Webpage tag generation method and device, equipment, storage medium and program product

Abstract

The invention relates to the technical field of computers, and provides a webpage tag generation method and device, equipment, a storage medium and a program product. The method comprises the following steps: acquiring internet behavior data of a plurality of users; the internet behavior data comprises to-be-classified webpage links and access entity interaction information; based on the access entity interaction information of each user, user grouping is carried out through a clustering algorithm, and a user group is generated; determining a group tag of the user group; and determining a webpage label of the to-be-classified webpage link based on the group label. According to the mode, the users in the same user group generally have similar access behavior modes, that is, the users in the same user group are more likely to access webpages of the same type, and the group label can reflect the classification of the webpages accessed by the user group, so that the user group can access the webpages in the scene of webpage content deficiency, and the user group can access the webpages in the scene of webpage content deficiency. The webpage label of the to-be-classified webpage link can be determined according to the group label, accurate classification of the webpage and accurate generation of the webpage label are realized, and the application scene is wider.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a method, apparatus, device, storage medium, and program product for generating web page tags. Background Art

[0002] When classifying web pages, it is necessary to generate corresponding web page tags according to the web page classification results. For example, after completing the web page classification, generate a "education category" web page tag for the web page link of an education website.

[0003] In some related technologies, a web page classification method based on web page content analysis can be adopted, that is, extract information such as text content, title, and keywords in the web page to be classified, and use this information to classify the web page, so as to generate corresponding web page tags for the web page to be classified. This classification method generally relies on text mining technology and natural language processing technology, and is applicable to scenarios where the web page content can be clearly obtained.

[0004] However, in scenarios where the web page content is missing, incomplete, or inaccurate, since the specific content of the web page cannot be directly obtained for recognition, it is easy to cause inaccurate web page classification results and difficult to accurately generate web page tags. Summary of the Invention

[0005] Embodiments of this application provide a method, apparatus, device, storage medium, and program product for generating web page tags, which are used to solve the technical problem that in scenarios where the web page content is missing, incomplete, or inaccurate, the web page classification results in the prior art are inaccurate and it is difficult to accurately generate web page tags.

[0006] In a first aspect, an embodiment of this application provides a method for generating web page tags, including: obtaining the Internet access behavior data of multiple users; the Internet access behavior data includes the web page link to be classified and access entity interaction information, and the web page link to be classified is a web page link with missing web page content; based on the access entity interaction information of each user, perform user grouping through a clustering algorithm to generate user groups; determine the group tags of the user groups; based on the group tags, determine the web page tags of the web page link to be classified.

[0007] In one embodiment, the access entity interaction information includes the access entity name; based on the access entity interaction information of each user, perform user grouping through a clustering algorithm to generate user groups, including: input the access entity name of each user into a large language model respectively for semantic analysis to obtain the semantic embedding vector of each user output by the large language model; the semantic embedding vector is used to characterize the role and characteristics of the access entity; based on each semantic embedding vector, generate the user behavior vector of each user respectively; based on all user behavior vectors, perform user grouping through a clustering algorithm to generate user groups.

[0008] In one embodiment, a user has at least one semantic embedding vector, and a semantic embedding vector is generated based on an access entity name; wherein, a user behavior vector is obtained by performing a weighted average operation based on each semantic embedding vector of a user and the corresponding weighting coefficient of each semantic embedding vector; the weighting coefficient is determined based on the number of interactions and interaction duration between a user and an access entity.

[0009] In one embodiment, a user has at least one semantic embedding vector, and a semantic embedding vector is generated based on an access entity name; wherein, a user behavior vector is determined based on each semantic embedding vector of a user and the corresponding attention weight of each semantic embedding vector; the attention weight is generated by a feedforward neural network.

[0010] In one embodiment, determining the group label of a user group includes: determining the labels of the prominent access links and high-frequency access entities of the user group; the prominent access links are web page links whose target group index is greater than a first preset threshold, and the high-frequency access entities are access entities whose access frequency is greater than a second preset threshold; based on the labels of the prominent access links and high-frequency access entities, determining the group label of the user group and determining the group label weight of the group label.

[0011] In one embodiment, determining the web page label of a web page link to be classified based on the group label includes: determining whether the web page link to be classified belongs to the user group; if the web page link to be classified belongs to the user group, using the group label as the web page label of the web page link to be classified and using the group label weight as the web page label weight of the web page label.

[0012] In a second aspect, an embodiment of the present application provides a web page label generation device, including: an acquisition module for acquiring the Internet access behavior data of multiple users; the Internet access behavior data includes a web page link to be classified and access entity interaction information, and the web page link to be classified is a web page link with missing web page content; a user grouping module for grouping users based on the access entity interaction information of each user through a clustering algorithm to generate user groups; a group label determination module for determining the group label of the user group; a web page label determination module for determining the web page label of the web page link to be classified based on the group label.

[0013] In a third aspect, an embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and running on the processor, and when the processor executes the computer program, it implements any one of the above web page label generation methods.

[0014] Fourthly, an embodiment of the present application provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements any one of the above-mentioned web page label generation methods.

[0015] Fifthly, an embodiment of the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements any one of the above-mentioned web page label generation methods.

[0016] The web page label generation method, device, equipment, storage medium and program product provided by the embodiments of the present application first group users using a clustering algorithm according to the access entity interaction information of multiple users to generate user groups, determine the group labels of the user groups, and then determine the web page labels of the web page links to be classified according to the group labels. Through the above method, since users in the same user group usually have similar access behavior patterns, that is, users in the same user group are more likely to access the same type of web pages, and the group label can reflect the web page classification accessed by the user group. Therefore, in the scenario where the web page content is missing, the web page labels of the web page links to be classified can be determined according to the group labels, realizing the accurate classification of web pages and the precise generation of web page labels, and having a wider range of applicable scenarios. Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 is one of the flow diagrams of the web page label generation method provided by the embodiments of the present application.

[0019] Figure 2 is the second flow diagram of the web page label generation method provided by the embodiments of the present application.

[0020] Figure 3 is the structural diagram of the web page label generation device provided by the embodiments of the present application.

[0021] Figure 4 is the structural diagram of the electronic device provided by the embodiments of the present application. Detailed Embodiments

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the following will clearly and completely describe the technical solutions in this application with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are some, but not all, of the embodiments of this application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application without creative efforts shall fall within the scope of protection of this application.

[0023] Please refer to Figure 1 and Figure 2 , Figure 1 which is one of the flow diagrams of the web tag generation method provided by the embodiments of this application, Figure 2 and Figure 1 which is the second flow diagram of the web tag generation method provided by the embodiments of this application. As shown, in the embodiments of this application, the web tag generation method includes steps S110 to S140, and the specific steps are as follows: S110: Obtain the Internet behavior data of multiple users.

[0024] The Internet behavior data includes web page links to be classified and access entity interaction information, and the web page links to be classified are web page links with missing web page content.

[0025] Specifically, as Figure 2 shown, the Internet behavior data of multiple users (i.e., the user population in Figure 2 ) can be collected from operator data; the Internet behavior data of each user includes the web page URL (Uniform Resource Locator) data accessed by the user, as well as the interaction information of the user with different access entities.

[0026] Among them, the web page URL data accessed by the user includes web page links for which web page content can be obtained, as well as web page links to be classified with missing, incomplete, or inaccurate web page content.

[0027] Optionally, the access entities include official accounts, mini-programs, applications, service components, etc.

[0028] Optionally, the Internet behavior data can be collected in real time or periodically through the operator's log system, and relevant metadata can also be extracted, such as access timestamps, access frequencies, and other features.

[0029] ]>Optionally, after the collection of the Internet behavior data is completed, the Internet behavior data is subjected to data cleaning and normalization processing, such as removing duplicate or invalid web page URLs, eliminating abnormal data, formatting the access entity names, etc., to ensure the integrity and consistency of the data.

[0030] S120: Group users through a clustering algorithm based on the access entity interaction information of each user to generate user groups.

[0031] The purpose of clustering is to divide multiple users with similar access behavior patterns into the same user group, so as to enable the generation of labels and behavior analysis at the group level.

[0032] Specifically, according to the access entity interaction information of each user, group users through a clustering algorithm to generate at least one user group.

[0033] S130: Determine the group label of the user group.

[0034] Specifically, determine the group label of each user group and the group label weight corresponding to each group label.

[0035] S140: Based on the group label, determine the web page label of the web page link to be classified.

[0036] Specifically, if there is only one user group, determine whether the web page link to be classified belongs to this user group; if the web page link to be classified belongs to this user group, use the group label of this user group as the web page label of the web page link to be classified, and use the group label weight corresponding to this group label as the web page label weight of the web page label.

[0037] It can be understood that if there are multiple user groups, first determine which user group the web page link to be classified belongs to; after determining that the web page link to be classified belongs to a certain user group, then use the group label of this user group as the web page label of the web page link to be classified, and use the group label weight corresponding to this group label as the web page label weight of the web page label.

[0038] The web page label generation method provided by the embodiments of this application first groups users through a clustering algorithm according to the access entity interaction information of multiple users to generate user groups, determines the group labels of the user groups, and then determines the web page labels of the web page links to be classified according to the group labels. Through the above method, since users in the same user group usually have similar access behavior patterns, that is, users in the same user group are more likely to access the same type of web pages, and the group label can reflect the web page classification accessed by the user group, therefore, in the scenario where the web page content is missing, the web page label of the web page link to be classified can be determined according to the group label, realizing the accurate classification of the web page and the accurate generation of the web page label, and having a wider applicable scenario.

[0039] In some embodiments, accessing entity interaction information includes accessing entity names; based on the accessing entity interaction information of each user, user groups are generated through a clustering algorithm, including: inputting the accessing entity names of each user into a large language model respectively for semantic analysis to obtain the semantic embedding vectors of each user output by the large language model; the semantic embedding vectors are used to characterize the roles and features of the accessing entities; based on each semantic embedding vector, user behavior vectors of each user are generated respectively; based on all user behavior vectors, user groups are generated through a clustering algorithm.

[0040] After the data preprocessing is completed, for the accessing entity names in the accessing entity interaction information (such as official account names and mini-program names), a large language model (LLM) can be used for semantic analysis to generate high-dimensional semantic embedding vectors. These semantic embedding vectors can reflect the roles of the corresponding accessing entities and capture the implicit features in the accessing entity names through semantic similarity.

[0041] For example, for a mini-program containing words such as "education" or "tutoring", its semantic embedding vector will be closer to the education theme, and the large language model will automatically generate relevant feature representations through context training.

[0042] Optionally, the large language model is a pre-trained language model, such as a BERT model, a GPT model, etc., which can perform embedding representation on texts and generate high-dimensional vectors.

[0043] Specifically, since a user may interact with multiple accessing entities, the accessing entity interaction information of a user may contain multiple accessing entity names; for each user, all the accessing entity names of the user are input into the large language model respectively for semantic analysis, and the semantic embedding vectors corresponding to each accessing entity name output by the large language model can be obtained; among them, one semantic embedding vector is used to characterize the role and features of one accessing entity.

[0044] Therefore, the process of the large language model generating semantic embedding vectors can be represented by the following formula: ; where is the semantic embedding vector of the accessing entity name, is the encoder of the large language model, is the accessing entity name.

[0045] Through this process, the large language model can extract the semantic features of the accessing entities from the accessing entity names and generate high-dimensional semantic embedding vectors for further processing.

[0046] Further, for each user, a user behavior vector of the user is generated according to all semantic embedding vectors of the user.

[0047] Further, according to the user behavior vectors of all users, user grouping is performed through a clustering algorithm to generate at least one user group.

[0048] The purpose of clustering is to divide multiple users with similar access behavior patterns into the same user group, so that label generation and behavior analysis can be performed at the group level. This process not only depends on the access entity interaction information of users, but also needs to combine the analysis of significant access links and the labels of frequently accessed entities to calculate the corresponding group labels and group label weights for each user group.

[0049] Specifically, the user behavior vectors of all users are used as input data, and an unsupervised clustering algorithm (such as the K-means clustering algorithm, DBSCAN clustering algorithm, or hierarchical clustering algorithm, etc.) is used to group users to obtain at least one clustering result.

[0050] Among them, each clustering result represents a user group, and users within the same user group have similar access behavior patterns.

[0051] Optionally, for each user group, the behavioral similarity between all users within the user group is measured by calculating the similarity between the user behavior vectors of each user within the user group.

[0052] Optionally, the cosine similarity is used to measure the behavioral similarity between all users within the user group. The calculation formula of the cosine similarity is as follows: ; Among them, represents the cosine similarity between user and user , is the user behavior vector of user , is the user behavior vector of user .

[0053] Optionally, based on the similarity matrix or directly using the user behavior vectors of all users, K-means clustering is performed. The K-means clustering algorithm updates the cluster centers by minimizing the distance between each user behavior vector and its belonging cluster center until convergence, generating at least one user group.

[0054] Through this process, each user is assigned to a cluster, thus forming several user groups (i.e., sets of users with similar access behavior patterns).

[0055] The web page label generation method provided by the embodiments of the present application performs semantic analysis on the accessed entity name through a large language model to generate a high-dimensional semantic embedding vector, which can accurately capture the actual use implied in the accessed entity name, making up for the deficiency that the prior art cannot effectively understand the user's access intention; in addition, by clustering the user behavior vectors, the efficient extraction of group behavior patterns is realized, the processing complexity is reduced, and the classification efficiency is significantly improved, which can meet the needs of large-scale user data processing.

[0056] In some embodiments, a user has at least one semantic embedding vector, and one semantic embedding vector is generated based on one accessed entity name; wherein, one user behavior vector is obtained by performing a weighted average operation based on each semantic embedding vector of a user and the corresponding weighted coefficient of each semantic embedding vector; the weighted coefficient is determined based on the number of interactions and interaction duration between a user and an accessed entity.

[0057] It can be understood that since a user may interact with multiple accessed entities, the accessed entity interaction information of a user may include multiple accessed entity names; for each user, all the accessed entity names of the user are respectively input into the large language model for semantic analysis, and the semantic embedding vectors corresponding to each accessed entity name output by the large language model can be obtained, and one semantic embedding vector is generated based on one accessed entity name.

[0058] Furthermore, for each user, the user behavior vector of the user can be further generated by analyzing the interaction behavior of the user with each accessed entity (such as the number of interactions and interaction duration, etc.).

[0059] Assume that a user has interaction information with multiple accessed entities, and the interaction information includes the number of interactions and interaction duration. The number of interactions between the user and the th accessed entity is denoted as , and the interaction duration between the user and the th accessed entity is denoted as . Then, the user behavior vector of the user can be calculated, and the user behavior vector characterizes the behavior pattern of the user on each accessed entity. Among them, the expression of the user behavior vector is as follows: ; Among them, represents the user behavior vector of the user, is the weighted coefficient corresponding to the semantic embedding vector of the th accessed entity, is the semantic embedding vector of the th accessed entity of the user, is the number of all accessed entities that the user has interacted with.

[0060] Among them, the weighting coefficient can be based on the number of interactions and the interaction duration to determine: .

[0061] As can be seen from the above formula, the more interactions a user has with a certain accessed entity and the longer the interaction duration, the greater the contribution of the user to the semantic embedding vector of the accessed entity. Through such an average weighting method, the significant interaction features in the user behavior will be more reflected in the final user behavior vector.

[0062] In some embodiments, a user has at least one semantic embedding vector, and a semantic embedding vector is generated based on an accessed entity name; among them, a user behavior vector is determined based on each semantic embedding vector of a user and the attention weight corresponding to each semantic embedding vector; the attention weight is generated by a feed-forward neural network.

[0063] It can be understood that the weighted average method directly calculates the contribution weight of each semantic embedding vector based on the interaction information of the user with each accessed entity (such as the number of interactions and the interaction duration). Although this method is relatively simple and easy to implement, it may not fully consider the complex patterns in the user behavior.

[0064] In order to better capture the interaction patterns of the user on different accessed entities, an attention mechanism can be introduced to adaptively assign corresponding attention weights to the semantic embedding vectors of each accessed entity. At this time, the expression of the user behavior vector is as follows: ; Among them, represents the user behavior vector of the user, is the semantic embedding vector of the th accessed entity of the user, is the number of all accessed entities that the user has interacted with, is the attention weight corresponding to the semantic embedding vector of the th accessed entity, represents the normalization processing of the attention weights of all semantic embedding vectors so that their sum is 1.

[0065] The calculation of the attention weight can be implemented through a feed-forward neural network. The input is the semantic embedding vector of the accessed entity and the overall interaction behavior characteristics of the user, and the output is the attention weight corresponding to each semantic embedding vector. The attention mechanism can dynamically adjust the importance of each accessed entity in the user behavior according to the context, so as to further improve the expression ability of the final user behavior vector when the user behavior has complex or irregular patterns.

[0066] It should be noted that the weighted average method or the attention mechanism is two different methods for generating user behavior vectors, but the processes of generating user behavior vectors by both can be uniformly represented by the following formula: ; Among them, represents the fusion process of the weighted average method or the attention mechanism, and finally generates a user representation (i.e., user behavior vector) that integrates various behavior features.

[0067] The web page tag generation method provided in the embodiments of the present application finally generates a high-dimensional user behavior vector through the weighted average method or the attention mechanism. This vector synthesizes the interaction information of the user with each accessed entity and can effectively reflect the user's interests and intentions. This vector can be further used for tasks such as group clustering and tag generation to support subsequent web page classification and recommendation work.

[0068] In some embodiments, determining the group tags of a user group includes: determining the significant access links of the user group and the tags of frequently accessed entities; the significant access links are web page links whose target group index is greater than a first preset threshold, and the frequently accessed entities are accessed entities whose access frequency is greater than a second preset threshold; based on the tags of the significant access links and the frequently accessed entities, determine the group tags of the user group and determine the group tag weights of the group tags.

[0069] Once users are clustered into multiple user groups, the corresponding group tags can be calculated for each user group. The generation of group tags depends on two aspects: the significant access links of the user group and the tags of the frequently accessed entities within the user group, that is, the tags of the frequently accessed entities.

[0070] In the embodiments of the present application, the significant access link is a web page link whose target group index (Target Group Index, TGI) is greater than a first preset threshold. TGI is an index that measures the access significance of a certain user group to a specific web page link. The higher the TGI, the more significant the access of the user group to the web page link. By counting the web page links accessed by all users in the user group and their access frequencies, the TGI of each web page link within the user group can be calculated.

[0071] Among them, the target group index of the web page link The calculation formula is: ; It should be noted that since a web page link uniquely corresponds to a web page URL, the URL is used to represent the web page link in the above formula.

[0072] Among user groups, web links with high TGI values are considered significant access links for that user group. The classification tags corresponding to the significant access links can be used as the group tags for the corresponding user group because these web links reflect the preferences of the user group for certain specific content or services. For example, if a certain user group frequently accesses educational websites, its significant access links may include multiple educational web page URLs, indicating that the user group may be interested in educational content.

[0073] In addition to significant access links, the tags of frequently accessed entities also play an important role in generating user group tags. Therefore, by counting the names of the access entities with high access frequencies within a user group and combining their semantic tags, group tags can be generated. Specifically, for the access entities frequently accessed within a user group, their classification tags are used as the group tags for that user group, and different group tag weights are assigned to each group tag according to the access frequencies within the user group. The calculation formula for the group tag weight is: ; where, is the group tag weight of the th user group, is the contribution of the th user in the th user group to the group tag (such as the number of interactions and interaction duration), is the total number of users in the th user group, is the number of users in the th user group who have accessed this frequently accessed entity.

[0074] Group tags are generated based on the strength of the internal connections within the user group. For example, if most users within a user group frequently access a certain educational access entity, that user group will be assigned an "educational" group tag, and the group tag weight of the group tag reflects the degree of preference of that user group for educational content.

[0075] The group tags for each user group can be calculated based on the tags of significant access links and frequently accessed entities. Therefore, after completing group clustering, the tags of significant access links and frequently accessed entities can be separately counted for each user group to generate group tags.

[0076] Optionally, the group tags can be a weighted combination of these significant access links and frequently accessed entities, and weights can be assigned to each group tag according to information such as access frequencies. The tags of significant access links and access entities with higher access frequencies will be assigned higher weights, indicating that these tags better represent the characteristics of the user group.

[0077] It should be noted that a user group may have multiple group tags, and all group tags can form a tag library.

[0078] For example, if a user group frequently accesses educational websites, its prominent access links may include multiple URLs of educational web pages, so the user group can be given the group tag of "educational"; at the same time, since the tag of the frequently accessed entity of this user group is "technology", the user group can be given the group tag of "technology"; at this time, the user group has two group tags, and these two group tags constitute the tag library of this user group. Since the user group has two group tags, therefore, the group tag weights of the two group tags can be determined respectively according to information such as the interaction times and interaction duration between the user group's access to educational web page URLs and technology access entities. For example, the group tag weight of the "educational" group tag is 80%, and the group tag weight of the "technology" group tag is 20%.

[0079] It can be understood that as the online behavior data of each user in the user group is continuously updated, the group tags can be dynamically adjusted. For example, when the users in a user group increase the access frequency to a certain type of web page link, the group tags of the user group may change, and the tag library will be updated accordingly.

[0080] In some embodiments, based on the group tags, determining the web page tags of the web page link to be classified includes: determining whether the web page link to be classified belongs to a user group; if the web page link to be classified belongs to a user group, then using the group tag as the web page tag of the web page link to be classified, and using the group tag weight as the web page tag weight of the web page tag.

[0081] After generating group tags for each user group, it is necessary to determine the web page tags of the web page link to be classified based on the group tags of each user group.

[0082] Specifically, if there is only one user group, determine whether the web page link to be classified belongs to this user group; if the web page link to be classified belongs to this user group, then use the group tag of this user group as the web page tag of the web page link to be classified, and use the group tag weight corresponding to this group tag as the web page tag weight of the web page tag.

[0083] It can be understood that if there are multiple user groups, first determine which user group the web page link to be classified belongs to; after determining that the web page link to be classified belongs to a certain user group, then use the group tag of this user group as the web page tag of the web page link to be classified, and use the group tag weight corresponding to this group tag as the web page tag weight of the web page tag.

[0084] Understandably, since a user group may have multiple group tags, there will also be multiple web page tags for the web page links to be classified. The weight of each web page tag is the same as the weight of its corresponding group tag. At this time, the weight of each web page tag can be expressed by the following formula: .

[0085] As can be seen from the above process, the web page tags of the web page links to be classified depend on the group tags of the user groups they belong to, and the weights of the web page tags are consistent with the weights of the group tags. In this way, according to the tag information of the user groups, corresponding tags can be assigned to each web page link to be classified that the user accesses, so as to achieve accurate web page classification and intent analysis.

[0086] Over time, the online behavior patterns and access preferences of each user in each user group may change. Therefore, the tag library needs to be continuously updated. Each recalculation of the group tags may form new web page tags or update the existing web page tags, which can ensure that the tag library can reflect the changes in user behavior in a timely manner.

[0087] The web page tag generation method provided by the embodiments of this application dynamically generates and updates web page tags based on user group behavior, combined with the tags of significant access links and frequently accessed entities, making the web page tags more real-time and accurate, and overcoming the limitations of the static tag library and difficulty in adapting to changes in the prior art.

[0088] The embodiments of this application also provide a web page tag generation device. Please refer to Figure 3 , Figure 3 which is a schematic structural diagram of the web page tag generation device provided by the embodiments of this application. In the embodiments of this application, the web page tag generation device includes an acquisition module 310, a user grouping module 320, a group tag determination module 330, and a web page tag determination module 340.

[0089] The acquisition module 310 is used to acquire the online behavior data of multiple users.

[0090] The online behavior data includes web page links to be classified and access entity interaction information, and the web page links to be classified are web page links with missing web page content.

[0091] The user grouping module 320 is used to group users based on the access entity interaction information of each user through a clustering algorithm to generate user groups.

[0092] The group tag determination module 330 is used to determine the group tags of the user groups.

[0093] A web page label determination module 340 is configured to determine the web page label of a web page link to be classified based on the group label.

[0094] In some embodiments, the access entity interaction information includes the access entity name.

[0095] A user grouping module 320 is configured to input the access entity name of each user into a large language model respectively for semantic analysis to obtain the semantic embedding vector of each user output by the large language model; the semantic embedding vector is used to characterize the role and features of the access entity; based on each semantic embedding vector, generate the user behavior vector of each user respectively; based on all user behavior vectors, perform user grouping through a clustering algorithm to generate user groups.

[0096] In some embodiments, a user has at least one semantic embedding vector, and one semantic embedding vector is generated based on one access entity name; wherein, one user behavior vector is obtained by performing a weighted average operation based on each semantic embedding vector of a user and the weighted coefficient corresponding to each semantic embedding vector; the weighted coefficient is determined based on the interaction times and interaction duration between a user and the access entity.

[0097] In some embodiments, a user has at least one semantic embedding vector, and one semantic embedding vector is generated based on one access entity name; wherein, one user behavior vector is determined based on each semantic embedding vector of a user and the attention weight corresponding to each semantic embedding vector; the attention weight is generated by a feedforward neural network.

[0098] In some embodiments, a group label determination module 330 is configured to determine the labels of the significant access links and high-frequency access entities of the user group; the significant access link is a web page link whose target group index is greater than a first preset threshold, and the high-frequency access entity is an access entity whose access frequency is greater than a second preset threshold; based on the labels of the significant access links and high-frequency access entities, determine the group label of the user group, and determine the group label weight of the group label.

[0099] In some embodiments, a web page label determination module 340 is configured to determine whether the web page link to be classified belongs to the user group; if the web page link to be classified belongs to the user group, then use the group label as the web page label of the web page link to be classified, and use the group label weight as the web page label weight of the web page label.

[0100] An embodiment of the present application further provides an electronic device, Figure 4 is a schematic structural diagram of the electronic device provided by the embodiment of the present application, as Figure 4As shown, the electronic device may include: a processor 410, a communications interface 420, a memory 430, and a communication bus 440. Among them, the processor 410, the communications interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 may call the logical instructions in the memory 430 to execute the web page tag generation method.

[0101] In addition, when the logical instructions in the above-mentioned memory 430 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0102] The embodiment of the present application also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the web page tag generation method provided by the above-mentioned various methods.

[0103] The embodiment of the present application also provides a computer program product. The computer program product includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the web page tag generation method provided by the above-mentioned various methods.

[0104] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative labor.

[0105] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating web page tags, characterized in that, Including: Obtain the Internet usage behavior data of multiple users; The Internet usage behavior data includes web page links to be classified and access entity interaction information, and the web page links to be classified are web page links with missing web page content; Based on the access entity interaction information of each user, perform user grouping through a clustering algorithm to generate user groups; Determine the group label of the user group; Based on the group label, determine the web page label of the web page link to be classified.

2. The web page label generation method according to claim 1, characterized in that, The access entity interaction information includes the access entity name; The performing user grouping through a clustering algorithm based on the access entity interaction information of each user to generate user groups includes: Input the access entity name of each user into a large language model respectively for semantic analysis to obtain the semantic embedding vector of each user output by the large language model; the semantic embedding vector is used to characterize the role and features of the access entity; Based on each semantic embedding vector, generate the user behavior vector of each user respectively; Based on all the user behavior vectors, perform user grouping through the clustering algorithm to generate the user groups.

3. The web page label generation method according to claim 2, characterized in that One user has at least one semantic embedding vector, and one semantic embedding vector is generated based on one access entity name; Wherein, one user behavior vector is obtained by performing weighted average operation based on each semantic embedding vector of one user and the corresponding weighting coefficient of each semantic embedding vector; the weighting coefficient is determined based on the interaction times and interaction duration between one user and the access entity.

4. The method for generating web page tags according to claim 2, wherein One user has at least one semantic embedding vector, and one semantic embedding vector is generated based on one access entity name; Wherein, one user behavior vector is determined based on each semantic embedding vector of one user and the corresponding attention weight; the attention weight is generated by a feedforward neural network.

5. The method for generating web page tags according to claim 1, wherein The determining the group label of the user group includes: Determine the labels of the significant access links and high-frequency access entities of the user group; the significant access links are web page links with a target group index greater than a first preset threshold, and the high-frequency access entities are access entities with an access frequency greater than a second preset threshold; Based on the labels of the significant access links and the high-frequency access entities, determine the group label of the user group and determine the group label weight of the group label.

6. The method for generating web page tags according to claim 5, wherein, The determining the web page label of the web page link to be classified based on the group label includes: Judge whether the web page link to be classified belongs to the user group; If the web page link to be classified belongs to the user group, use the group label as the web page label of the web page link to be classified and use the group label weight as the web page label weight of the web page label.

7. A web page label generating device, characterized in that, Including: An acquisition module for acquiring the Internet usage behavior data of multiple users; The Internet usage behavior data includes web page links to be classified and access entity interaction information, and the web page links to be classified are web page links with missing web page content; A user grouping module, configured to group users through a clustering algorithm based on the access entity interaction information of each user to generate user groups; A group label determination module, configured to determine the group label of the user group; A web page label determination module, configured to determine the web page label of the to-be-classified web page link based on the group label.

8. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the web page label generation method according to any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the web page label generation method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the web page label generation method according to any one of claims 1 to 6.

Images