Information recommendation method, device, storage medium, and program product

Abstract

Embodiments of the present application provide an information recommendation method and device, a storage medium and a program product. The method comprises: obtaining operation information of a user, and determining a target user type to which the user belongs from a plurality of user types based on the operation information; determining a target recommendation strategy matched with the target user type from a plurality of recommendation strategies for determining to-be-recommended information according to a preset correspondence relationship; determining first to-be-recommended information and a corresponding first matching degree according to the target recommendation strategy and the operation information; obtaining skill information of the user; determining second to-be-recommended information and a corresponding second matching degree according to a similarity between the skill information and a skill information tag corresponding to a preset post cluster; and determining recommended information according to the first to-be-recommended information, the first matching degree, the second to-be-recommended information and the second matching degree. The embodiments of the present application can achieve accurate information recommendation.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, and in particular relates to an information recommendation method, device, storage medium and program product. Background Technology

[0002] With the popularization of internet recruitment technology, recruitment platforms have formed an intelligent matching system based on user profiles and job descriptions. Currently, the service architecture and algorithm model of recruitment platforms are mainly built around the job-seeking characteristics of white-collar workers. They generally achieve two-way screening and recommendation between resumes and job postings by collecting structured text data such as users' educational background, work experience, and skill tags.

[0003] However, the employment patterns of blue-collar workers are characterized by strong geographical dependence and non-standardized skill descriptions, which are fundamentally different from those of white-collar workers. Therefore, how to accurately deliver job information to blue-collar workers is a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0004] This application provides an information recommendation method, device, storage medium, and program product that can achieve accurate information recommendation.

[0005] In a first aspect, embodiments of this application provide an information recommendation method, comprising: acquiring user operation information, and determining a target user type to which the user belongs from multiple user types based on the operation information; determining a target recommendation strategy that matches the target user type from multiple recommendation strategies used to determine information to be recommended according to a preset correspondence, wherein the preset correspondence includes the correspondence between different user types and recommendation strategies; determining a first piece of information to be recommended and a corresponding first matching degree based on the target recommendation strategy and the operation information; acquiring user skill information; determining a second piece of information to be recommended and a corresponding second matching degree based on the similarity between the skill information and skill information tags corresponding to a preset job cluster; and determining recommended information based on the first piece of information to be recommended, the first matching degree, the second piece of information to be recommended, and the second matching degree.

[0006] In one embodiment, both the first and second recommended information include location information and value matching information. Determining recommended information based on the first recommended information, a first matching degree, the second recommended information, and a second matching degree includes: calculating spatial distance constraints corresponding to the value matching information in the first and second recommended information according to a preset value distance regression equation; calculating spatial distance based on the location information and user baseline location information in the first and second recommended information; filtering the first and second recommended information based on the spatial distance constraints and spatial distance to obtain filtered first and second recommended information; calculating a third matching degree based on the spatial distance and value matching information in the filtered first and second recommended information; and determining recommended information from the filtered first and second recommended information based on the first matching degree, the second matching degree, and the third matching degree.

[0007] In one implementation, the preset value distance regression equation is shown in formula (1): Where x represents value matching information; y represents spatial distance; , , For coefficients, This is the error term.

[0008] In one implementation, the target user type includes interactive users: Based on the target recommendation strategy and operation information, determining the first information to be recommended and its corresponding first matching degree includes: acquiring user basic features and job features; constructing a bipartite graph consisting of multiple nodes based on the user features and job features, wherein the nodes include user nodes representing user basic features and job nodes representing job features, and the edges connecting the nodes are constructed based on operation information; mapping the basic features corresponding to user nodes and the job features corresponding to job nodes respectively to obtain the basic vector corresponding to user nodes and the job vector corresponding to job nodes; for each node, determining the neighboring nodes in the bipartite graph and acquiring the neighboring node vectors corresponding to the neighboring nodes; for each node, weighted aggregation of the neighboring node vectors based on the edge weights of the edges between the node and each neighboring node to obtain a neighbor aggregation vector, and fusing the neighbor aggregation vector with the basic vector or job vector corresponding to the node to obtain an aggregated basic vector or aggregated job vector; and determining the first information to be recommended and its corresponding first matching degree based on the aggregated basic vector and aggregated job vector.

[0009] In one implementation, the target user type further includes a cold start type; before constructing a bipartite graph consisting of multiple nodes based on user characteristics and job characteristics, the method further includes: obtaining basic attribute information of the user and skill information of the user; determining user enhancement features based on the basic attribute information and skill information; constructing a bipartite graph consisting of multiple nodes based on user characteristics and job characteristics, including: constructing a bipartite graph consisting of multiple nodes based on user enhancement features and job characteristics.

[0010] In one embodiment, before determining the second recommended information and the corresponding second matching degree based on the similarity between the skill information and the skill information tags corresponding to the preset job clusters, the method further includes: obtaining job information corresponding to multiple jobs; for each job, extracting keywords from the job information according to a preset keyword frequency calculation formula to obtain a keyword feature vector corresponding to each job; clustering the multiple keyword feature vectors to obtain multiple job clusters; for each job cluster, obtaining the target job information of the job cluster, and determining the skill information tags corresponding to the job cluster based on the skill information in the target job information and preset filtering conditions.

[0011] In one implementation, multiple keyword feature vectors are clustered to obtain multiple job clusters, including: calculating a first distance value between keyword feature vectors; merging keyword feature vectors whose first distance values ​​satisfy a preset distance condition to obtain a merged initial job cluster; determining whether the initial job clusters meet a stopping condition; if the initial job clusters do not meet the stopping condition, for each initial job cluster, calculating the average vector of all keyword feature vectors within that initial job cluster, and calculating a second distance value between multiple average vectors; merging keyword feature vectors whose second distance values ​​satisfy a preset distance condition, updating the initial job clusters, and returning to the step of determining whether the initial job clusters meet the stopping condition, until the stopping condition is met, resulting in multiple job clusters.

[0012] Secondly, embodiments of this application provide an information recommendation device, the device comprising: The first acquisition module is used to acquire the user's operation information and determine the target user type to which the user belongs from multiple user types based on the operation information; The first determining module is used to determine a target recommendation strategy that matches the target user type from multiple recommendation strategies used to determine the information to be recommended, based on a preset correspondence relationship. The preset correspondence relationship includes the correspondence relationship between different user types and recommendation strategies. The second determining module is used to determine the first information to be recommended and the corresponding first matching degree based on the target recommendation strategy and operation information; The second acquisition module acquires the user's skill information; The third determination module is used to determine the second recommended information and the corresponding second matching degree based on the similarity between the skill information and the skill information tags corresponding to the preset job clusters; The fourth determining module is used to determine the recommended information based on the first information to be recommended, the first matching degree, the second information to be recommended, and the second matching degree.

[0013] Thirdly, embodiments of this application provide an information recommendation device, the device including: a processor and a memory storing computer program instructions; the processor executes the computer program instructions to implement the information recommendation method of the first aspect or any embodiment of the first aspect.

[0014] Fourthly, a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the information recommendation method of the first aspect or any embodiment of the first aspect.

[0015] Fifthly, embodiments of this application provide a computer program product in which instructions, when executed by a processor of an electronic device, cause the electronic device to perform an information recommendation method as described in the first aspect or any embodiment of the first aspect.

[0016] The information recommendation method, device, and computer storage medium of this application embodiment can acquire operation information and classify users into different user types accordingly. Then, it selects corresponding recommendation strategies for different user types to generate first recommended information, effectively solving the problems of recommendation systems failing to start or low-quality recommendation information in scenarios with new users or sparse data. This ensures that there is an appropriate recommendation information generation path under different data completeness levels. Simultaneously, by pre-setting job clusters and standardized skill information tags, the non-standardized, colloquial skill descriptions that are difficult to uniformly measure among blue-collar workers are converted into similarity calculations with each job cluster tag set, generating second recommended information and matching degree. This objectively and quantitatively evaluates the non-standardized skills and practical experience of blue-collar workers, reducing the limitation of relying solely on keyword string matching to identify implicit skills. Furthermore, by integrating behavioral recommendation results based on user type strategies with structured recommendation results based on skill tag similarity, the final recommended jobs not only respond to existing behavioral preferences but also uncover potential jobs matching the skill structure of cold-start users or users with sparse behavior, improving the accuracy, coverage, and multi-scenario adaptability of job recommendations for blue-collar workers. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating an embodiment of the information recommendation method provided in this application is shown; Figure 2 A schematic diagram of clustering provided in one embodiment of this application is shown; Figure 3 A flowchart illustrating an embodiment of the information recommendation method provided in this application is shown; Figure 4 A flowchart illustrating an embodiment of the information recommendation method provided in this application is shown; Figure 5 This is a schematic diagram of the structure of an information recommendation device provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an information recommendation device provided in an embodiment of this application. Detailed Implementation

[0019] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0020] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0021] It should be noted that the acquisition, storage, use, and processing of data in this application embodiment all comply with the relevant provisions of national laws and regulations.

[0022] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0023] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0024] With the popularization of internet recruitment technology, recruitment platforms have formed an intelligent matching system based on user profiles and job descriptions. Currently, the service architecture and algorithm models of recruitment platforms are mainly built around the job-seeking characteristics of white-collar workers. They typically achieve two-way screening and recommendation between resumes and job postings by collecting structured text data such as users' educational background, work experience, and skill tags. For example, using data such as skill tags and years of work experience filled in by users during platform registration as core data, a fuzzy string matching algorithm can be used to compare job requirement keywords (such as welder's certificate, electrician's certificate, etc.) with user information to achieve basic skills matching. It is understandable that the core logic for recommending jobs to white-collar workers is mainly based on collaborative filtering (i.e., recommending based on the behavior of similar users) or content-based text matching (i.e., matching resume keywords with job descriptions).

[0025] However, such job recommendations fail to meet the needs of blue-collar workers for geographical proximity, similar job types, and immediate employment. Specifically, the employment patterns of blue-collar workers are characterized by strong geographical dependence and non-standardized skill descriptions. Therefore, the job recommendation process only uses geographical location as a general screening condition, rather than a core calculation factor, which fails to meet the blue-collar workers' need for "immediate employment." Furthermore, relying solely on keyword matching to match skills with blue-collar workers lacks a dedicated skills verification and assessment technology system. Blue-collar jobs not only require hard skills but also rely on core qualities such as practical proficiency. These "non-standardized skills" are difficult for blue-collar workers to describe, and related technologies struggle to effectively assess them.

[0026] Therefore, how to accurately deliver job information to blue-collar workers is a technical problem that urgently needs to be solved in this field.

[0027] To address the problems of the prior art, embodiments of this application provide an information recommendation method, apparatus, storage medium, and program product. The information recommendation method provided in this application embodiment will be described first below.

[0028] Figure 1 A flowchart illustrating an embodiment of the information recommendation method provided in this application is shown. Figure 1 As shown, the method may include the following steps: S110. Obtain the user's operation information and determine the target user type from multiple user types based on the operation information.

[0029] S120. Based on the preset correspondence, determine the target recommendation strategy that matches the target user type from multiple recommendation strategies used to determine the information to be recommended.

[0030] S130. Based on the recommendation strategy and operation information, determine the first piece of information to be recommended and its corresponding first matching degree.

[0031] S140. Obtain the user's skill information.

[0032] S150. Based on the similarity between the skill information and the skill information tags corresponding to the preset job clusters, determine the second information to be recommended and the corresponding second matching degree.

[0033] S160. Determine the recommended information based on the first information to be recommended, the first matching degree, the second information to be recommended, and the second matching degree.

[0034] Figure 1 In the illustrated embodiment, operational information is acquired and users are categorized into different user types. Corresponding recommendation strategies are then selected for each user type to generate the first set of recommended information. This effectively solves the problem of recommendation systems failing to start or providing low-quality recommendations in scenarios with new users or sparse data, ensuring that there is a suitable recommendation information generation path under different levels of data completeness. Simultaneously, by pre-setting job clusters and standardized skill information tags, the non-standardized, colloquial skill descriptions that are difficult to uniformly measure among blue-collar workers are converted into similarity calculations with each job cluster tag set, generating the second set of recommended information and matching degree. This objectively and quantitatively assesses the "non-standardized skills" and practical experience of blue-collar workers, reducing the limitation of relying solely on keyword string matching to identify implicit skills. Furthermore, by integrating behavioral recommendation results based on user type strategies with structured recommendation results based on skill tag similarity, the final recommended jobs not only respond to existing behavioral preferences but also uncover potential jobs matching the skill structure of cold-start users or users with sparse behavior, improving the accuracy, coverage, and multi-scenario adaptability of job recommendations for blue-collar workers.

[0035] In some embodiments, in S110, user operation information can be obtained, and the target user type to which the user belongs can be determined from multiple user types based on the operation information.

[0036] For example, operational information can be behavioral data generated by users on the platform, such as browsing history, search keywords, clicks, favorites, applications, etc., which can reflect the user's intentions and preferences.

[0037] For example, user types can be categories based on the sparsity or richness of action information. For instance, users with no or very little action information (such as newly registered users) can be classified as cold-start users, while users who have accumulated sufficient action information can be classified as interactive users. By analyzing action information, the target user type to which a user currently belongs can be determined.

[0038] In some embodiments, in S120, a target recommendation strategy that matches the target user type is determined from a plurality of recommendation strategies for determining the information to be recommended, based on a preset correspondence.

[0039] The preset correspondence includes the correspondence between different user types and recommendation strategies.

[0040] The preset mapping relationship is the correspondence between different user types and different recommendation strategies. This mapping relationship can be determined by technical personnel based on the characteristics corresponding to different user types.

[0041] Recommendation strategies are different calculation rules or logics preset for generating information to be recommended. For example, the recommendation strategy configured for cold-start users can be different from the recommendation strategy configured for interactive users.

[0042] For example, based on the target user type stated by the user, a target recommendation strategy that matches the target user type is determined from multiple recommendation strategies used to determine the information to be recommended.

[0043] In some embodiments, in S130, the first information to be recommended and the corresponding first matching degree can be determined based on the target recommendation strategy and operation information.

[0044] The first piece of information to be recommended can be candidate recommendation information generated by applying a target recommendation strategy and utilizing operational information. For example, the first piece of information to be recommended can be a set of job postings.

[0045] The first matching score is used to quantify the degree of matching between the first piece of information to be recommended and the user.

[0046] In some alternative embodiments, the target user type includes interactive users. Deep-seated, implicit career preferences of users can be identified by constructing a heterogeneous user-job graph network.

[0047] A bipartite graph consisting of multiple nodes can be constructed. Nodes include user nodes representing basic user characteristics and job node nodes representing job characteristics, with edges connecting the nodes constructed based on operational information. The basic characteristics corresponding to user nodes and the job characteristics corresponding to job node nodes are mapped to obtain the basic vector for user nodes and the job vector for job node nodes. For each node, its neighboring nodes in the bipartite graph are determined, and their corresponding neighbor vectors are obtained. For each node, the neighbor vectors are weighted and aggregated according to the edge weights between the node and its neighbors to obtain a neighbor aggregate vector. This neighbor aggregate vector is then fused with the node's corresponding basic vector or job vector to obtain an aggregated basic vector or aggregated job vector. Based on the aggregated basic vector and aggregated job vector, the first piece of information to be recommended and its corresponding first matching degree are determined.

[0048] For example, a bipartite graph can include two types of nodes: user nodes and job nodes, and different types of nodes can be connected by edges. That is, edges can be established between different types of nodes based on user operation information. It can be understood that the edges between nodes can be used to reflect the interaction behavior between users and jobs.

[0049] Different weights can be assigned to edges based on the depth of interaction reflected in the action information. In one example, the greater the interaction depth, the greater the edge weight. For instance, if the user's action information is "favorite," an edge is created between the corresponding user node U and the job node J, with a weight coefficient of 5; or, if the user's action information is "schedule an interview," an edge is created between the corresponding user node and the job node, with a weight coefficient of 4; or, if the user's action information is "apply for service," an edge is created between the corresponding user node and the job node, with a weight coefficient of 3. Similarly, communication and click actions can be assigned weight coefficients of 2 and 1 respectively, and if there is no interaction, an edge connection is created.

[0050] Different types of nodes represent different information. Basic features in user nodes describe basic user information, such as city, gender, age, device, and spending level. Job information in job nodes describes basic job information, which may include job skills, value matching information, and work location.

[0051] For example, each node in the bipartite graph can be mapped to obtain a vector representation of each node.

[0052] In one example, an initial embedding layer can be used to transform the basic features and job features into low-dimensional dense basic vectors and job vectors, respectively. The basic vector can be represented as... The job vector can be represented as: . The number of user nodes in the bipartite graph; d represents the number of job nodes in the bipartite graph; d represents the embedding dimension. This indicates layer 0 (i.e., the initial embedding layer).

[0053] Neighbor nodes are all nodes in the bipartite graph that are directly connected to the current node by an edge. If the current node is a user node, the neighbor nodes are job nodes; conversely, if the current node is a job node, the neighbor nodes are user nodes. It's understood that the vector representation of the user node or job node needs to be updated. Furthermore, each neighbor node has a corresponding neighbor node vector.

[0054] For example, features of nodes in a bipartite graph can be aggregated using graph convolutional layers. These graph convolutional layers can include multiple convolutional layers. Each convolutional layer can be used to aggregate the features of a node with the features of its neighbors. In one example, the graph convolutional layer can employ a Light Graph Convolutional Network (LightGCN).

[0055] For example, the basic vector or job vector corresponding to a node can be weighted and fused with the neighbor node vectors corresponding to neighboring nodes according to their weights to obtain the neighbor aggregation vector. Wherein, if the current node is a user node, the neighbor aggregation vector can be calculated using the following formula (2): If the current node is a job node, the neighbor aggregation vector can be calculated using the following formula (3): in, It is the embedding of user u in layer I (i.e., the neighbor aggregation vector corresponding to the user node). It uses the embedding of job j at layer I (i.e., the neighbor aggregation vector corresponding to the job node). These are the neighboring job nodes of user u (i.e., the jobs that the user has clicked / favorited). It is the neighboring user node of position j; It is the number of user u's neighbors. It represents the number of users interacting with job j; weight. This is used to balance node intensity and prevent popular positions from excessively affecting the results; The edge weight is denoted as .

[0056] For example, the neighbor aggregation vector is fused with the base vector or job vector corresponding to the node to obtain the aggregated base vector or aggregated job vector.

[0057] In one example, multi-level aggregation can be performed to obtain either an aggregated base vector or an aggregated job vector. The aggregated base vector... It can be expressed as the following formula (4); the aggregated job vector It can be expressed as the following formula (5): Here, l represents the number of convolutional layers. It's understandable that in the first convolutional layer (l=1), the user vector only aggregates nodes of "directly interacting positions," such as the features of position B directly clicked by user A. In the first convolutional layer (l=2), the user vector aggregates the features of "interacting users of interactive positions." For example, if position B is clicked by user C, and user C then clicks on position D, then user A's embedding will indirectly absorb the features of position D. Therefore, it's understandable that in higher-level aggregation processes, implicit associations along longer paths can be captured, such as user A→position B→user C→position D→user E→position F. User A's embedding may contain features of position F, thus uncovering potential preferences across multiple steps.

[0058] For example, a user's preference score for a job can be calculated based on the aggregated base vector and the aggregated job vector. The first piece of information to be recommended and its corresponding first matching degree can then be determined based on the preference score.

[0059] In one example, the preference score can be calculated according to the following formula (6). : Where T represents transpose.

[0060] In another example, the user's preference scores for all job positions can be sorted in descending order, and a preset number of job positions can be output as the first list of recommended positions. Furthermore, the preference score can be used as the first match score for these first list of recommended positions. For instance, the top 20 job positions in the sorted job sequence representation can be output as the first list of recommended positions.

[0061] In this embodiment, a bipartite graph between users and job positions is established, and the interaction depth is quantified as edge weights. This allows the strength of user intent reflected by different operational information to be directly encoded into the connection strength of the graph structure. Based on this, during the vector aggregation stage, neighboring nodes with strong interaction depth contribute more information, while the contribution of neighboring nodes with weak interaction depth is suppressed. This enables the aggregated user vector to dynamically capture the preferences expressed by the user's actual behavior, avoiding the problem that relying solely on static resumes is insufficient to reflect changes in the user's true job-seeking intent. Simultaneously, the vectors of job position nodes are also enhanced after aggregating the characteristics of their interacting users, thus implicitly incorporating the characteristics of the user group interested in that position. This allows information discrepancies arising from non-standardized skill descriptions to be effectively supplemented and corrected by the user's actual operational information. Finally, recommendations are made based on the matching scores calculated using these enhanced vectors, improving the accuracy, personalization, and adaptability to sparse behavioral scenarios in job recommendations.

[0062] In some alternative embodiments, the target user type includes cold-start users. Further, user enhancement features can be determined by obtaining the user's basic attribute information and skill information, and based on the basic attribute information and skill information. Further, a bipartite graph consisting of multiple nodes is constructed based on the user enhancement features and job characteristics.

[0063] For example, basic attribute information can be the user's basic information, including at least the city, gender, age, device, and consumption level. Skill information is the user's description of their own abilities.

[0064] For example, user-enhanced features can be obtained by extracting features from basic attribute information and skill information using a pre-defined large language model. In one example, the pre-defined large language model is a Word2Vec model, which can be trained using basic attribute information, skill information, and skill information labels corresponding to job clusters.

[0065] For example, after obtaining enhanced user information, a bipartite graph consisting of multiple nodes can be constructed based on enhanced user features and job features. The edges connecting the nodes are constructed based on operation information. The basic features corresponding to user nodes and the job features corresponding to job nodes are mapped respectively, resulting in a basic vector for the user node and a job vector for the job node. For each node, its neighboring nodes in the bipartite graph are determined, and their corresponding neighboring node vectors are obtained. For each node, the neighboring vectors are weighted and aggregated according to the edge weights between the node and its neighboring nodes to obtain a neighboring aggregated vector. This neighboring aggregated vector is then fused with the node's corresponding basic vector or job vector to obtain an aggregated basic vector or aggregated job vector. Based on the aggregated basic vector and aggregated job vector, the first recommended information and its corresponding first matching degree are determined.

[0066] In one example, feature extraction can be performed on basic attribute information and skill information to obtain the user's initial embedding. and skill embedding Furthermore, for the initial user embedding... and skill embedding The vectors are concatenated to obtain the user enhancement vector. ,Right now = Furthermore, the generated As input, it participates in the graph propagation process of LightGCN to generate the first recommendation information and the corresponding first matching degree.

[0067] In this embodiment, by acquiring users' basic attribute information and skill information, the user characteristics of cold-start users are enhanced to alleviate the difficulties of cold start caused by data sparsity. Based on this, a bipartite graph is constructed using the enhanced user features and job features, enabling nodes in the graph to carry rich attribute semantics even in the absence of interaction records, ensuring the accuracy and robustness of subsequent graph convolutional message passing and preference prediction.

[0068] In some embodiments, in S140, the user's skill information is obtained.

[0069] Skill information can be non-standardized text provided by users describing their own abilities. For example, blue-collar workers may fill in scattered skill statements such as "can perform MIG welding", "have an electrician's certificate", or "have worked as a lathe operator for ten years".

[0070] In some embodiments, in S150, a second recommended information and a corresponding second matching degree are determined based on the similarity between the skill information and the skill information tags corresponding to the preset job cluster.

[0071] A pre-defined job cluster can be a set of blue-collar jobs pre-aggregated according to the commonalities of required skills. Each pre-defined job cluster includes at least a skill information tag to describe the skill requirements of that cluster. In addition, each pre-defined job cluster may also include value matching information (such as salary), geographic information, and one or more jobs contained within that cluster.

[0072] For example, the following table (1) shows several preset job clusters and their contents: Table (1)

[0073] For example, the second information to be recommended can be selected by calculating the similarity between the user's skill information and the skill information tags of each job cluster, and filtering out the matching job clusters and their corresponding jobs.

[0074] The second matching degree is a quantitative score measuring the similarity between user skill information and job cluster tag set, which can characterize the skill matching degree.

[0075] In one example, the second matching degree can be calculated using the cosine similarity calculation formula, which is shown in formula (7) below: Where match_score is the matching score corresponding to the second matching degree. Job cluster skill information tags; This refers to user skill information. Among them, , Both can be represented by vectors. It can be represented as ; It can be represented as The skill list includes [s1, s2, ..., sn], where sn represents the nth skill information filled in by the user. This indicates the average pooling operation.

[0076] In some optional embodiments, job information corresponding to multiple job positions can be obtained. For each job position, keywords are extracted from the job information according to a preset keyword frequency calculation formula to obtain a keyword feature vector corresponding to each job position. Multiple keyword feature vectors are clustered to obtain multiple job position clusters. For each job position cluster, the target job information of that cluster is obtained, and the skill information tags corresponding to that job position cluster are determined based on the skill information in the target job information and preset filtering conditions.

[0077] For example, job information can represent various text descriptions associated with each job, such as job title, description (including job responsibilities, job requirements and work content, etc.), work address, value matching information and other full information.

[0078] For example, job information can be cleaned to remove invalid and redundant content. In one example, data cleansing standardizes date formats, locations (unifying categories such as park, residential area, etc.), and value-matching information units (e.g., unifying them as monthly salary). Furthermore, missing data is filled using the mode, and job keywords (e.g., painting walls, plumbing, cleaning, etc.) are extracted based on job titles and descriptions.

[0079] For example, the preset keyword frequency calculation formula can be a statistical algorithm used to extract important words from job information, which can quantify the importance of words in job information. The preset keyword frequency calculation formula transforms unstructured job information into structured, computable feature vectors.

[0080] In one example, the preset keyword frequency calculation formula can be the Term Frequency-Inverse Document Frequency (TF-IDF) method, which reduces the weight of common words by weighting the statistical word frequencies. A vocabulary of the entire set of job postings can be constructed. For each job posting in the vocabulary, the TF-IDF value of each word in the vocabulary is calculated based on the corresponding job information. The job information for each job posting is described as a high-dimensional sparse vector, with a dimension equal to the size of the vocabulary, and each element corresponding to the TF-IDF score of a word. For each job posting's TF-IDF vector, the average of all non-zero elements is calculated, and this average is used as a threshold. Words with TF-IDF scores higher than this threshold are retained as key feature words representing that job posting.

[0081] The TF-IDF value can be calculated using the following formula (8): Where N is the total number of all job titles in the document, and DF(t) is the number of documents containing the word t. This represents the frequency of word t in job d. The inverse document frequency (IDF) is the logarithm of the reciprocal of the frequency of the word across the entire document set.

[0082] For example, if the vocabulary has 1,000 words, and 100 different words actually appear in the manufacturing-assembly worker job description document, calculate the TF-IDF score of these 100 words using the formula (6) above, and the score of the remaining 900 words is 0. Take these 100 non-zero TF-IDF scores and calculate their arithmetic mean. Further, retain the words in the document whose TF-IDF scores are higher than the average. This group of words will be used as the key feature words corresponding to the job after the document has been filtered.

[0083] For example, the keyword feature vector can be a vector composed of keywords extracted from each job position.

[0084] For example, clustering operations can be performed on keyword feature vectors to divide job positions into multiple job clusters. Each job cluster includes one or more keyword feature vectors.

[0085] In some optional embodiments, a first distance value can be calculated between keyword feature vectors, and keyword feature vectors whose first distance values ​​satisfy a preset distance condition can be merged to obtain a merged initial job cluster. It is then determined whether the initial job cluster meets a stopping condition. If it is determined that the initial job cluster does not meet the stopping condition, for each initial job cluster, the average vector of all keyword feature vectors within that initial job cluster is calculated, and a second distance value is calculated between multiple average vectors. Keyword feature vectors whose second distance values ​​satisfy the preset distance condition are merged, the initial job cluster is updated, and the step of determining whether the initial job cluster meets the stopping condition is returned, until the stopping condition is met, resulting in multiple job clusters.

[0086] For example, the keyword feature vector is a numerical feature vector that uniformly represents a job position, obtained by fusing features from job information. In one example, salary information is normalized to obtain a numerical vector; categorical features such as location and experience are One-Hot encoded; the job description text is calculated using TF-IDF to obtain a text vector; and the above vectors are horizontally concatenated to generate a fixed-length fused feature vector, which serves as the keyword feature vector for that job position.

[0087] The first distance value represents the similarity measure between the feature vectors of two keywords. In one example, the first distance value CosineDist(A,B) can be measured using cosine distance, as shown in formula (9) below: in, , These represent different keyword feature vectors.

[0088] The preset distance condition is used to determine whether to merge keyword feature vectors or job clusters. In one example, the preset distance condition is to merge the pair of vectors or clusters with the smallest distance value.

[0089] Stopping conditions are rules used to determine whether the clustering iteration process has terminated. In one example, the stopping condition is that after all positions are aggregated into a large cluster and a complete dendrogram is generated, the final cluster partitioning is determined by observing where the dendrogram is cut at a point with a large distance jump.

[0090] The average vector is the arithmetic mean of all keyword feature vectors within a given job cluster across all dimensions. This vector represents the center position of the cluster in the feature space.

[0091] The second distance value is the distance between the average vectors of different job clusters. In one example, the second distance value d(A,B) can be calculated using the average distance method shown in formula (10): Here, A and B represent different job clusters.

[0092] In one example, Figure 2 A schematic diagram of clustering provided in one embodiment of this application is shown. It will be understood that a bottom-up hierarchical clustering method can be employed. First, each job is considered as a separate cluster (e.g., ...). Figure 2 (As shown in Figure A). To improve the business rationality of the clustering results, before calculating similarity, the keyword feature vectors are preprocessed with weights based on business importance. For example, the feature dimensions are divided into three core dimensions: job functions and skills, work location and commuting, and salary and benefits, and weighted by 0.5, 0.3, and 0.2 respectively. After weighting, the first distance value (e.g., cosine distance) between all keyword feature vectors is calculated to form a distance matrix. The two keyword feature vectors whose first distance values ​​satisfy the preset distance condition (i.e., the smallest distance) are merged to obtain the initial job cluster (e.g., job function cluster). Figure 2 The multiple initial job clusters M shown.

[0093] Further, it is determined whether the initial job clusters meet the stopping condition. If it is determined that the initial job clusters do not meet the stopping condition, for each initial job cluster, the average vector of all keyword feature vectors within that initial job cluster is calculated, and a second distance value is calculated between multiple average vectors. The second distance value is calculated using the average distance method to comprehensively reflect the average similarity of all point pairs between clusters and avoid the influence of extreme values.

[0094] The two initial job clusters with the smallest second distance values ​​are merged, and the distance matrix is ​​updated until all jobs are aggregated into one large cluster and a complete dendrogram is generated. By observing the dendrogram and cutting at locations with large distance jumps, the final job cluster division is determined.

[0095] In this embodiment, by iteratively calculating the distance between keyword feature vectors and merging them based on preset distance conditions, job clusters can be constructed from the bottom up. In each iteration, the average vector within a cluster represents the overall semantics of the cluster, and then the second distance value between clusters is calculated based on the average vector for merging. This effectively filters out noise and abnormal fluctuations in individual job descriptions, making the clustering results more robust. Simultaneously, by setting stopping conditions, a reasonable number of clusters can be automatically determined, avoiding the subjectivity of manually setting the number of clusters and ensuring high similarity of jobs within clusters and clear separation of jobs between clusters. This process results in job clusters with high cohesion and high discriminative power in terms of business operations, providing a solid foundation for subsequent data processing.

[0096] For example, for each job cluster, the target job information for that cluster can be obtained, and skill information tags corresponding to that job cluster can be determined based on the skill information in the target job information and preset filtering conditions. Here, the target job information refers to the features corresponding to a certain cluster. The skill information tags are obtained from the target job information based on skill-related fields and preset filtering rules.

[0097] In one example, for each generated job cluster, its representative labels and statistical features are extracted, specifically including: Identifying representative jobs: Statistically selecting the most frequently occurring job names in the cluster. Extracting core skill labels: Statistically calculating the skill vectors of all jobs within the cluster, calculating the frequency or percentage of each skill, and selecting the 5 most frequently occurring skills as the core skill labels for the cluster. Calculating salary ranges: Collecting salary data for all jobs within the cluster, calculating the median, 25th percentile (Q1), and 75th percentile (Q3), expressed in the form [Q1-Q3], such as "5K (thousand) - 10K / month". Statistically analyzing geographical distribution: Statistically analyzing the work locations of all jobs within the cluster, calculating the frequency or percentage of each location, and selecting the 3 most frequently occurring locations as the main geographical distribution areas for the cluster. Defining cluster names: Based on the core features of the cluster, defining a business-meaning name for each job cluster. The final output is a list of job clusters, with each record containing: cluster name, representative job, core skill label, salary range, and main geographical distribution.

[0098] Furthermore, in one or more embodiments of this application, the generated job clusters are evaluated, and the evaluation methods include quantitative evaluation and manual review.

[0099] The silhouette coefficient can be used to measure the tightness between a job and its own cluster, as well as its separation from other clusters. The overall silhouette coefficient of the system is the average of the silhouette coefficients of all samples. The closer this value is to 1, the better the clustering effect. The silhouette coefficient of a single sample i is calculated using the following formula (11): Where a(i) represents the average distance from sample i to all other samples in the same cluster. The smaller the value, the closer the sample is to its own cluster. b(i) represents the minimum average distance from sample i to all other samples in the same cluster. The larger the value, the more separated the sample is from other clusters.

[0100] Furthermore, it is understandable that when s(i) is close to 1, it indicates that the clustering is reasonable, and the sample point is close to its own cluster center and far away from other clusters; when s(i) is close to 0, it indicates that the sample point is on the boundary between two clusters; when s(i) is close to -1, it indicates that the sample point may have been misassigned.

[0101] Technical personnel review the generated job clusters and their characteristics to verify their rationality, interpretability, and business value, and make necessary manual adjustments based on the review results. Specifically, the review may include: intra-cluster consistency (randomly selecting several jobs within each cluster to verify the high similarity of their core skill tags, job content, and requirements); inter-cluster distinguishability (verifying the clarity, rationality, and effectiveness of core skill tags and job profiles across different clusters); tag value (verifying the accuracy, unambiguity, and effectiveness of the automatically generated "top 5 core skill tags" in representing the core requirements of the job in that cluster); business logic (verifying whether the clustering results align with common sense in the recruitment market or among industry experts); and application value (verifying whether the generated clusters and tags can be directly used to improve the filtering logic of recommendation systems, user profile matching, or job posting guidance for recruiters).

[0102] In this embodiment, by acquiring job information from multiple positions, and using a preset keyword frequency calculation formula, keywords are automatically extracted from unstructured job text to generate structured keyword feature vectors, achieving semantic dimensionality reduction of a large amount of job information. Based on this, the keyword feature vectors are clustered, automatically classifying semantically similar jobs into multiple job clusters, solving the problems of information silos and feature sparsity caused by diverse job names and colloquial descriptions. For each cluster, skill information tags are extracted from its target job information based on skill information and preset screening conditions, enabling each job cluster to be represented by a standardized skill dimension. This process transforms the original, complex job data into a clearly defined, semantically consistent, and tagged job cluster system, providing high-quality, highly interpretable job node features for subsequent construction of recommendation models such as bipartite graphs, improving the system's generalization ability and business adaptability in downstream matching and recommendation tasks.

[0103] S160. Determine the recommended information based on the first information to be recommended, the first matching degree, the second information to be recommended, and the second matching degree.

[0104] The recommendation information is generated by fusing the first piece of information to be recommended and its first matching degree with the second piece of information to be recommended and its second matching degree, such as by weighted sorting, deduplication, or combination.

[0105] For example, the first matching degree and the second matching degree can be weighted and fused to determine the recommendation information. In one example, for users who have interacted, the fused matching degree can be calculated using the following formula (12); for users who are cold-starting, the fused matching degree can be calculated using the following formula (13): in, , These are the weighting coefficients. =0.4; =0.7. ; ; .

[0106] In this embodiment, by acquiring operational information and classifying users into different user types accordingly, and then selecting corresponding recommendation strategies for different user types to generate the first recommended information, the problem of recommendation systems failing to start or having low-quality recommendation information in scenarios with new users or sparse data is effectively solved. This ensures that there is an appropriate recommendation information generation path under different levels of data completeness. Simultaneously, by pre-setting job clusters and standardized skill information tags, the non-standardized, colloquial skill descriptions that are difficult to uniformly measure among blue-collar workers are converted into similarity calculations with each job cluster tag set to generate the second recommended information and matching degree. This objectively and quantitatively assesses the "non-standardized skills" and practical experience of blue-collar workers, reducing the limitation of relying solely on keyword string matching to identify implicit skills. Furthermore, by integrating the behavioral recommendation results based on user type strategies with the structured recommendation results based on skill tag similarity, the final recommended jobs not only respond to existing behavioral preferences but also uncover potential jobs matching the skill structure of cold-start users or users with sparse behavior, improving the accuracy, coverage, and multi-scenario adaptability of job recommendations for blue-collar workers.

[0107] In another optional embodiment, in order to meet the needs of blue-collar users for timely job placement, as another implementation of this application, this application also provides another implementation of the information recommendation method, as detailed in the following embodiments.

[0108] Figure 3 A flowchart illustrating an embodiment of the information recommendation method provided in this application is shown. Figure 3 As shown, the method may include the following steps: S310. Based on the preset value distance regression equation, calculate the spatial distance constraints corresponding to the value matching information in the first and second recommended information respectively.

[0109] For example, the first set of recommended information and the second set of recommended information can represent two sets of job positions to be screened.

[0110] The preset value distance regression equation is used to quantify the difference between value matching information and distance. Value matching information can include quantifiable indicators such as salary and benefits, which have varying levels of intensity. This equation can fit the acceptable commuting distance for a user under the corresponding value matching information.

[0111] Spatial distance constraints can be a distance tolerance dynamically generated by a value distance regression equation based on the degree of value matching information.

[0112] In some optional embodiments, the preset value distance regression equation is shown in Equation (1): Where x represents value matching information; y represents spatial distance; , , For coefficients, This is the error term.

[0113] It is understandable that the preset value distance regression equation shown in formula (11) can be obtained by training a k-th order polynomial regression model. The k-th order polynomial regression model can be expressed as: .

[0114] Understandably, the presupposed value distance regression equation is a second-order polynomial regression scheme. The introduction of the quadratic term can effectively capture the downward convex curve relationship where the distance growth rate gradually slows down as value matching information increases, which is more in line with actual decision-making scenarios.

[0115] S320. Calculate the spatial distance based on the location information in the first and second information to be recommended, as well as the user's baseline location information.

[0116] For example, user reference location information refers to the geographical location specified by the user as a commuting reference point, such as the user's place of residence, current location, or the center point of the desired work area.

[0117] For example, the location information (such as job coordinates) in the first and second information to be recommended is obtained, and the user's baseline location information (such as the user's address) is obtained, and the physical spatial distance between the two (such as commuting straight-line distance or driving distance) is calculated.

[0118] In one example, spatial distance can be calculated using the Haversine formula (as shown in formula (14) below). It is understood that latitude and longitude can be converted from degrees to radians, and the output... 1, 2, λ1, λ2. Calculate the difference in longitude and latitude between the two points and output Δ. Substituting Δλ into the Haversine formula, we calculate the parameter (a) and then use the inverse trigonometric function atan² to calculate the spherical distance (c). Finally, we multiply this by the Earth's radius, and then multiply the calculated angular distance (c) by the Earth's radius (R) to obtain the actual distance (d).

[0119] in, 1, 2 represents the latitude (in radians) of the two points, and λ1 and λ2 represent the longitude (in radians) of the points; Δ = 2 1 (Latitude difference); Δλ = λ2 λ1 (longitude difference); R: Earth's radius.

[0120] S330 filters the first and second recommended information based on spatial distance constraints and spatial distance, resulting in the filtered first and second recommended information.

[0121] For example, candidate positions are filtered by real spatial distance and spatial distance constraints, retaining positions whose real spatial distance is less than or equal to the spatial distance constraints. That is, only positions whose commuting distance is within the user's tolerance range for the value of the position are recommended, thus obtaining the filtered recommendation information.

[0122] S340. Calculate the third matching degree based on the spatial distance and value matching information in the first recommended information after filtering and the second recommended information after filtering.

[0123] For example, within the candidate set after initial screening (including the first recommended information and the second recommended information after screening), a third matching degree is calculated using the selected spatial distance and value matching information. The third matching degree represents the final score of the position in terms of value matching information and distance after excluding obviously unreasonable, long-distance, low-value positions.

[0124] In one example, the third matching score can be calculated using the following formula (15): in, The minimum value among the value matching information in the first piece of information to be recommended and the second piece of information to be recommended; The maximum value among the value matching information in the first piece of information to be recommended and the second piece of information to be recommended; It can be represented as Here, we can take the average of the first set of recommended information and the second set of recommended information after filtering.

[0125] S350. Based on the first matching degree, the second matching degree, and the third matching degree, determine the recommended information from the filtered first information to be recommended and the filtered second recommended information.

[0126] For example, a first matching score measuring behavioral preferences and a second matching score measuring content / skill matching are obtained, and combined with the third matching score calculated above, a multi-factor weighted fusion is performed. Based on the fused total score, the final recommendation information to be recommended to the user is determined from the filtered candidate job information. For example, a job recommendation list is output in descending order of score.

[0127] In one example, fusion can be performed using the following formula (16): in, , The weighting coefficient can be set. =0.3; =0.7.

[0128] In this embodiment, a preset value distance regression equation is used to dynamically generate acceptable spatial distance constraints based on the value matching information of the job positions. This allows high-value positions to have a more lenient commuting tolerance range, avoiding the loss of quality opportunities due to mechanical distance truncation. Meanwhile, low-value positions are only pushed to users when they are nearby, thereby improving the cost-effectiveness of recommendations and the user's actual commuting willingness. Based on the dynamic constraint filtering, the filtered spatial distance and value information are further integrated to calculate a third matching degree, which is then combined with a first matching degree based on behavioral preferences and a second matching degree based on content for multi-factor comprehensive decision-making. This hierarchical filtering and integration strategy ensures that the recommendation results strictly meet the user's geographical accessibility requirements while accurately measuring the comprehensive attractiveness of the job from multiple dimensions. Ultimately, it outputs highly personalized recommendation information that balances commuting costs and job quality, significantly improving recommendation conversion rate and interpretability.

[0129] Below, in conjunction with Figure 4 The following examples illustrate information recommendation methods.

[0130] Figure 4 A flowchart illustrating an embodiment of the information recommendation method provided in this application is shown, as follows: Figure 4 As shown, the system first performs preprocessing on the raw job information, including information extraction and data cleaning to extract core job information. Then, Natural Language Processing (NLP) technology is used for feature extraction, transforming the job data into standardized vectors that the model can process. Finally, a hierarchical clustering algorithm is used to output structured job cluster descriptions. In terms of user preference perception, the system adopts a dual-track approach: on the one hand, it collects users' personal skills and job intentions information actively filled in on the platform and performs semantic matching with job cluster descriptions to accurately identify users' explicit job intentions; on the other hand, it integrates users' historical job interaction data and user operator attribute features, combining them with job cluster descriptions to conduct small-sample behavioral data modeling. For cold-start users with no platform interaction records, a pre-trained regressor can be imported to generate personalized pseudo-Embeddings for these users, solving the recommendation problem for cold-start users.

[0131] Furthermore, the data enters the core algorithm module, which relies on the LightGCN graph convolutional recommendation algorithm to integrate three types of information: explicit user intentions, behavioral embeddings of ordinary users, and pseudo-embeddings of cold-start users. This process calculates, generates, and outputs an initial recommendation list of job preferences for all users.

[0132] Finally, the intelligent filtering and sorting process for commuting is implemented. This involves extracting the latitude, longitude, and salary levels of initially recommended positions, while also obtaining the latitude and longitude of the user's permanent residence based on their mobile operator attributes. Through dynamic modeling, the maximum acceptable commuting distance threshold for the user is calculated based on salary, and the actual commuting distance between the position and the user's residence is precisely calculated using a formula. Data comparison is then performed to eliminate positions with commuting distances exceeding the user's acceptable limit. The remaining eligible positions are then comprehensively sorted, ultimately outputting a final list of recommended positions that matches the user's job search needs, salary expectations, and commuting ability.

[0133] In this embodiment, a multimodal blue-collar user modeling architecture is constructed, integrating domain-specific Word2Ve semantic matching for explicit needs, LightGCN graph model mining for implicit behaviors, and transfer learning of basic attributes for cold-start users. A decision tree is built based on behavioral activity, explicit input completeness, and cold-start attribute validity. Abnormal recommendations are proactively blocked for users with missing attributes, improving the matching accuracy in cold-start scenarios and covering the needs of various blue-collar users across all scenarios, thus overcoming the cold-start recommendation dilemma. Furthermore, dynamic modeling using multinomial regression of salary and commuting distance is employed. Based on salary and historical behavioral data, a multinomial regression equation is fitted to accurately fit the law of "diminishing marginal returns on commuting tolerance due to salary increases." Coupled with salary-driven commuting threshold calculation and Haversine spherical distance, a risk attenuation factor is dynamically injected, simultaneously filtering high-risk commuting positions and prioritizing recommendations for nearby high-paying positions, reducing user commuting costs and better aligning with the job-seeking decision-making logic of blue-collar workers. Compared to traditional linear models, this approach can more effectively improve the job recommendation matching rate.

[0134] Figure 5 This is a schematic diagram of an information recommendation device provided in an embodiment of this application. Figure 5 As shown, the device 500 may include: The first acquisition module 501 is used to acquire the user's operation information and determine the target user type to which the user belongs from multiple user types based on the operation information; The first determining module 502 is used to determine a target recommendation strategy that matches the target user type from multiple recommendation strategies for determining the information to be recommended, based on a preset correspondence relationship. The preset correspondence relationship includes the correspondence relationship between different user types and recommendation strategies. The second determining module 503 is used to determine the first information to be recommended and the corresponding first matching degree based on the target recommendation strategy and operation information; The second acquisition module 504 acquires the user's skill information; The third determining module 505 is used to determine the second recommended information and the corresponding second matching degree based on the similarity between the skill information and the skill information tags corresponding to the preset job clusters; The fourth determining module 506 is used to determine the recommended information based on the first information to be recommended, the first matching degree, the second information to be recommended, and the second matching degree.

[0135] In one embodiment, both the first and second recommended information include location information and value matching information. The fourth determining module 506 determines the recommended information based on the first recommended information, the first matching degree, the second recommended information, and the second matching degree in the following manner: Based on a preset value distance regression equation, the spatial distance constraints corresponding to the value matching information in the first and second recommended information are calculated respectively; the spatial distance is calculated based on the location information and the user's baseline location information in the first and second recommended information; the first and second recommended information are filtered based on the spatial distance constraints and the spatial distance to obtain filtered first and second recommended information; a third matching degree is calculated based on the spatial distance and value matching information in the filtered first and second recommended information; and recommended information is determined from the filtered first and second recommended information based on the first matching degree, the second matching degree, and the third matching degree.

[0136] In one implementation, the preset value distance regression equation is shown in formula (1): Where x represents value matching information; y represents spatial distance; , , For coefficients, This is the error term.

[0137] In one implementation, the target user type includes interactive users; the first determining module 502 determines the first information to be recommended and the corresponding first matching degree according to the target recommendation strategy and operation information in the following manner: obtaining user basic features and job features; constructing a bipartite graph consisting of multiple nodes based on the user features and job features, wherein the nodes include user nodes representing user basic features and job nodes representing job features, and the edges connecting the nodes are constructed based on operation information; mapping the basic features corresponding to the user nodes and the job features corresponding to the job nodes respectively to obtain the basic vector corresponding to the user nodes and the job vector corresponding to the job nodes; for each node, determining the neighbor nodes of the node in the bipartite graph and obtaining the neighbor node vector corresponding to the neighbor nodes; for each node, performing weighted aggregation of the neighbor node vectors according to the edge weights of the edges between the node and each neighbor node to obtain a neighbor aggregation vector, and fusing the neighbor aggregation vector with the basic vector or job vector corresponding to the node to obtain an aggregated basic vector or an aggregated job vector; determining the first information to be recommended and the corresponding first matching degree based on the aggregated basic vector and the aggregated job vector.

[0138] In one embodiment, the target user type further includes a cold start type; before constructing a bipartite graph consisting of multiple nodes based on user characteristics and job characteristics, the first determining module 502 is further configured to: obtain the user's basic attribute information and the user's skill information; determine the user's enhanced characteristics based on the basic attribute information and skill information; the first determining module 502 constructs a bipartite graph consisting of multiple nodes based on user characteristics and job characteristics in the following manner: constructing a bipartite graph consisting of multiple nodes based on user enhanced characteristics and job characteristics.

[0139] In one embodiment, before determining the second recommended information and the corresponding second matching degree based on the similarity between the skill information and the skill information tags corresponding to the preset job clusters, the third determining module 505 is further configured to: obtain job information corresponding to multiple jobs; for each job, extract keywords from the job information according to a preset keyword frequency calculation formula to obtain a keyword feature vector corresponding to each job; cluster the multiple keyword feature vectors to obtain multiple job clusters; for each job cluster, obtain the target job information of the job cluster, and determine the skill information tag corresponding to the job cluster based on the skill information in the target job information and preset filtering conditions.

[0140] In one embodiment, the third determining module 505 is further configured to cluster multiple keyword feature vectors to obtain multiple job clusters: calculate a first distance value between keyword feature vectors; merge keyword feature vectors whose first distance value satisfies a preset distance condition to obtain a merged initial job cluster; determine whether the initial job cluster meets a stopping condition; if the initial job cluster does not meet the stopping condition, for each initial job cluster, calculate the average vector of all keyword feature vectors within the initial job cluster, and calculate a second distance value between multiple average vectors; merge keyword feature vectors whose second distance value satisfies the preset distance condition, update the initial job cluster, and return to the step of determining whether the initial job cluster meets the stopping condition, until the stopping condition is met, to obtain multiple job clusters.

[0141] Figure 6 A schematic diagram of the hardware structure of the information recommendation device provided in an embodiment of this application is shown.

[0142] The information recommendation device may include a processor 601 and a memory 602 storing computer program instructions.

[0143] Specifically, the processor 601 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0144] Memory 602 may include mass storage for data or instructions. For example, and not limitingly, memory 602 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. In one instance, memory 602 may include removable or non-removable (or fixed) media, or memory 602 may be non-volatile solid-state memory. Memory 602 may be internal or external to the information delivery device.

[0145] In one instance, memory 602 may be read-only memory (ROM). In one instance, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or flash memory, or a combination of two or more of these.

[0146] Memory 602 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Therefore, generally, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of this disclosure.

[0147] The processor 601 reads and executes computer program instructions stored in the memory 602 to achieve... Figure 1 The information recommendation method in the illustrated embodiment.

[0148] In one example, the information recommendation device may further include a communication interface 603 and a bus 604. For example, Figure 6 As shown, the processor 601, memory 602, and communication interface 603 are connected through bus 604 and complete communication with each other.

[0149] The communication interface 603 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0150] Bus 604 includes hardware, software, or both, that couples components of an information recommendation device together. For example, and not as a limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 604 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, this application contemplates any suitable bus or interconnect.

[0151] This information recommendation device can execute the information recommendation method described in this application based on user operation information and user skill information, thereby achieving a combination of... Figure 1 and Figure 3 The method of recommending information described.

[0152] Furthermore, in conjunction with the information recommendation methods described in the above embodiments, this application embodiment can provide a computer storage medium for implementation. This computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the information recommendation methods described in the above embodiments.

[0153] This application also provides a computer program product, including a computer program that, when executed by a processor, implements any of the information recommendation methods described in the above embodiments.

[0154] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0155] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, read-only memory (ROM), flash memory, erasable read-only memory (EROM), floppy disks, compact disc read-only memory (CD-ROM), optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0156] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0157] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0158] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. An information recommendation method, characterized in that, include: Obtain user operation information, and determine the target user type to which the user belongs from multiple user types based on the operation information; According to a preset correspondence, a target recommendation strategy that matches the target user type is determined from multiple recommendation strategies used to determine the information to be recommended. The preset correspondence includes the correspondence between different user types and recommendation strategies. Based on the target recommendation strategy and the operation information, the first information to be recommended and the corresponding first matching degree are determined; Obtain the user's skill information; Based on the similarity between the skill information and the skill information tags corresponding to the preset job clusters, the second information to be recommended and the corresponding second matching degree are determined. Recommended information is determined based on the first information to be recommended, the first matching degree, the second information to be recommended, and the second matching degree.

2. The information recommendation method according to claim 1, characterized in that, Both the first and second recommended information include location information and value matching information; The step of determining recommendation information based on the first information to be recommended, the first matching degree, the second information to be recommended, and the second matching degree includes: Based on the preset value distance regression equation, the spatial distance constraints corresponding to the value matching information in the first and second information to be recommended are calculated respectively. Calculate the spatial distance based on the location information in the first and second information to be recommended, as well as the user's baseline location information; Based on the spatial distance constraint and the spatial distance, the first information to be recommended and the second information to be recommended are filtered to obtain the filtered first information to be recommended and the filtered second information to be recommended. The third matching degree is calculated based on the spatial distance in the first recommended information after filtering, the second recommended information after filtering, and the value matching information; Based on the first matching degree, the second matching degree, and the third matching degree, the recommended information is determined from the filtered first information to be recommended and the filtered second recommended information.

3. The information recommendation method according to claim 2, characterized in that, The preset value distance regression equation is shown in formula (1): Where x represents the value matching information; y represents the spatial distance; , , For coefficients, This is the error term.

4. The information recommendation method according to claim 1, characterized in that, The target user types include interactive users; The step of determining the first information to be recommended and the corresponding first matching degree based on the target recommendation strategy and the operation information includes: Obtain basic user characteristics and job characteristics; A bipartite graph consisting of multiple nodes is constructed based on the user basic characteristics and the job characteristics. The nodes include user nodes representing the user basic characteristics and job nodes representing the job characteristics, and the edges connecting the nodes are constructed based on the operation information. The basic features corresponding to the user node and the job features corresponding to the job node are mapped respectively to obtain the basic vector corresponding to the user node and the job vector corresponding to the job node; For each node, determine the neighboring nodes of the node in the bipartite graph, and obtain the neighboring node vector corresponding to the neighboring node; For each node, the neighbor node vectors are weighted and aggregated according to the edge weights of the edges between the node and each neighbor node to obtain a neighbor aggregate vector. The neighbor aggregate vector is then fused with the base vector or job vector corresponding to the node to obtain the aggregated base vector or aggregated job vector. Based on the aggregated base vector and the aggregated job vector, the first information to be recommended and the corresponding first matching degree are determined.

5. The information recommendation method according to claim 4, characterized in that, The target user types also include cold start users; Before constructing the bipartite graph consisting of multiple nodes based on the user's basic characteristics and the job characteristics, the method further includes: Obtain the user's basic attribute information and the user's skill information; The user enhancement features are determined based on the basic attribute information and the skill information. The construction of a bipartite graph consisting of multiple nodes based on the user's basic characteristics and the job characteristics includes: A bipartite graph consisting of multiple nodes is constructed based on the user enhancement features and the job features.

6. The information recommendation method according to claim 1, characterized in that, Before determining the second recommended information and the corresponding second matching degree based on the similarity between the skill information and the skill information tags corresponding to the preset job cluster, the method further includes: Obtain job information for multiple job positions; For each of the aforementioned positions, keywords are extracted from the position information according to a preset keyword frequency calculation formula to obtain a keyword feature vector corresponding to each of the aforementioned positions; Clustering the feature vectors of multiple keywords yields multiple job clusters; For each job cluster, obtain the target job information for that job cluster, and determine the skill information tag corresponding to that job cluster based on the skill information in the target job information and preset filtering conditions.

7. The information recommendation method according to claim 6, characterized in that, The clustering of multiple keyword feature vectors yields multiple job clusters, including: Calculate the first distance between the keyword feature vectors; The keyword feature vectors whose first distance values ​​satisfy the preset distance conditions are merged to obtain the merged initial job clusters; Determine whether the initial job cluster meets the stopping condition; If it is determined that the initial job cluster does not meet the stopping condition, for each initial job cluster, the average vector of all keyword feature vectors in the initial job cluster is calculated, and the second distance value between multiple average vectors is calculated. The keyword feature vectors whose second distance values ​​satisfy the preset distance conditions are merged, the initial job cluster is updated, and the step of determining whether the initial job cluster meets the stopping condition is returned until the stopping condition is met, thus obtaining the multiple job clusters.

8. An information recommendation device, characterized in that, The device includes: a processor and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement the information recommendation method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer storage medium stores computer program instructions, which, when executed by a processor, implement the information recommendation method as described in any one of claims 1-7.

10. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the information recommendation method as described in any one of claims 1-7.

Images