on this, the online recommendation process re-ranks the static recommendation results according to the user's current behavior and status, providing users with more accurate, personalized recommendations. The algorithm proposed in this embodiment overcomes the shortcomings of existing technologies, which require users to first complete registration information or be recommended a list of popular financial products, as well as query results in this algorithm system. Based on this information, implicit connections between users and financial products are found to complete recommendations. This embodiment constructs an implicit social network based on the user's descriptive tags, providing a recommendation path for each recommended product, improving the interpretability of the recommendation results.

[0094] Example 2

[0095] Figure 2 This is a schematic diagram of the structure of a federated comparative learning recommendation model according to an embodiment of the present invention. Building on the previous embodiment and based on the structural features of the federated comparative learning recommendation model, this embodiment provides a detailed explanation of the steps in the previous embodiment, including inputting a bipartite interaction graph set into the federated comparative learning recommendation model to obtain a list of target product recommendations.

[0096] like Figure 2 As shown in the figure, the joint contrastive learning recommendation model includes an initial embedding layer, a recommendation task main view channel, and a contrast task implicit user relationship graph channel.

[0097] Among them, the initial embedding layer of the joint contrastive learning recommendation model randomly generates node representations including the following three types of nodes: user nodes, product nodes, and description feature nodes.

[0098] In actual operation, a specific task is performed in the recommendation task main view channel of the joint contrastive learning recommendation model: the user-product interaction graph, user nodes, and product nodes are input, and graph convolution calculation is performed on the user-product interaction graph. The calculation method is shown in the following formula:

[0099]

[0100] in, and They represent the information aggregation results of user node u and product node i after convolutional propagation at the l+1th layer in the user-product bipartite interaction graph; and In the user-product bipartite interaction graph, it represents the set of neighbor nodes of user node u and product node i. Aggregating the outputs of the three convolution layers, we get the local user node representation and the local product node representation. The calculation method is as follows:

[0101]

[0102] in, and They respectively represent the local user node representation and the local product node representation obtained in the user-product bipartite interaction graph.

[0103] Next, the prediction layer can be used to perform recommendation tasks. The inner product is used to calculate the similarity score between the local user node and the local product node, which is used to represent the priority weight of the recommendation. The top K products are recommended to the user as the product recommendation list. The calculation method is shown in the following formula:

[0104]

[0105] in, is the similarity score.

[0106] In the specific implementation process, the personal Bayesian loss function can be used to optimize the node representation and improve the recommendation performance. The calculation method is shown in the following formula:

[0107]

[0108] Among them, (u,i) and represents the interaction between user u and product i observed in the current batch data; N u represents the set of all items that user u interacts with in the current mini-batch data; (u, j) and represents the similarity score between the negative sample pair and the negative sample pair; σ represents the sigmoid activation function.

[0109] In the actual operation process, specific tasks are performed in the implicit user relationship graph channel of the comparison task: input user-product feature graph and implicit user-user relationship graph, user node, description feature node, perform three-layer graph convolution calculation on the user-product feature graph and implicit user-user relationship graph respectively, and aggregate the output to describe the user node using product features. Local hidden user node The calculation process is the same as the above graph convolution calculation method, so I will not go into details here.

[0110] After that, we can construct a contrastive learning task to align user nodes in different channels. The calculation method is as follows:

[0111]

[0112] Among them, E u Represents the global user node representation that describes user interests from different perspectives; MEAN() represents the spatial attention aggregation mechanism based on feedforward neural network.

[0113] Use contrastive learning function to optimize local user nodes and global user nodes. The specific calculation method is shown in the following formula:

[0114]

[0115] Among them, E u , They represent the global node representation and local node representation of the same user node, E j Indicates the negative sample pairs of nodes of the same type except this node in the current minimum batch data; τ represents the temperature parameter of the softmax function; B represents the same minimum batch data.

[0116] In traditional recommendation algorithms, most of the observed interaction data between users and items is concentrated in a few top items, which only occupy a small part of the item set, and the interaction data is unevenly distributed. At this time, there is a "long tail problem", and the recommendation algorithm cannot accurately model the user's interest preferences to make further personalized recommendations. The algorithm proposed in the embodiment of the present invention uses descriptive labels generated in the transaction process from the user's perspective, including investment preferences and user levels, and uses descriptive labels of the product itself, such as product risk rating, from the perspective of financial products. It constructs the user's implicit social network and uses a graph convolutional neural network to model and learn the user's implicit investment preference representation. It constructs a corresponding bipartite interaction graph based on the interaction information between the user and the product, uses a graph convolutional neural network to model and learn the user's explicit investment preferences, and uses a graph contrast learning network to align the user's explicit investment preferences with the implicit investment preferences, fully mining the user's implicit social network, so as to effectively alleviate the problem of incomplete user investment preference representation and inaccurate recommendations caused by sparse transaction signals, as well as the problem of long-tail financial product recommendation.

[0117] Example 3

[0118] Figure 3 This is a schematic diagram of the structure of a product recommendation device in an embodiment of the present invention. This embodiment is applicable to product recommendation situations. The device can be implemented in software and / or hardware. The device can be integrated into any device that provides product recommendation functions, such as Figure 3 As shown, the product recommendation device specifically includes: an acquisition and determination module 201 and an input module 202.

[0119] The acquisition and determination module 201 is configured to acquire historical user transaction data and a product description feature set, and determine a bipartite interaction graph set based on the historical user transaction data and the product description feature set;

[0120] The input module 202 is configured to input the set of bipartite interaction graphs into a joint contrastive learning recommendation model to obtain a target product recommendation list.

[0121] Optionally, the set of bipartite interaction graphs includes: a user-product bipartite interaction graph, a user-product feature graph, and an invisible user-user relationship graph;

[0122] The input module 202 includes:

[0123] A first input unit is configured to input the user-product bipartite interaction graph into a joint contrastive learning recommendation model to obtain a local user node and a local product node;

[0124] A first determining unit, configured to determine an initial product recommendation list according to a similarity score between the local user node and the local product node;

[0125] A second input unit is configured to input the user-product feature graph and the hidden user-user relationship graph into the joint contrastive learning recommendation model to obtain user nodes and local hidden user nodes described using product features;

[0126] A second determining unit is configured to determine a global user node based on the local user node, the user node described using product features, and the local hidden user node;

[0127] An updating unit is used to update the initial product recommendation list according to the distance between the local user node and the global user node in the vector space to obtain a target product recommendation list.

[0128] Optionally, the joint contrastive learning recommendation model includes: an initial embedding layer, a recommendation task main view channel, and a contrast task implicit user relationship graph channel;

[0129] The initial embedding layer includes: user nodes, product nodes, and description feature nodes;

[0130] The first input unit is specifically used for:

[0131] The user-product bipartite interaction graph, the user node, and the product node are input into the recommendation task main view channel to obtain a local user node and a local product node.

[0132] Optionally, the second input unit is specifically configured to:

[0133] The comparison task implicit user relationship graph channel is input according to the user-product feature graph, the implicit user-user relationship graph, the user node and the description feature node to obtain user nodes and local implicit user nodes described using product features.

[0134] Optionally, the acquisition and determination module 201 is specifically configured to:

[0135] generating a user-product bipartite interaction graph based on the historical user transaction data;

[0136] generating a user-product feature graph based on the historical user transaction data and the product description feature set;

[0137] Generate an invisible user-user relationship graph based on the user-product bipartite interaction graph;

[0138] A set of bipartite interaction graphs is determined according to the user-product bipartite interaction graph, the user-product feature graph, and the invisible user-user relationship graph.

[0139] Optionally, the device further includes:

[0140] a push and acquisition unit, configured to push the target product recommendation list and acquire interaction behaviors and transaction status for the target product recommendation list;

[0141] A determination and updating unit is configured to determine a target product from the target product recommendation list according to the interaction behavior and the transaction status, and to reorder the target product recommendation list based on the target product.

[0142] The above-mentioned product can execute the product recommendation method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0143] Example 4

[0144] Figure 4 A schematic diagram of the structure of an electronic device 30 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0145] like Figure 4 As shown, the electronic device 30 includes at least one processor 31 and a memory, such as a read-only memory (ROM) 32, a random access memory (RAM) 33, etc., which is communicatively connected to the at least one processor 31. The memory stores a computer program that can be executed by the at least one processor. The processor 31 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 32 or the computer program loaded from the storage unit 38 into the random access memory (RAM) 33. Various programs and data required for the operation of the electronic device 30 can also be stored in the RAM 33. The processor 31, ROM 32, and RAM 33 are connected to each other via a bus 34. An input / output (I / O) interface 35 is also connected to the bus 34.

[0146] Multiple components in the electronic device 30 are connected to the I / O interface 35, including an input unit 36, such as a keyboard, a mouse, etc.; an output unit 37, such as various types of displays, speakers, etc.; a storage unit 38, such as a magnetic disk, an optical disk, etc.; and a communication unit 39, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 39 allows the electronic device 30 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0147] The processor 31 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the processor 31 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 31 executes the various methods and processes described above, such as the product recommendation method:

[0148] Obtaining historical user transaction data and a product description feature set, and determining a bipartite interaction graph set based on the historical user transaction data and the product description feature set;

[0149] The set of bipartite interaction graphs is input into a joint contrastive learning recommendation model to obtain a target product recommendation list.

[0150] In some embodiments, the product recommendation method can be implemented as a computer program that is tangibly contained in a computer-readable storage medium, such as a storage unit 38. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 30 via the ROM 32 and / or the communication unit 39. When the computer program is loaded into the RAM 33 and executed by the processor 31, one or more steps of the product recommendation method described above can be performed. Alternatively, in other embodiments, the processor 31 can be configured to execute the product recommendation method in any other appropriate manner (for example, by means of firmware).

[0151] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system comprising at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0152] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0153] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0154] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0155] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0156] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0157] In one embodiment, the present invention further includes a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the product recommendation method of any embodiment of the present invention.

[0158] The computer program product may be implemented by writing computer program code for performing the operations of the present invention in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0159] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0160] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A product recommendation method, characterized in that: include: Obtaining historical user transaction data and a product description feature set, and determining a bipartite interaction graph set based on the historical user transaction data and the product description feature set; The set of bipartite interaction graphs is input into a joint contrastive learning recommendation model to obtain a target product recommendation list.

2. The method according to claim 1, wherein the set of bipartite interaction graphs comprises: User-product bipartite interaction diagram, user-product feature diagram, and invisible user-user relationship diagram; The set of bipartite interaction graphs is input into the joint contrastive learning recommendation model to obtain a target product recommendation list, including: Inputting the user-product bipartite interaction graph into a joint contrastive learning recommendation model to obtain local user nodes and local product nodes; Determining an initial product recommendation list based on similarity scores between the local user node and the local product node; Inputting the user-product feature graph and the hidden user-user relationship graph into the joint contrastive learning recommendation model to obtain user nodes described by product features and local hidden user nodes; Determine a global user node according to the local user node, the user node described using product features, and the local hidden user node; The initial product recommendation list is updated according to the distance between the local user node and the global user node in the vector space to obtain a target product recommendation list.

3. The method according to claim 2, characterized in that The joint contrastive learning recommendation model includes: an initial embedding layer, a recommendation task main view channel, and a contrast task implicit user relationship graph channel; The initial embedding layer includes: user nodes, product nodes, and description feature nodes; Inputting the user-product bipartite interaction graph into the joint contrastive learning recommendation model to obtain local user nodes and local product nodes, including: The user-product bipartite interaction graph, the user node, and the product node are input into the recommendation task main view channel to obtain a local user node and a local product node.

4. The method according to claim 2, characterized in that The user-product feature graph and the hidden user-user relationship graph are input into the joint contrastive learning recommendation model to obtain user nodes and local hidden user nodes described by product features, including: The comparison task implicit user relationship graph channel is input according to the user-product feature graph, the implicit user-user relationship graph, the user node and the description feature node to obtain user nodes and local implicit user nodes described using product features.

5. The method according to claim 1, wherein Determining a bipartite interaction graph set based on the historical user transaction data and the product description feature set includes: generating a user-product bipartite interaction graph based on the historical user transaction data; generating a user-product feature graph based on the historical user transaction data and the product description feature set; Generate an invisible user-user relationship graph based on the user-product bipartite interaction graph; A set of bipartite interaction graphs is determined according to the user-product bipartite interaction graph, the user-product feature graph, and the invisible user-user relationship graph.

6. The method according to claim 1, characterized in that After obtaining the target product recommendation list, it also includes: Push the target product recommendation list and obtain the interactive behavior and transaction status for the target product recommendation list; A target product is determined from the target product recommendation list according to the interaction behavior and the transaction status, and the target product recommendation list is reordered based on the target product.

7. A product recommendation device, characterized in that: include: an acquisition and determination module, configured to acquire historical user transaction data and a product description feature set, and determine a set of bipartite interaction graphs based on the historical user transaction data and the product description feature set; An input module is used to input the set of bipartite interaction graphs into a joint contrastive learning recommendation model to obtain a target product recommendation list.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor. The computer program is executed by the at least one processor to enable the at least one processor to perform the product recommendation method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the product recommendation method according to any one of claims 1 to 6 when executed.

10. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the product recommendation method according to any one of claims 1 to 6.

Images