Recommendation processing method and device, computer equipment and storage medium
By constructing a heterogeneous information network and a semantic recognition model, the similarity between new accounts and historical accounts is analyzed, which solves the problem of low accuracy in cold start for new accounts and achieves efficient product recommendation without the acquisition of social information.
Patent Information
- Application Number
- CN202411726933.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-28
- Publication Date
- 2025-02-11
AI Technical Summary
Traditional recommendation systems have low accuracy when new accounts are cold-starting, and they require access to the account's private social information, which raises privacy and authorization issues.
A heterogeneous information network is constructed based on the attribute information of new and historical accounts. The similarity of nodes is analyzed through a semantic recognition model, and the data is classified to recommend products.
It improves the accuracy and precision of cold start recommendations for new accounts without needing to obtain account social information, and uses common group information for product recommendations.
Smart Images
Figure CN119417575A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of big data technology, and in particular to a recommendation processing method, apparatus, computer equipment, storage medium and computer program product. Background Art
[0002] The recommendation system is mainly used to recommend items that an account may like. For a platform, its accounts and recommended items are constantly growing and changing. How to recommend items that satisfy new accounts or how to recommend new items to accounts that like it are the key research contents of the recommendation system.
[0003] In traditional technology, the friend information of a new account on a social networking site is usually imported, and then the items that the new account's social friends like are recommended to the new account. However, this method requires obtaining the account's private social information, and also involves the account and the privacy and authorization information contained in the account, and the accuracy is not high. Therefore, there is a problem of low cold start accuracy in recommending items to new accounts. Summary of the invention
[0004] Based on this, it is necessary to provide a recommendation processing method, device, computer equipment, computer-readable storage medium and computer program product that can improve the accuracy of cold start recommendations for new accounts in response to the above technical problems.
[0005] In a first aspect, the present application provides a recommendation processing method. The method comprises:
[0006] Based on the account attribute information of the new account and the account attribute information of the historical account, construct a heterogeneous information network of the account attribute information;
[0007] Acquire multiple groups of node wandering sequences of the heterogeneous information network, and train a semantic recognition model based on the multiple groups of node wandering sequences;
[0008] Obtaining similarity information between nodes in the heterogeneous information network through the semantic recognition model;
[0009] Classify each node in the heterogeneous information network according to the similarity information to obtain a target classification result of each node;
[0010] According to the target classification result, the new account is recommended to obtain a target recommendation result for the new account.
[0011] In one embodiment, obtaining multiple groups of node walk sequences of the heterogeneous information network, and training a semantic recognition model based on the multiple groups of node walk sequences includes:
[0012] For the account attribute information of the nodes in the heterogeneous information network, setting attribute conversion information and wandering path length information;
[0013] According to the attribute conversion information and the walking path length information, performing multiple random walk processes on the nodes in the heterogeneous information network to obtain the multiple groups of node walking sequences;
[0014] Based on each group of the node walk sequences, the semantic recognition models to be trained are iteratively trained to obtain a plurality of trained semantic recognition models.
[0015] In one embodiment, obtaining similarity information between nodes in the heterogeneous information network through the semantic recognition model includes:
[0016] Obtaining multiple similarity matrices of each node through multiple semantic recognition models;
[0017] The multiple similarity matrices are averaged to obtain similarity information between the nodes.
[0018] In one embodiment, each node in the heterogeneous information network is classified according to the similarity information to obtain a target classification result of each node, including:
[0019] Classifying each node in the heterogeneous information network according to the similarity information to obtain an initial classification result of each node;
[0020] The outliers in the initial classification result are reclassified to obtain the target classification result of each node.
[0021] In one embodiment, the target classification result includes a plurality of classification groups; each of the classification groups includes a node representing the new account and a node representing the historical account;
[0022] The step of performing recommendation processing on the new account according to the target classification result to obtain a target recommendation result for the new account includes:
[0023] According to the historical information corresponding to the historical accounts in each classification group, the new accounts in each classification group are recommended to obtain the target recommendation results of the new accounts in each classification group.
[0024] In one embodiment, based on the account attribute information of the new account and the account attribute information of the historical account, a heterogeneous information network of the account attribute information is constructed, including:
[0025] Performing feature extraction and data preprocessing on the account information of the new account to obtain account attribute information of the new account;
[0026] Performing feature extraction and data preprocessing on the account information of the historical account to obtain account attribute information of the historical account;
[0027] The heterogeneous information network is constructed according to the account attribute information of the new account and the historical account.
[0028] In a second aspect, the present application also provides a recommendation processing device. The device comprises:
[0029] A grid construction module, used to construct a heterogeneous information network of the account attribute information based on the account attribute information of the new account and the account attribute information of the historical account;
[0030] A model training module, used to obtain multiple groups of node wandering sequences of the heterogeneous information network, and train a semantic recognition model based on the multiple groups of node wandering sequences;
[0031] A similarity acquisition module, used to obtain similarity information between nodes in the heterogeneous information network through the semantic recognition model;
[0032] A node classification module, used to classify each node in the heterogeneous information network according to the similarity information to obtain a target classification result of each node;
[0033] The account recommendation module is used to perform recommendation processing on the new account according to the target classification result to obtain the target recommendation result of the new account.
[0034] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:
[0035] Based on the account attribute information of the new account and the account attribute information of the historical account, construct a heterogeneous information network of the account attribute information;
[0036] Acquire multiple groups of node wandering sequences of the heterogeneous information network, and train a semantic recognition model based on the multiple groups of node wandering sequences;
[0037] Obtaining similarity information between nodes in the heterogeneous information network through the semantic recognition model;
[0038] Classify each node in the heterogeneous information network according to the similarity information to obtain a target classification result of each node;
[0039] According to the target classification result, the new account is recommended to obtain a target recommendation result for the new account.
[0040] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:
[0041] Based on the account attribute information of the new account and the account attribute information of the historical account, construct a heterogeneous information network of the account attribute information;
[0042] Acquire multiple groups of node wandering sequences of the heterogeneous information network, and train a semantic recognition model based on the multiple groups of node wandering sequences;
[0043] Obtaining similarity information between nodes in the heterogeneous information network through the semantic recognition model;
[0044] Classify each node in the heterogeneous information network according to the similarity information to obtain a target classification result of each node;
[0045] According to the target classification result, the new account is recommended to obtain a target recommendation result for the new account.
[0046] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:
[0047] Based on the account attribute information of the new account and the account attribute information of the historical account, construct a heterogeneous information network of the account attribute information;
[0048] Acquire multiple groups of node wandering sequences of the heterogeneous information network, and train a semantic recognition model based on the multiple groups of node wandering sequences;
[0049] Obtaining similarity information between nodes in the heterogeneous information network through the semantic recognition model;
[0050] Classify each node in the heterogeneous information network according to the similarity information to obtain a target classification result of each node;
[0051] According to the target classification result, the new account is recommended to obtain a target recommendation result for the new account.
[0052] The above-mentioned recommendation processing method, device, computer equipment, storage medium and computer program product construct a heterogeneous information network of account attribute information based on the account attribute information of the new account and the account attribute information of the historical account; obtain multiple groups of node walk sequences of the heterogeneous information network, and train a semantic recognition model based on the multiple groups of node walk sequences; obtain similarity information between nodes in the heterogeneous information network through the semantic recognition model; classify each node in the heterogeneous information network according to the similarity information to obtain the target classification result of each node; perform recommendation processing on the new account according to the target classification result to obtain the target recommendation result of the new account. By adopting this method, without obtaining the social information of the account, the account attribute information of the new account and the historical account is used to model the heterogeneous information network, and then the heterogeneous information network and the semantic recognition model are used to discover the group relationship between the new account and the historical account, so as to use the common information between the groups to recommend products to the new account, which greatly improves the cold start accuracy of recommending products to the new account. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] Figure 1 A schematic diagram of a process flow of a recommended processing method in an embodiment;
[0054] Figure 2 A flowchart of the steps of training a semantic recognition model in one embodiment;
[0055] Figure 3 A schematic diagram of the distribution of nodes before clustering in one embodiment;
[0056] Figure 4 A schematic diagram of the distribution of nodes after clustering in one embodiment;
[0057] Figure 5 is a schematic diagram of a heterogeneous information network and a random walk in one embodiment;
[0058] Figure 6 A schematic flow chart of a recommended processing method in another embodiment;
[0059] Figure 7 is a structural block diagram of a recommended processing device in one embodiment;
[0060] Figure 8 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION
[0061] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0062] It should be noted that the account information (including but not limited to account device information, account personal information, account attribute information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the account or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.
[0063] In one embodiment, Figure 1 As shown, a recommendation processing method is provided. This embodiment takes the method applied to a terminal as an example. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:
[0064] Step S101, constructing a heterogeneous information network of account attribute information based on the account attribute information of the new account and the account attribute information of the historical account.
[0065] Among them, new accounts refer to accounts registered on the platform for the first time, and historical accounts refer to accounts that have been registered in the past.
[0066] Among them, heterogeneous information network is an extended model of homogeneous networks with only the same attributes.
[0067] Specifically, after the account is authorized, the information of the new account can be collected in a legal and compliant manner in combination with expert experience or business rules, and then the account attribute information of the new account can be extracted from the account information of the new account. It is also possible to screen out several typical accounts from the database of existing accounts (i.e. historical accounts, old accounts) in combination with expert experience or business rules, and then perform the same processing on the account information of the typical account as the account information of the new account, so as to extract the account attribute information of the typical account from the account information of the typical account. Then, based on the account attribute information of the new account and the account attribute information of the typical account, a heterogeneous information network of account attribute information is constructed.
[0068] The account information of the new account and the account information of the typical account may be the information obtained when the account is registered or the basic information of the account. It should be noted that the account information of the new account and the account information of the typical account do not need to obtain the account social information.
[0069] Step S102, obtaining multiple groups of node wandering sequences of a heterogeneous information network, and training a semantic recognition model based on the multiple groups of node wandering sequences.
[0070] Among them, the node walk sequence refers to the walk path sequence obtained after random walks between nodes in a heterogeneous information network.
[0071] Among them, the semantic recognition model can be built based on deep learning algorithms (such as word2vec). Word2vec (word embedding) can convert discrete variables (nodes) into continuous vectors (multidimensional vectors). Since the idea of the language model is that the appearance of a word is affected by the context of the word, the basic idea of word2vec is also to use words to predict words (nearby words).
[0072] Specifically, the terminal can select each node in the heterogeneous information network in turn, and use the node as the initial node for random walk processing, and finally obtain multiple groups (set as k groups) of node walk sequences. The multiple groups of node walk sequences are used as the training corpus of the semantic recognition model. According to the semantic information of the node walk sequence, the initial semantic recognition model is used to train the multiple groups of node walk sequences respectively, and k semantic recognition models are obtained.
[0073] Step S103, obtaining similarity information between nodes in the heterogeneous information network through a semantic recognition model.
[0074] Among them, similarity information is used to describe the similarity between nodes in a heterogeneous information network.
[0075] Specifically, semantic recognition processing is performed on each node in the heterogeneous information network through multiple semantic recognition models, so as to analyze the similarity information between each node according to the semantic recognition results.
[0076] Step S104, classifying each node in the heterogeneous information network according to the similarity information to obtain a target classification result for each node.
[0077] Specifically, the initial classification results of each node are obtained by classifying the similarity information; then the outliers, isolated points and abnormal points in the initial classification results are reclassified to finally obtain the target classification results of each node.
[0078] Step S105: perform recommendation processing on the new account according to the target classification result to obtain the target recommendation result of the new account.
[0079] Specifically, the terminal recommends the purchase product information of the historical accounts belonging to the same classification group to the new account according to the target classification result, and finally obtains the target recommendation result for the new account.
[0080] In the above recommendation processing method, a heterogeneous information network of account attribute information is constructed based on the account attribute information of the new account and the account attribute information of the historical account; multiple groups of node walk sequences of the heterogeneous information network are obtained, and a semantic recognition model is trained based on the multiple groups of node walk sequences; similarity information between nodes in the heterogeneous information network is obtained through the semantic recognition model; each node in the heterogeneous information network is classified according to the similarity information to obtain the target classification result of each node; and the new account is recommended according to the target classification result to obtain the target recommendation result of the new account. By adopting this method, a heterogeneous information network is modeled through the account attribute information of the new account and the historical account without obtaining the social information of the account, and then the group relationship between the new account and the historical account is discovered by using the heterogeneous information network and the semantic recognition model, so as to use the common information between the groups to recommend products to the new account, which greatly improves the cold start accuracy of recommending products to the new account.
[0081] In one embodiment, Figure 2 As shown, the above step S102, obtaining multiple groups of node wandering sequences of a heterogeneous information network, and training a semantic recognition model based on the multiple groups of node wandering sequences, specifically includes the following contents:
[0082] Step S201, setting attribute conversion information and wandering path length information for the account attribute information of the nodes in the heterogeneous information network.
[0083] Specifically, the terminal can set the account attribute information for the nodes in the heterogeneous information network and set the walking rules of the random walk algorithm, which includes attribute conversion information and walking path length information. Among them, the attribute conversion information is used to describe the attribute order rules determined in the path walk. For example, for attributes A, B and C, A->B->C->A->B->C... and A->C->B->A->C->B... can all be used as attribute conversion information. The walking path length is used to describe the maximum length of each walking path. When the walking path length is reached, the walking process of this round is terminated.
[0084] Step S202, performing multiple random walks on the nodes in the heterogeneous information network according to the attribute conversion information and the walk path length information, and obtaining multiple groups of node walk sequences.
[0085] Multiple rounds of random walks are performed in a heterogeneous information network. Specifically, in each round of random walks, each node is selected as the initial node in turn in the heterogeneous information network, and random walks are performed according to attribute conversion information and walk path length information to obtain a node walk sequence for each round; after K rounds of walks, K groups of node walk sequences can be obtained.
[0086] Step S203, based on each group of node walk sequences, iteratively train the semantic recognition models to be trained to obtain multiple trained semantic recognition models.
[0087] Specifically, the terminal may use each group of node walk sequences to iteratively train the semantic recognition models to be trained, thereby training multiple semantic recognition models.
[0088] In this embodiment, multiple random walks are performed on nodes in a heterogeneous information network using attribute conversion information and walk path length information, and then multiple groups of node walk sequences obtained by the walks are used to train multiple trained semantic recognition models. In this way, subsequent steps can use the semantic recognition model to analyze the similarities between different nodes in the heterogeneous information network, thereby obtaining the similarities between different accounts, laying the foundation for product recommendations for new accounts and effectively improving the accuracy of product recommendation cold start.
[0089] In one embodiment, the above step S103 obtains similarity information between nodes in a heterogeneous information network through a semantic recognition model, which specifically includes the following contents: obtaining multiple similarity matrices of each node through multiple semantic recognition models; averaging multiple similarity matrices to obtain similarity information between nodes.
[0090] Specifically, the account attribute information of each node in the heterogeneous information network is input into each semantic recognition model respectively, so as to calculate the cosine distance between each node in the heterogeneous information network through each semantic recognition model, and then output multiple similarity matrices of each node; the terminal calculates the average value of multiple similarity matrices as the similarity information between each node.
[0091] In this embodiment, similarity information between nodes is obtained through multiple semantic recognition models, and the group association between nodes is analyzed using the similarity information to provide the information required for cold start for product recommendations for new accounts.
[0092] In one embodiment, the above step S104 classifies each node in the heterogeneous information network according to the similarity information to obtain the target classification result of each node, which specifically includes the following contents: according to the similarity information, unsupervised classification is performed on each node in the heterogeneous information network to obtain the initial classification result of each node; outliers in the initial classification result are re-classified to obtain the target classification result of each node.
[0093] Specifically, Figure 3 This is a schematic diagram of the distribution of nodes before clustering. Using an unsupervised clustering algorithm, the nodes corresponding to the similarity information are clustered to obtain the initial clustering results of each node. Figure 4is a schematic diagram of the distribution of nodes after clustering. Figure 4 As shown, the initial clustering result includes multiple classification groups, some of which contain nodes of new customers and nodes of existing customers, while some of which contain only nodes of new customers. Reclassify each node in the classification group that contains only nodes of new customers to ensure that each classification group in the final target classification result contains both nodes representing new accounts and nodes of historical accounts. For example, a similarity threshold can be set, and by judging the similarity threshold and similarity information, each node in the classification group that contains only nodes of new customers can be reclassified.
[0094] Among them, the unsupervised clustering algorithm can be a density-based clustering algorithm (Density-Based Spatial Clustering of Applications with Noise, DBSCAN).
[0095] In this embodiment, unsupervised classification processing is performed on each node in the heterogeneous information network based on similarity information to obtain an initial classification result of each node; the outliers in the initial classification result are also re-divided so that the target classification result of each node simultaneously includes nodes representing new accounts and nodes of historical accounts, so that when subsequent account recommendations are performed, product recommendations can be made for new accounts based on relevant information of historical accounts in the same classification grouping. Without the need to obtain account social information, the cold start accuracy of recommending products to new accounts is greatly improved.
[0096] In one embodiment, the target classification result includes multiple classification groups; each classification group includes nodes representing new accounts and nodes representing historical accounts. The above step S105, based on the target classification result, performs recommendation processing on the new account to obtain the target recommendation result of the new account, specifically includes the following content: based on the historical information corresponding to the historical account in each classification group, performs recommendation processing on the new account in each classification group to obtain the target recommendation result of the new account in each classification group.
[0097] Specifically, each classification group in the target classification result contains nodes representing new accounts and nodes representing historical accounts (or existing accounts). The terminal can determine the target recommendation results based on the historical information corresponding to the historical accounts in each classification group, and send the target recommendation results to each new account in the classification group. Without the need for the new account to authorize its social information, the cold start of recommendation processing for the new account is also completed.
[0098] In this embodiment, the association between the new account and the historical account is established through the heterogeneous information network and classification processing, and then the historical information of the historical accounts in the same classification group in the target classification result is used to perform recommendation processing for the new account in the same classification group. Without the need for the new account to authorize its social information, the cold start of the recommendation processing for the new account is also completed, and the common information of the groupware is fully utilized to improve the accuracy of the cold start of the recommendation processing for the new account. The attribute information and historical purchase product information of the historical account can also be used to improve the precision of the cold start recommendation processing for the new account, avoiding directly recommending popular products to the new account.
[0099] In one embodiment, the above step S101 constructs a heterogeneous information network of account attribute information based on the account attribute information of the new account and the account attribute information of the historical account, which specifically includes the following contents: performing feature extraction processing and data preprocessing on the account information of the new account to obtain the account attribute information of the new account; performing feature extraction processing and data preprocessing on the account information of the historical account to obtain the account attribute information of the historical account; and constructing a heterogeneous information network based on the account attribute information of the new account and the historical account.
[0100] Specifically, feature extraction is performed on the account information of the new account to obtain the initial attribute information of the new account, and then data preprocessing is performed on the initial attribute information of the new account, such as dirty data elimination processing, missing value processing, etc., to obtain the account attribute information of the new account. Similarly, feature extraction is performed on the account information of the historical account to obtain the initial attribute information of the historical account, and then data preprocessing is performed on the initial attribute information of the historical account, such as dirty data elimination processing, missing value processing, etc., to obtain the account attribute information of the historical account.
[0101] Furthermore, a heterogeneous information network is constructed based on the account attribute information of new accounts and historical accounts. Figure 5 The figure is a schematic diagram of a heterogeneous information network and random walk, where the solid and dotted lines represent the connection relationship of different attributes of each node, and the values on them represent the jump probability; the random walk algorithm is used to generate different node walk sequences in the heterogeneous network, and the right side is an example of 3 node walk sequences generated by random walk. Nodes can be defined for new accounts and historical accounts respectively, and the relationship between nodes can be described by edges. For example, purchase relationship: connecting historical account nodes and product nodes, indicating that the historical account has purchased a certain product; preference relationship: such as the preference of historical accounts for specific types of products; participation relationship: the relationship between historical accounts and activity nodes, indicating that the historical account has participated in a certain promotion activity; similarity relationship: the correlation and similarity between historical accounts and new accounts.
[0102] In this embodiment, the account attribute information of the new account and the historical account is obtained by performing feature extraction processing and data preprocessing on the account information of the new account and the historical account respectively; then, based on the account attribute information of the new account and the historical account, the effective construction of the heterogeneous information network is realized, so that the subsequent steps can analyze the relationship between different accounts based on the heterogeneous information network, thereby improving the accuracy of cold start recommendations for new accounts.
[0103] In one embodiment, Figure 6 As shown, another recommended processing method is provided, which is described by taking the method applied to a terminal as an example, and includes the following steps:
[0104] Step S601: construct a heterogeneous information network of account attribute information based on the account attribute information of the new account and the account attribute information of the historical account.
[0105] Step S602, setting attribute conversion information and wandering path length information for the account attribute information of the nodes in the heterogeneous information network.
[0106] Step S603, performing multiple random walks on the nodes in the heterogeneous information network according to the attribute conversion information and the walk path length information, and obtaining multiple groups of node walk sequences.
[0107] Step S604, based on each group of node walk sequences, iteratively train the semantic recognition models to be trained to obtain multiple trained semantic recognition models.
[0108] Step S605, obtaining similarity information between nodes in the heterogeneous information network through a semantic recognition model.
[0109] Step S606, classify each node in the heterogeneous information network according to the similarity information to obtain a target classification result of each node; the target classification result includes multiple classification groups; each classification group includes nodes representing new accounts and nodes of historical accounts.
[0110] Step S607, according to the historical information corresponding to the historical accounts in each classification group, the new accounts in each classification group are recommended to obtain the target recommendation results of the new accounts in each classification group.
[0111] The above recommendation processing method can achieve the following beneficial effects: without obtaining account social information, a heterogeneous information network is modeled through the account attribute information of new accounts and historical accounts, and then the group relationship between new accounts and historical accounts is discovered by using the heterogeneous information network and semantic recognition model, so as to use the common information between groups to recommend products to new accounts, greatly improving the cold start accuracy of recommending products to new accounts.
[0112] In order to more clearly illustrate the recommendation processing method provided by the embodiment of the present disclosure, the above recommendation processing method is specifically described below with a specific embodiment. Another recommendation processing method is provided, which can be applied to a terminal, and specifically includes the following contents:
[0113] (1) Collect new customer information based on expert experience / business rules, and then pre-process the data, mainly including dirty data and missing value processing, to extract the corresponding customer attribute information;
[0114] (2) Combine expert experience / business rules to select several typical customers from the existing customer database, then perform the same data preprocessing on these customer data and add them to the new customer data preprocessed in step (1) to build a heterogeneous network based on customer attributes (a heterogeneous network is an extended model of a homogeneous network with only the same attributes);
[0115] (3) Based on the correlation / similarity of the attributes between the nodes in the heterogeneous network, the random walk algorithm is used to generate the node walk sequence according to the predetermined attribute conversion and walk path length rules; the attribute conversion rule refers to the attribute order determined in the path walk, for example, for attributes A, B and C, A->B->C->A->B->C... and A->C->B->A->C->B... can both be used as attribute conversion rules; the walk path length rule refers to determining the maximum length of each walk path to end a walk process;
[0116] (4) Select each node in the heterogeneous network in turn, take the node as the initial node, repeat step (3), obtain multiple groups (set as k groups) of node walk sequences, and save the walk paths as the training corpus of word2vec;
[0117] (5) Based on the semantic information of the node sequence, use the deep learning algorithm (word2vec) to train the k groups of node walk sequences in step (4) respectively to obtain k word2vec models;
[0118] (6) According to the training results of step (4), for each word2vec model, the cosine distance between nodes is calculated and the similarity matrix of the nodes is constructed. Finally, k similarity matrices can be obtained;
[0119] (7) Calculate the average of the k similarity matrices in step (6) to obtain the final node similarity matrix;
[0120] (8) Using the unsupervised learning (DBSCAN) algorithm, cluster the similarity matrix in step (7) to obtain the pre-grouping result;
[0121] (9) Reprocessing the outliers / isolated points in the pre-grouping results in step (8) (e.g., setting a similarity threshold and grouping data based on business experience) to obtain the final grouping results;
[0122] (10) Based on the grouping results and the historical information of existing customers, recommend relevant items to new customers in the same group.
[0123] In this embodiment, without obtaining the customer's private social information, the correlation / similarity between different individuals is modeled based on the basic attribute information provided by the customer, and the common information between groups is fully utilized to improve the accuracy of product recommendations / cold starts. It is also possible to combine expert experience / business experience, use existing customer attributes and purchased product information, and improve the sophistication and accuracy of customer cold start recommendations to avoid directly pushing popular products.
[0124] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.
[0125] Based on the same inventive concept, the embodiment of the present application also provides a recommendation processing device for implementing the recommendation processing method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more recommendation processing device embodiments provided below can refer to the limitations on the recommendation processing method above, and will not be repeated here.
[0126] In one embodiment, Figure 7 As shown, a recommendation processing device 700 is provided, including: a grid construction module 701, a model training module 702, a similarity acquisition module 703, a node classification module 704 and an account recommendation module 705, wherein:
[0127] The grid construction module 701 is used to construct a heterogeneous information network of account attribute information based on the account attribute information of the new account and the account attribute information of the historical account.
[0128] The model training module 702 is used to obtain multiple groups of node wandering sequences of a heterogeneous information network, and train a semantic recognition model based on the multiple groups of node wandering sequences.
[0129] The similarity acquisition module 703 is used to obtain similarity information between nodes in the heterogeneous information network through a semantic recognition model.
[0130] The node classification module 704 is used to classify each node in the heterogeneous information network according to the similarity information to obtain the target classification result of each node.
[0131] The account recommendation module 705 is used to perform recommendation processing on the new account according to the target classification result to obtain the target recommendation result of the new account.
[0132] In one embodiment, the model training module 702 is also used to set attribute conversion information and wandering path length information for the account attribute information of the nodes in the heterogeneous information network; perform multiple random walks on the nodes in the heterogeneous information network according to the attribute conversion information and the wandering path length information to obtain multiple groups of node wandering sequences; based on each group of node wandering sequences, iteratively train the semantic recognition model to be trained to obtain multiple trained semantic recognition models.
[0133] In one embodiment, the similarity acquisition module 703 is further used to obtain multiple similarity matrices of each node through multiple semantic recognition models; and average the multiple similarity matrices to obtain similarity information between each node.
[0134] In one embodiment, the node classification module 704 is also used to classify each node in the heterogeneous information network according to similarity information to obtain an initial classification result for each node; and to reclassify the outliers in the initial classification result to obtain a target classification result for each node.
[0135] In one embodiment, the target classification result includes multiple classification groups; each classification group includes nodes representing new accounts and nodes of historical accounts. The account recommendation module 705 is also used to recommend new accounts in each classification group according to the historical information corresponding to the historical accounts in each classification group, and obtain the target recommendation result of the new accounts in each classification group.
[0136] In one embodiment, the grid construction module 701 is also used to perform feature extraction processing and data preprocessing on the account information of the new account to obtain the account attribute information of the new account; perform feature extraction processing and data preprocessing on the account information of the historical account to obtain the account attribute information of the historical account; and construct a heterogeneous information network based on the account attribute information of the new account and the historical account.
[0137] Each module in the above-mentioned recommendation processing device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module.
[0138] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 8 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a recommended processing method is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.
[0139] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0140] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.
[0141] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.
[0142] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.
[0143] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.
[0144] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0145] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. A recommendation processing method, characterized in that: The method comprises: Based on the account attribute information of the new account and the account attribute information of the historical account, construct a heterogeneous information network of the account attribute information; Acquire multiple groups of node wandering sequences of the heterogeneous information network, and train a semantic recognition model based on the multiple groups of node wandering sequences; Obtaining similarity information between nodes in the heterogeneous information network through the semantic recognition model; Classify each node in the heterogeneous information network according to the similarity information to obtain a target classification result of each node; According to the target classification result, the new account is recommended to obtain a target recommendation result for the new account.
2. The method according to claim 1, characterized in that The step of obtaining multiple groups of node walk sequences of the heterogeneous information network and training a semantic recognition model based on the multiple groups of node walk sequences includes: For the account attribute information of the nodes in the heterogeneous information network, setting attribute conversion information and wandering path length information; According to the attribute conversion information and the walking path length information, performing multiple random walk processes on the nodes in the heterogeneous information network to obtain the multiple groups of node walking sequences; Based on each group of the node walk sequences, the semantic recognition models to be trained are iteratively trained to obtain a plurality of trained semantic recognition models.
3. The method according to claim 2, characterized in that The obtaining of similarity information between nodes in the heterogeneous information network through the semantic recognition model includes: Obtaining multiple similarity matrices of each node through multiple semantic recognition models; The multiple similarity matrices are averaged to obtain similarity information between the nodes.
4. The method according to claim 1, characterized in that: The classifying process is performed on each node in the heterogeneous information network according to the similarity information to obtain a target classification result of each node, including: Classifying each node in the heterogeneous information network according to the similarity information to obtain an initial classification result of each node; The outliers in the initial classification result are reclassified to obtain the target classification result of each node.
5. The method according to claim 1, characterized in that: The target classification result includes a plurality of classification groups; each of the classification groups includes a node representing the new account and a node representing the historical account; The step of performing recommendation processing on the new account according to the target classification result to obtain a target recommendation result for the new account includes: According to the historical information corresponding to the historical accounts in each classification group, the new accounts in each classification group are recommended to obtain the target recommendation results of the new accounts in each classification group.
6. The method according to claim 1, characterized in that The step of constructing a heterogeneous information network of the account attribute information based on the account attribute information of the new account and the account attribute information of the historical account includes: Performing feature extraction and data preprocessing on the account information of the new account to obtain account attribute information of the new account; Performing feature extraction and data preprocessing on the account information of the historical account to obtain account attribute information of the historical account; The heterogeneous information network is constructed according to the account attribute information of the new account and the historical account.
7. A recommendation processing device, characterized in that: The device comprises: A grid construction module, used to construct a heterogeneous information network of the account attribute information based on the account attribute information of the new account and the account attribute information of the historical account; A model training module, used to obtain multiple groups of node wandering sequences of the heterogeneous information network, and train a semantic recognition model based on the multiple groups of node wandering sequences; A similarity acquisition module, used to obtain similarity information between nodes in the heterogeneous information network through the semantic recognition model; A node classification module, used to classify each node in the heterogeneous information network according to the similarity information to obtain a target classification result of each node; The account recommendation module is used to perform recommendation processing on the new account according to the target classification result to obtain the target recommendation result of the new account.
8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.
10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.