A personalized job matching recommendation method and system based on knowledge graph

Abstract

The present invention discloses a method and system for personalized job matching based on a knowledge graph. The method collects job postings published by recruiting companies, processes the data, and extracts job triples through relationship extraction to obtain a job knowledge graph. The method also collects historical job search behavior data from users, including clicks, comments, and favorites, to obtain a user-preferred job data set. The user-preferred job data set is used as a seed, and the RippleNet algorithm is applied to extract multi-hop job information and its relationships from the job knowledge graph to obtain triplets of entity interactions between the user and the job knowledge graph. A multi-task recommendation model based on the knowledge graph and incorporating user preferences is constructed and trained, comprising a user-graph entity interaction module, a recommendation module, and a job-graph entity interaction module. A ranked job list is then derived based on the scores of user-job interactions. The present invention can improve recommendation results and provide personalized recommendations that meet user needs.

Description

Technical Field

[0001] The present invention relates to the field of information recommendation technology, and in particular to a method and system for recommending personalized job matching based on a knowledge graph. Background Art

[0002] Online recruitment has become the mainstream channel for job hunting and recruitment. The massive amount of job postings and recruitment information makes it difficult for recruiting companies and job seekers to find the right match. Recruiting companies need to sift through a vast amount of resumes to find candidates that meet their needs. At the same time, job seekers need to find personalized positions that meet their interests and hobbies within a vast pool of available positions. To meet this personalized recruitment need, job seekers' interests are identified by analyzing their search, browsing, and review behaviors. These positions are then recommended to the user. However, traditional recommendation systems suffer from cold start and data sparsity issues, and they ignore users' individual interests and preferences, which impacts the accuracy and interpretability of recommendation systems.

[0003] Introducing knowledge graphs (KGs) into recommendation systems (RSs) provides a new solution to the problems existing in traditional job recommendation systems. Knowledge graphs contain rich semantic information, which can solve the data sparsity problem of current recommendation systems to a certain extent. At the same time, the application of knowledge graph reasoning can supplement the auxiliary data of the recommendation system. By introducing the semantic information and attribute information of knowledge graph triples, users' personal interests and hobbies can be discovered at a deeper level, which can improve the accuracy and explainability of personalized recommendations. Recommendation systems based on knowledge graphs have become a popular branch of recommendation system research.

[0004] Collaborative filtering algorithms are commonly used in job recommendations. These recommendations contain massive amounts of user resumes and job positions. Users typically only click, browse, and rate a subset of these positions, resulting in sparse user and job rating matrices and hindering the calculation of similar user groups. If a new user has never clicked or rated a position in the system, or a newly added position has never been rated, then this new user or position will not be recommended in traditional recommendation systems. This is the data sparsity and cold start problem. A common approach to addressing sparsity and cold start issues is to introduce additional auxiliary information into the job recommendation algorithm, such as social network information, user or job attributes, and contextual information. This auxiliary information enriches the user and job descriptions in the system, complementing the sparsity of interactive information and enabling more personalized recommendations. Knowledge graphs, with their rich semantic information, can be incorporated into recommendation systems as auxiliary information, addressing sparsity and cold start issues and improving recommendation accuracy.

[0005] Existing recommendation algorithms often ignore users' behavioral preferences or do not consider the weight of users' preferences, resulting in unreasonable recommended items. Current recommendation algorithms that integrate knowledge graphs take into account the structure and semantic information of knowledge graphs, but often ignore the connection between users and knowledge graphs. Summary of the Invention

[0006] In response to the shortcomings of the existing technology, the present invention proposes a method and system for personalized job matching based on knowledge graphs. By introducing the user's preferences and interests, the job graph learning module and the recommendation module are trained simultaneously using a multi-task learning method, and the feature vectors of job seekers and recruitment positions are obtained. Finally, the obtained job seeker feature vectors and the recruitment position feature vectors are similarly calculated, and the calculation results are recommended to the user, thereby achieving personalized job matching.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] A method for recommending personalized job matching based on a knowledge graph, comprising the following steps:

[0009] Step 1: Collect job postings from recruiting companies, process the data, and extract job triples through relationships. Then, perform knowledge fusion to obtain a job knowledge graph.

[0010] Step 2: Collect the user's historical job search behavior data, including clicks, browsing, comments, and collections, to obtain the user's preferred job data set;

[0011] Step 3: Using the user's preferred job dataset as a seed, apply the RippleNet algorithm to obtain multi-hop job information and its relationships from the job knowledge graph, and obtain triples of the user's interaction with the entity of the job knowledge graph;

[0012] Step 4: Build and train a multi-task recommendation model based on the knowledge graph and integrating user preferences, including a user-graph entity interaction module, a recommendation module, and a position-graph entity interaction module;

[0013] Obtain the relationship between the user and the knowledge graph through the user-graph entity interaction module;

[0014] Through the position-graph entity interaction module, the similarity relationship between positions and graph entities is obtained;

[0015] Through the recommendation module, we can get the score of the interaction between users and positions;

[0016] Step 5: Based on the scores of the interactions between the user and the positions, a ranked list of positions recommended to the user is obtained.

[0017] Furthermore, in step 3, the triplet of the user's interaction with the entity of the job knowledge graph is represented as <h u ,r u ,t u >, where h u is the header entity, indicating the user; r u is a relation or attribute; u It is the tail entity;

[0018] Through the user-graph entity interaction module, the relationship between the user and the knowledge graph is obtained, specifically including:

[0019] In the user-graph entity interaction module, t u 、r u The information is extracted through deep neural network respectively, and t u ′、r u '; h u Perform cross-compression information extraction with the user set u in the job knowledge graph to obtain h u ' and u'; and r u ′ and h u ′ Input the deep neural network to get the predicted value of the tail entity Finally, the prediction value is evaluated using the similarity function prediction results.

[0020] Furthermore, the job position triple is represented as <h v ,r v ,t v >, where h v is the head entity, indicating the position; r v is the relationship from the head entity to the tail entity; v is the tail entity;

[0021] Through the position-graph entity interaction module, the similarity relationship between positions and graph entities is obtained, including:

[0022] In the position-graph entity interaction module, r v and t v Through deep neural network information extraction, we can get t v ″、r v '; h v Perform cross-compression information extraction with the position set v in the position knowledge graph to obtain h v ′ and v′; and r v ′ and h v ′ Input the deep neural network to get the predicted value of the tail entity Finally, the similarity function is used to evaluate the predicted value prediction results.

[0023] Furthermore, in the recommendation module, the user's preferred job data set is input into a bidirectional LSTM network based on the attention mechanism to obtain the user's interest preference vector; and the user's interest preference vector, u′, and v″ are fused to obtain the interaction probability between the user and all jobs.

[0024] Furthermore, after the multi-task recommendation model based on the knowledge graph and integrating user preferences is trained, the user feature vector, the position feature vector and the user's interest preference vector are saved.

[0025] Furthermore, the step 5 is implemented through the following sub-steps:

[0026] According to the user ID, the corresponding user feature vector and user interest vector are indexed; the similarity between the user and all positions is calculated using the recommendation formula after integrating the user's interest preferences; the similarities are sorted, and the top n positions with the maximum similarity are selected and recommended to the user.

[0027] Furthermore, the deep neural network uses a multi-layer perceptron.

[0028] A knowledge graph-based personalized job matching recommendation system, including a user-graph entity interaction module, a recommendation module, a job-graph entity interaction module, and a cross-compression module;

[0029] The cross-compression module includes a cross-compression unit 1 and a cross-compression unit 2; the cross-compression unit 1 is used to automatically learn the high-order interaction features between the user and the entities of the job knowledge graph; the cross-compression unit 2 is used to automatically learn the high-order interaction features between the user's preferred job and the job entities of the job knowledge graph;

[0030] The user-graph entity interaction module is used to learn the triple relationship between the user and the entity of the job knowledge graph;

[0031] The position-graph entity interaction module is used to learn the triple relationship between the user's preferred position and the position entity of the position knowledge graph;

[0032] The recommendation module is used to integrate the high-order interaction features between the user and the entities of the job knowledge graph, the high-order interaction features between the user's preferred positions and the job entities of the job knowledge graph, and the user's preferences to predict the interaction scores between the user and all positions in the job knowledge graph.

[0033] An electronic device, comprising:

[0034] one or more processors;

[0035] A storage device for storing one or more programs. When the one or more programs are executed by the electronic device, the electronic device implements the method for recommending personalized matching of people and jobs based on the knowledge graph.

[0036] A computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method for recommending personalized matching of people and jobs based on a knowledge graph.

[0037] The beneficial effects of the present invention are as follows:

[0038] (1) The present invention's personalized job matching recommendation method and system based on knowledge graph integrates user behavioral preferences and proposes a hybrid recommendation system model based on multi-task learning that integrates user preferences. It also considers the relationship between users and graphs, as well as the relationship between job entities and graph entities. It considers both the knowledge graph architecture and semantic information, as well as the connection between users and knowledge graphs, and effectively utilizes knowledge graph information to enhance recommendation performance.

[0039] (2) Based on the user's resume information and browsing, clicking, and evaluating behaviors on the job information they are interested in, a user's click-browsing job history dataset is established as a seed. The RippleNet algorithm is applied to extract the connection between users and the knowledge graph, and the connection between this connection and the knowledge graph is integrated into the recommendation task to improve the recommendation performance.

[0040] (3) The model simultaneously learns the semantic matching and recommendation tasks between users and knowledge graphs. Similarly, it can also learn the semantic matching and recommendation tasks between positions and knowledge graph entities. The recommendation module incorporates user interest preferences, allowing the recommendation system to learn user preference information, thereby improving recommendation results and providing personalized recommendations that meet user needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flowchart of the framework of the personalized job recommendation method based on knowledge graph.

[0042] Figure 2 A flowchart constructed for the job knowledge graph.

[0043] Figure 3 This is a partial schematic diagram of the job knowledge graph.

[0044] Figure 4 A schematic diagram of obtaining multi-hop job information and its relationships from the job knowledge graph using the Ripple algorithm.

[0045] Figure 5This is a schematic diagram of the knowledge graph-based multi-task personalized recommendation service model; the left picture is the user-graph entity interaction module; the middle picture is the recommendation module, and the right picture is the position-graph entity interaction module.

[0046] Figure 6 Cross Compression Unit - CC u and cross compression unit 2 CC v Schematic diagram of cross compression.

[0047] Figure 7 It is a bidirectional LSTM network architecture based on the attention mechanism. DETAILED DESCRIPTION

[0048] The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments, and the purpose and effects of the present invention will become more apparent. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0049] The knowledge graph-based personalized job matching recommendation method and system of the present invention introduces a knowledge graph to solve the sparsity of recommendation data and improve the accuracy of recommendation ranking. By constructing a job knowledge graph, information such as positions and attributes is integrated and aligned, and the semantic information and attribute information in the graph are used to explore the user's deep preferences; the different relationship link information in the knowledge graph is used to facilitate the divergence of recommendation results; at the same time, the job graph can connect the user's historical records and recommendation results to enhance the interpretability of job recommendations.

[0050] like Figure 1 As shown, the knowledge graph-based personalized job matching recommendation method of the present invention includes the following steps.

[0051] Step 1: Collect job postings published by recruiting companies, process the data, and extract job triples through relationships. Then, perform knowledge fusion to obtain a job knowledge graph.

[0052] like Figure 2 As shown in the figure, we first construct the ontology of the knowledge graph in the field of job recruitment, using a top-down approach. Then, we collect and process data based on job data from major job recruitment websites, perform named entity recognition and relationship extraction on the data based on the ontology, and form triples in the form of <entity, relationship, entity> or <entity, attribute, attribute value>, and perform knowledge fusion on heterogeneous data sources. Among them, the job entities released by enterprises mainly include the following 8 attributes: job title, work location, work experience, educational requirements, job description, company name, industry, job requirements, etc. The job seeker information is reflected in the form of a resume, mainly including: name, age, job search intention, education, work location, work experience, project experience, personal skills, etc. The recruitment position triple is represented as <hv ,r v ,t v >, where h v is the head entity, indicating the position; r v is the relationship from the head entity to the tail entity; v is the tail entity. The job knowledge graph is represented by G, and G={(h v ,r v ,t v )|h v ,t v ∈E,r v ∈R}. Where E and R represent the entity set and relationship set in the knowledge graph. Figure 3 This is a partial diagram of a job posting knowledge graph. The user set in the job posting knowledge graph is represented as U = u1, u2, ..., and the job posting set is represented as V = v1, v2, ....

[0053] Step 2: Collect the user's historical job search behavior data, including clicks, browsing, comments, and collections, to obtain a set of job data that the user prefers.

[0054] The user's preferred job data set is represented as v u , based on the job data set of user preferences, establish the interaction matrix Y between job seekers and job openings, Characterize the user's implicit feedback, defined as follows:

[0055]

[0056] in, =1 indicates that there is an invisible interaction between the job seeker u and the position v, such as click behavior, browsing behavior, collection behavior, resume submission behavior and evaluation behavior.

[0057] Step 3: Take the user's preferred job dataset as a seed, apply the RippleNet algorithm to obtain multi-hop job information and its relationships from the job knowledge graph, and obtain triples of the user's entity interaction with the job knowledge graph.

[0058] The triplet of the user's interaction with the entity of the job knowledge graph is represented as (user, relationship, entity). The process of applying the RippleNet algorithm to obtain multi-hop job information and its relationships from the job knowledge graph is as follows: taking the user's behavior data as the center of interest, simulate the user's interest spreading outward layer by layer on the knowledge graph, and the process of continuous attenuation is similar to ripples in water. Finally, the obtained RippleNet (user, relationship, graph entity) triplet is used in training to construct a more comprehensive user representation vector. The propagation process of RippleNet is as follows: Figure 4As shown. The final triplet of the interaction between the user and the entity of the job knowledge graph is represented as <h u ,r u ,t u >, where h u is the header entity, indicating the user; r u is a relation or attribute; u Is the tail entity.

[0059] Step 4: Build and train a multi-task recommendation model based on the knowledge graph and integrating user preferences, including a user-graph entity interaction module, a recommendation module, and a position-graph entity interaction module; obtain the relationship between the user and the knowledge graph through the user-graph entity interaction module; obtain the similarity relationship between the position and the graph entity through the position-graph entity interaction module; and obtain the score of the interaction between the user and the position through the recommendation module.

[0060] The structure of the model is as follows Figure 5 As shown in the figure, the model uses multi-task learning to train the knowledge graph module and the recommendation system module simultaneously, fully considering the user-job graph entity interaction and job-graph entity interaction, modeling user representation and job entity representation from these two aspects, and using them as the final input of the recommendation algorithm. At the same time, a bidirectional LSTM (Long Short-Term Memory) network with an attention mechanism extracts user preferences from the user's historical behavioral interaction series into the recommendation model, thereby giving personalized person-job matching recommendation results.

[0061] To facilitate the introduction of each module, the multi-layer perceptron is used as a representative of the deep neural network.

[0062] 1. Cross Compression Module (CC u ,C v )

[0063] The cross compression module includes a cross compression unit 1 and a cross compression unit 2, wherein the cross compression unit 1 CC u Used to automatically learn the high-order interaction features between users and entities in the job knowledge graph; the cross compression unit CC v Used to automatically learn high-order interaction features between user preferred positions and position entities in the position knowledge graph.

[0064] The cross compression unit is proposed by the MKR (Multi-Task Feature Learning for KnowledgeGraph Enhanced Recommendation) model, which automatically learns the high-order feature interaction features between items and entities in the KG in the recommendation system. Its structure is as follows Figure 6 shown.

[0065] Here we use cross compression unit 2 CC v As an example, a cross compression unit CC u Similar to it.

[0066] For a job position v and an entity h corresponding to the job position v in the knowledge graph, first construct the cross matrix Where d is the dimension of v and h.

[0067]

[0068] Recruitment position v and entity h through cross compression unit 2 CC v The output is:

[0069]

[0070] in, is the cross unit weight and bias parameter. Adjusting this parameter can simultaneously learn the two tasks of knowledge graph semantic matching and recommendation. u The same is true for the calculation of the feature vector of user u.

[0071] 2. Position-graph entity interaction module

[0072] The job-graph entity interaction module uses a deep semantic matching mechanism to convert the job position triples <h v ,r v ,t v > v and t v Through deep neural network information extraction, we can get t v ″、r v ″; v Perform cross-compression information extraction with the position set v in the position knowledge graph to obtain h v ′ and v′; and r v ″ and h v Input the deep neural network to get the predicted value of the tail entity Finally, the similarity function is used to evaluate the predicted value prediction results.

[0073] h′ v =CC v (…CC v (v,h v ))[h]

[0074] r′ v =MLP(…MLP(r v ))

[0075] t′ v =MLP(…MLP(t v ))

[0076]

[0077] Among them, CC v stands for compressed cross unit 2, MLP is a multi-layer perceptron, MLP(x) = σ(wx+b), where w is the weight parameter and σ is the nonlinear activation sigmoid function. The similarity function is:

[0078]

[0079] 3. User-graph entity interaction module

[0080] The user-graph entity interaction module learns the relationship between users and knowledge graph entities. u 、r u The information is extracted through deep neural network respectively, and t u ′、r u '; h u Perform cross-compression information extraction with the user set u in the job knowledge graph to obtain h u ' and u'; and r u ′ and h u ′ Input the deep neural network to get the predicted value of the tail entity Finally, the similarity function is used to evaluate the predicted value prediction results.

[0081] h′ u =CC u (…CC u (u,h u ))[h]

[0082] r′ u =MLP(…MLP(r u ))

[0083] t′ u =MLP(…MLP(t u ))

[0084]

[0085] Among them, CC u stands for compressed cross unit, MLP is multi-layer perceptron, MLP(x) = σ(wx+b), where w is the weight parameter and σ is the nonlinear activation sigmoid function. The similarity function is:

[0086]

[0087] 4. Recommendation module

[0088] (1) User interest preference acquisition

[0089] In the recommendation module, a bidirectional LSTM network plus an attention mechanism is used to optimize the user preference learning ability. The network input is the user job search process interaction behavior data series (seq uv LSTM can solve problems such as long-term memory and gradient in back propagation. It uses a gating mechanism to control input, memory and other information and make predictions at the current time step. It can extract user preferences from the user and position interaction sequence, and use the attention mechanism to adjust the weight of each position, so that users' interest preferences for positions can be extracted more accurately. The network structure is as follows: Figure 7 The calculation formula is as follows:

[0090] h out =BILSTM(seq uv )

[0091]

[0092]

[0093] α=sofrmax(w T tanh(c n ,h out ))

[0094] r=h out α T

[0095] uip=MLP(r)

[0096] Among them, h out is the result of the bidirectional LSTM output hidden layer, α nj h n ,h j The attention weight, w α and w T is the parameter matrix, α is the attention weight, r is the LSTM output after adding the weight, and then the user interest preference (UIP) is obtained through MLP. The user interest preference is abbreviated as: UIP = AttentionLSTM (seq uv ).

[0097] (2) Recommended

[0098] The input of the recommendation module is the user interaction matrix Y and the job seeker historical behavior data series (User interaction term sequence) denoted as Sequv The recommendation module inputs the user's preferred job data set into a bidirectional LSTM network based on the attention mechanism to obtain the user's interest preference vector; and fuses the user's interest preference vector, u', and v' to obtain the interaction probability between the user and all jobs. Among them, given the user feature vector u, after the cross compression unit CC u After MLP processing, u′ is obtained as:

[0099] u cc =CC u (u,h u )[u]

[0100] u′=MLP(MLP(…MLP(u cc )))

[0101] Similarly, the processed v is:

[0102] v cc =CC v (v,h v )[v]

[0103] v′=MLP(MLP(…MLP(v cc )))

[0104] The recommendation formula after integrating user interest preferences is:

[0105]

[0106] Among them, λ is the weight of user interest preference, AttentionLSTM(seq uv ) is the user's interest preference.

[0107] The loss function of the multi-task recommendation model based on knowledge graph and integrating user preferences is:

[0108]

[0109] Where θ is the cross entropy loss function, L rs is the loss of the recommendation module, is the loss value of fitting the (user, relationship, entity) triplet of the user and job graph entity interaction, is the loss value of the knowledge graph entity and entity interaction (position, relationship, entity) triple fitting, λ is the regularization coefficient, is the regularization term.

[0110] After the model is trained, it saves the user feature vector, the position feature vector, and the user's interest preference vector.

[0111] Step 5: Based on the scores of the interactions between the user and the positions, a ranked list of positions recommended to the user is obtained.

[0112] In this step, the corresponding user feature vector and user interest vector are indexed according to the user ID; the similarity between the user and all positions is calculated using the recommendation formula after integrating the user's interest preferences; the similarities are sorted, and the top n positions with the maximum similarity are selected and recommended to the user.

[0113] In order to realize the above-mentioned knowledge graph-based personalized matching recommendation method, the present invention also discloses a knowledge graph-based personalized matching recommendation system for people and jobs, including a user-graph entity interaction module, a recommendation module, a job-graph entity interaction module, and a cross-compression module;

[0114] The cross-compression module includes a cross-compression unit 1 and a cross-compression unit 2; the cross-compression unit 1 is used to automatically learn the high-order interaction features between the user and the entities of the job knowledge graph; the cross-compression unit 2 is used to automatically learn the high-order interaction features between the user's preferred job and the job entities of the job knowledge graph;

[0115] The user-graph entity interaction module is used to learn the triple relationship between the user and the entity of the job knowledge graph;

[0116] The position-graph entity interaction module is used to learn the triple relationship between the user's preferred position and the position entity of the position knowledge graph;

[0117] The recommendation module is used to integrate the high-order interaction features between the user and the entities of the job knowledge graph, the high-order interaction features between the user's preferred positions and the job entities of the job knowledge graph, and the user's preferences to predict the interaction scores between the user and all positions in the job knowledge graph.

[0118] The embodiment of the knowledge graph-based personalized matching recommendation system for people and jobs of the present invention can be applied to any device with data processing capabilities, and the any device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of any device with data processing capabilities where it is located reading the corresponding computer program instructions in the non-volatile memory into the memory and running them. From the hardware level, in addition to the processor, memory, network interface, and non-volatile memory, the any device with data processing capabilities where the device in the embodiment is located can also include other hardware according to the actual functions of the any device with data processing capabilities, which will not be described in detail.

[0119] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0120] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present invention. A person of ordinary skill in the art can understand and implement the present invention without inventive work.

[0121] An embodiment of the present invention also provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the method for recommending personalized matching of people and jobs based on knowledge graphs in the above embodiment is implemented.

[0122] The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the aforementioned embodiments, such as a hard disk or memory. The computer-readable storage medium may also be an external storage device, such as a plug-in hard disk, a smart memory card (SmartMedia card, SMC), an SD card, a flash card (Flash card), etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capabilities. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and may also be used to temporarily store data that has been output or is to be output.

[0123] Those skilled in the art will understand that the foregoing descriptions are merely preferred embodiments of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art will still be able to modify the technical solutions described in the foregoing examples or substitute equivalents for some of the technical features therein. Any modifications, equivalent substitutions, etc. made within the spirit and principles of the invention shall be included within the scope of protection of the invention.

Claims

1. A personalized job matching recommendation method based on knowledge graph, characterized by: The method comprises the following steps: Step 1: Collect job postings from recruiting companies, process the data, and extract job triples through relationships. Then, perform knowledge fusion to obtain a job knowledge graph. Step 2: Collect the user's historical job search behavior data, including clicks, browsing, comments, and collections, to obtain the user's preferred job data set; Step 3: Using the user's preferred job dataset as a seed, apply the RippleNet algorithm to obtain multi-hop job information and its relationships from the job knowledge graph, and obtain triples of the user's interaction with the entity of the job knowledge graph; Step 4: Build and train a multi-task recommendation model based on the knowledge graph and integrating user preferences, including a user-graph entity interaction module, a recommendation module, and a position-graph entity interaction module; Obtain the relationship between the user and the knowledge graph through the user-graph entity interaction module; Through the position-graph entity interaction module, the similarity relationship between positions and graph entities is obtained; Through the recommendation module, we can get the score of the interaction between users and positions; Step 5: Based on the scores of the interactions between the user and the positions, a ranked list of positions recommended to the user is obtained; In step 3, the triplet of the user's interaction with the entity of the job knowledge graph is represented as <h u ,r u ,t u >, where h u is the header entity, indicating the user; r u is a relation or attribute; u It is the tail entity; Through the user-graph entity interaction module, the relationship between the user and the knowledge graph is obtained, specifically including: In the user-graph entity interaction module, t u 、r u The information is extracted through deep neural network respectively, and t u ´、r u ´; will h u Perform cross-compression information extraction with the user set u in the job knowledge graph to obtain h u ´ and u´; and r u ´ and h u Input the deep neural network to get the predicted value of the tail entity , and finally use the similarity function to evaluate the predicted value The prediction results; The job position triple is represented as <h v ,r v ,t v >, where h v is the head entity, indicating the position; r v is the relationship from the head entity to the tail entity; v is the tail entity; Through the position-graph entity interaction module, the similarity relationship between positions and graph entities is obtained, including: In the position-graph entity interaction module, r v and t v Through deep neural network information extraction, we can get t v ´、r v ´; will h v Perform cross-compression information extraction with the position set v in the position knowledge graph to obtain h v ´ and v´; and r v ´ and h v Input the deep neural network to get the predicted value of the tail entity , and finally use the similarity function to evaluate the predicted value The prediction results; In the recommendation module, the user's preferred job data set is input into a bidirectional LSTM network based on the attention mechanism to obtain the user's interest preference vector; and the user's interest preference vector, u', and v' are fused to obtain the interaction probability between the user and all jobs. .

2. The method for recommending personalized job matching based on knowledge graph according to claim 1 is characterized in that: After the multi-task recommendation model based on the knowledge graph and integrating user preferences is trained, the user feature vector, position feature vector and user interest preference vector are saved.

3. The method for recommending personalized job matching based on knowledge graph according to claim 2 is characterized in that: The step 5 is implemented by the following sub-steps: According to the user ID, the corresponding user feature vector and user interest vector are indexed; the similarity between the user and all positions is calculated using the recommendation formula after integrating the user's interest preferences; the similarities are sorted, and the top n positions with the maximum similarity are selected and recommended to the user.

4. The method for recommending personalized job matching based on knowledge graph according to claim 1, characterized in that: The deep neural network uses a multi-layer perceptron.

5. A personalized job matching recommendation system based on knowledge graph, characterized by: The system is used to implement the knowledge graph-based personalized job matching recommendation method according to claim 1, and the system includes a user-graph entity interaction module, a recommendation module, a job-graph entity interaction module, and a cross-compression module; The cross-compression module includes a cross-compression unit 1 and a cross-compression unit 2; the cross-compression unit 1 is used to automatically learn the high-order interaction features between the user and the entities of the job knowledge graph; the cross-compression unit 2 is used to automatically learn the high-order interaction features between the user's preferred job and the job entities of the job knowledge graph; The user-graph entity interaction module is used to learn the triple relationship between the user and the entity of the job knowledge graph; The position-graph entity interaction module is used to learn the triple relationship between the user's preferred position and the position entity of the position knowledge graph; The recommendation module is used to integrate the high-order interaction features between the user and the entities of the job knowledge graph, the high-order interaction features between the user's preferred positions and the job entities of the job knowledge graph, and the user's preferences to predict the interaction scores between the user and all positions in the job knowledge graph.

6. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs, which, when executed by the electronic device, enables the electronic device to implement the knowledge graph-based personalized job matching recommendation method as described in any one of claims 1 to 4.

7. A computer-readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for recommending personalized matching of people and jobs based on a knowledge graph as described in any one of claims 1 to 4 is implemented.

Images