CN116127186A - Knowledge-graph-based individual matching recommendation method and system for person sentry - Google Patents
Knowledge-graph-based individual matching recommendation method and system for person sentry Download PDFInfo
- Publication number
- CN116127186A CN116127186A CN202211583266.4A CN202211583266A CN116127186A CN 116127186 A CN116127186 A CN 116127186A CN 202211583266 A CN202211583266 A CN 202211583266A CN 116127186 A CN116127186 A CN 116127186A
- Authority
- CN
- China
- Prior art keywords
- user
- entity
- graph
- job
- knowledge
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2457—Query processing with adaptation to user needs
- G06F16/24578—Query processing with adaptation to user needs using ranking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/28—Databases characterised by their database models, e.g. relational or object models
- G06F16/284—Relational databases
- G06F16/288—Entity relationship models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/10—Office automation; Time management
- G06Q10/105—Human resources
- G06Q10/1053—Employment or hiring
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Business, Economics & Management (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Resources & Organizations (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- Pure & Applied Mathematics (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Economics (AREA)
- Algebra (AREA)
- Animal Behavior & Ethology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a personalized matching recommendation method and system for a person post based on a knowledge graph, wherein the method collects recruitment position information issued by a recruitment enterprise, performs data processing, and obtains a recruitment position triplet through relation extraction to obtain a position knowledge graph; collecting historical job hunting behavior data of a user including click browsing, comment and collection, and obtaining a job data set preferred by the user; taking a job data set preferred by a user as a seed, and obtaining multi-hop job information and a relation thereof from the job knowledge graph by applying a RippleNet algorithm to obtain a triplet of entity interaction between the user and the job knowledge graph; the method comprises the steps of constructing and training a multitask recommendation model based on a knowledge graph and fusing user preferences, wherein the multitask recommendation model comprises a user-graph entity interaction module, a recommendation module and a position-graph entity interaction module; and obtaining a job ranking list according to the scores of the interactions of the user and the job. The invention can improve the recommendation result and give the recommendation requirement meeting the individuation of the user.
Description
Technical Field
The invention relates to the technical field of information recommendation, in particular to a personalized matching recommendation method and system for person posts based on a knowledge graph.
Background
At present, online recruitment has become a main channel for job hunting and recruitment, and massive job hunting information and recruitment information make it difficult for recruiters and recruiters to select matching. Recruitment enterprises need to find job seekers meeting enterprise demands from a large number of resume information of job seekers, and meanwhile, job seekers need to find personalized positions meeting own interests from a large number of recruitment positions to match. In order to meet personalized job demands, the user's interest in job-seeking positions is found by analyzing the actions of searching, browsing, evaluating and the like of job seekers, and then the position information of interest of the job seekers is recommended to the user for display. However, the conventional recommendation system has the problems of cold start and sparse data, and personalized interest preference of the user is ignored, so that the accuracy and the interpretability of the recommendation system are affected.
The Knowledge Graph (KG) is introduced into the recommendation system (Recommend Systems, RS) to provide a new solution for solving the problems of the traditional job-seeking recommendation system, the Knowledge Graph contains abundant semantic information, the problem of data sparsity of the current recommendation system can be solved to a certain extent, meanwhile, the auxiliary data of the recommendation system can be supplemented by the reasoning of the application Knowledge Graph, and the personal interest of the user can be found in a deeper level by introducing the Knowledge Graph triplet semantic information and attribute information, so that the personalized recommendation accuracy and the interpretability can be improved, and the recommendation system based on the Knowledge Graph becomes the popular branch field of the recommendation system in the research direction.
In the field of job recommendation, a collaborative filtering algorithm is generally adopted, recommendation data comprise massive user resume and job data, and users only click, browse and evaluate part of the job data, so that a user and job scoring matrix is sparse, and calculation of similar user groups is affected; if a new user never clicks and evaluates a position in the system, or a newly added position is never evaluated, then the new user or position is not recommended in the traditional recommendation system, which is a data sparsity and cold start problem. One common idea for solving the sparsity and cold start problems is to additionally introduce some auxiliary information as input, such as social network information, attributes of users or positions, context information, etc., into the position recommendation algorithm. The auxiliary information can enrich the information description of the user and the position in the system, well supplement the sparseness of the interaction information in the system, and enable the recommendation effect to be more personalized. The knowledge graph has rich semantic information, can be used as auxiliary information to be introduced into a recommendation system, solves the problems of sparsity and cold start, and improves the accuracy of recommendation.
The existing recommendation algorithm often ignores the behavior preference of the user or does not consider the weight occupied by the preference of the user, so that the recommended articles are unreasonable, and the current recommendation algorithm fused with the knowledge graph considers the structure and semantic information of the knowledge graph, but often ignores the connection between the user and the knowledge graph.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a personalized matching recommendation method and system based on a knowledge graph post.
The aim of the invention is achieved by the following technical scheme:
a personalized matching recommendation method based on knowledge graph person post includes the following steps:
step one: collecting recruitment information issued by a recruitment enterprise, performing data processing, extracting through a relationship to obtain a recruitment triad, and performing knowledge fusion to obtain a job knowledge graph;
step two: collecting historical job hunting behavior data of a user including click browsing, comment and collection, and obtaining a job data set preferred by the user;
step three: taking the position data set preferred by the user as a seed, and applying a RippleNet algorithm to acquire multi-hop position information and the relation thereof from the position knowledge graph to acquire a triplet of entity interaction between the user and the position knowledge graph;
step four: the method comprises the steps of constructing and training a multitask recommendation model based on a knowledge graph and fusing user preferences, wherein the multitask recommendation model comprises a user-graph entity interaction module, a recommendation module and a position-graph entity interaction module;
obtaining a relationship between a user and the knowledge graph through a user-graph entity interaction module;
obtaining a similarity relation between the position and the map entity through a position-map entity interaction module;
obtaining the score of the interaction between the user and the position through a recommendation module;
step five: and obtaining a job ranking list recommended to the user according to the scores of the user and the job interactions.
Further, in the third step, the triples of the user interaction with the entity of the job position knowledge graph are expressed as<h u ,r u ,t u >Wherein h is u Is a header entity representing a user; r is (r) u Is a relationship or attribute; t is t u Is a tail entity;
obtaining a relationship between a user and the knowledge graph through a user-graph entity interaction module, wherein the relationship comprises the following steps:
in the user-map entity interaction module, t is set u 、r u Respectively extracting information through a deep neural network to respectively obtain t u ′、r u 'A'; will h u Cross-compressing information extraction is carried out on the user set u in the job position knowledge graph, and h is obtained respectively u 'and u'; and r is set to u ' and h u ' input deep neural network, obtain predicted value of tail entity
Finally, the predicted value is evaluated by using a similarity function>
Is a predicted result of (a).
Further, the recruitment position triad is represented as<h v ,r v ,t v >Wherein h is v A head entity representing a position; r is (r) v Is the relationship from the head entity to the tail entity; t is t v Is a tail entity;
obtaining a similarity relation between the position and the map entity through a position-map entity interaction module, wherein the similarity relation comprises the following specific steps:
in the position-map entity interaction module, r is calculated v And t v Respectively extracting information through a deep neural network to respectively obtain t v ″、r v 'A'; will h v Performing cross compression information extraction with the position set v in the position knowledge graph to obtain h respectively v 'and v'; and r is set to v ' and h v ' input deep neural network, obtain predicted value of tail entity
Finally, the predicted value is evaluated by using a similarity function>
Is a predicted result of (a).
Further, in a recommendation module, inputting the job data set of the user preference into a bidirectional LSTM network based on an attention mechanism to obtain a user interest preference vector; and fusing the user interest preference vectors, u ', v'; obtaining the interaction probability of the user and all positions
Further, after the training of the multitask recommendation model based on the knowledge graph and fused with the user preference is completed, the user feature vector, the feature vector of the position and the interest preference vector of the user are stored.
Further, the fifth step is realized by the following sub-steps:
indexing corresponding user feature vectors and user interest vectors according to the user ID; calculating the similarity between the user and all positions by using a recommendation formula after fusing the user interest preference; and sorting the similarity, and selecting n positions before the maximum similarity to recommend to the user.
Further, the deep neural network adopts a multi-layer perceptron.
A personalized matching recommendation system based on knowledge graph person post comprises a user-graph entity interaction module, a recommendation module, a position-graph entity interaction module and a cross compression module;
the cross compression module comprises a cross compression unit I and a cross compression unit II; the cross compression unit is used for automatically learning high-order interaction characteristics between a user and the entity of the job position knowledge graph; the second cross compression unit is used for automatically learning high-order interaction characteristics between positions preferred by the user and position entities of the position knowledge graph;
the user-map entity interaction module is used for learning a triplet relation between a user and the entity of the job position knowledge map;
the position-map entity interaction module is used for learning a triplet relation between positions preferred by the user and position entities of the position knowledge map;
the recommendation module is used for fusing high-order interaction characteristics between the user and the entity of the job position knowledge graph, high-order interaction characteristics between the job position preferred by the user and the job position entity of the job position knowledge graph, and user preference, and predicting interaction scores between the user and all job positions of the job position knowledge graph.
An electronic device, comprising:
one or more processors;
and the storage device is used for storing one or more programs, and when the one or more programs are executed by the electronic equipment, the electronic equipment realizes the personalized matching recommendation method based on the knowledge graph.
A computer readable medium having stored thereon a computer program which when executed by a processor implements the knowledge-graph-based person post personalized matching recommendation method.
The beneficial effects of the invention are as follows:
(1) According to the personalized matching recommendation method and system based on the knowledge graph and the sentry, behavior preference of the user is fused, a mixed recommendation system model based on the knowledge graph and the multitask learning of the user preference is provided, meanwhile, the relationship between the user and the graph and the relationship between the position entity and the graph entity are considered, the knowledge graph architecture and semantic information are considered, the relationship between the user and the knowledge graph is also considered, and the knowledge graph information is effectively utilized to enhance recommendation performance;
(2) According to the resume information of the user and the position information of interest, browsing, clicking, evaluating and other actions, establishing a history data set of the user clicking, browsing the position as a seed, extracting the contact between the user and the knowledge graph by using a RippleNet algorithm, and fusing the contact and the contact of the knowledge graph into a recommendation task to improve the recommendation performance.
(3) The model learns the two tasks of matching and recommending the user and the knowledge graph semantics simultaneously, and can learn the two tasks of matching and recommending the position and the knowledge graph entity semantics. The recommendation module adds interest preference factors of the user, so that the recommendation system learns the user preference information, thereby improving the recommendation result and giving out recommendation demands meeting the individuation of the user.
Drawings
Fig. 1 is a schematic flow diagram of a personalized position recommendation method framework based on a knowledge graph.
Fig. 2 is a flowchart of job knowledge graph construction.
Fig. 3 is a schematic diagram of part of a job knowledge graph.
Fig. 4 is a schematic diagram of acquiring multi-hop job information and its relationship from a job knowledge graph through a happle algorithm.
FIG. 5 is a schematic diagram of a knowledge-graph-based multitasking personalized recommendation service model; wherein, the left graph is a user-graph entity interaction module; the middle diagram is a recommendation module, and the right diagram is a position-map entity interaction module.
FIG. 6 shows a cross compression unit CC u And cross compression unit two CC v Schematic diagram of cross compression.
Fig. 7 is a bi-directional LSTM network architecture based on an attention mechanism.
Detailed Description
The objects and effects of the present invention will become more apparent from the following detailed description of the preferred embodiments and the accompanying drawings, it being understood that the specific embodiments described herein are merely illustrative of the invention and not limiting thereof.
According to the knowledge-graph-based individual matching recommendation method and system for the sentry, the knowledge graph is introduced to solve the sparsity of recommendation data, the accuracy of recommendation ordering is improved, the knowledge graph is constructed, information such as the sentry and the attribute is fused and aligned, and the deep preference of a user is mined by utilizing semantic information and attribute information in the graph; different relation link information in the knowledge graph is utilized, so that the divergence of the recommendation result is facilitated; meanwhile, the post map can be connected with the history record and the recommendation result of the user, so that the interpretation of the post recommendation is enhanced.
As shown in FIG. 1, the personalized matching recommendation method based on the knowledge graph sentry comprises the following steps.
Step one: collecting recruitment information issued by recruitment enterprises, performing data processing, extracting through the relationship to obtain recruitment triples, and performing knowledge fusion to obtain a position knowledge graph.
As shown in FIG. 2, an ontology of a knowledge graph in the field of human post recruitment is first constructed, a top-down method is adopted, then data acquisition and processing are carried out according to position data from each large job-seeking recruitment website, and named entity identification and relation extraction are carried out according to the ontology, so that a triplet form is formed<Entity, relationship, entity>Or (b)<Entity, attribute value>Knowledge fusion is performed on heterogeneous data sources. The job entity published by the enterprise mainly comprises the following 8 attributes: job title, job site, working experience, academic requirements, job description, company name, industry, job requirements, etc. The job-seeking user information is embodied in a resume form and mainly comprises: name, age, intent to search for job, academic, job site, work experience, project experience, personal skills, etc. The recruitment position triad is represented as<h v ,r v ,t v >Wherein h is v A head entity representing a position; r is (r) v From head entity to tail entityA relationship; t is t v Is a tail entity. The job position knowledge graph is represented by G, and g= { (h) v ,r v ,t v )|h v ,t v ∈E,r v E R }. Wherein E and R represent entity sets and relationship sets in the knowledge-graph. Fig. 3 is a partial schematic view of one of the recruitment position knowledge maps. User set in job position knowledge graph is expressed as u=u 1 ,u 2 ,. the recruitment position set is denoted v=v 1 ,v 2 ,...。
Step two: and collecting historical job hunting behavior data of the user including click browsing, comment and collection, and obtaining a job position data set preferred by the user.
The set of user-preferred job position data is denoted as v u Based on the user's preferred job data set, an interaction matrix Y of job-seeking users and recruitment positions is established,
implicit feedback characterizing a user is defined as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
=1 indicates that there is a stealth interaction between job seeker u and job v, such as click behavior, browse behavior, collection behavior, delivery resume behavior, evaluation behavior, and the like.
Step three: and taking the position data set preferred by the user as a seed, and applying a RippleNet algorithm to acquire multi-hop position information and the relation thereof from the position knowledge graph to obtain a triplet of entity interaction between the user and the position knowledge graph.
The triples of user interactions with the entities of the job knowledge graph are expressed as (user, relationship, entity). The process of acquiring multi-hop position information and the relation thereof from the position knowledge graph by applying the RippleNet algorithm is as follows: simulating a user by taking behavior data of the user as an interest centerThe interests are outwards diffused layer by layer on the knowledge graph, and the process is continuously attenuated, like the ripple in water. And finally, participating the obtained RippleNet (user, relation, atlas entity) triples in training, and constructing a more comprehensive user representation vector. The propagation process of RippleNet is shown in FIG. 4. The finally obtained triples of the entity interaction between the user and the position knowledge graph are expressed as<h u ,r u ,t u >Wherein h is u Is a header entity representing a user; r is (r) u Is a relationship or attribute; t is t u Is the tail entity.
Step four: the method comprises the steps of constructing and training a multitask recommendation model based on a knowledge graph and fusing user preferences, wherein the multitask recommendation model comprises a user-graph entity interaction module, a recommendation module and a position-graph entity interaction module; obtaining a relationship between a user and the knowledge graph through a user-graph entity interaction module; obtaining a similarity relation between the position and the map entity through a position-map entity interaction module; and obtaining the score of the interaction between the user and the position through a recommendation module.
The structure of the model is shown in fig. 5, the model adopts multitask learning to simultaneously train and learn a knowledge graph module and a recommendation system module, fully considers user-position graph entity interaction and position-graph entity interaction, models user representation and position entity representation from the two aspects, takes the user representation and the position entity representation as final input of a recommendation algorithm, and simultaneously extracts user preferences from a historical behavior interaction series of a user to the recommendation model by a bidirectional LSTM (Long Short-Term Memory) network added with an attention mechanism, thereby providing personalized person post matching recommendation results.
In the following, for convenience of description, the deep neural network selects a multi-layer perceptron as a representative to describe.
1. Cross compression module (CC) u ,C v )
The cross compression module comprises a cross compression unit I and a cross compression unit II, wherein the cross compression unit I CC u For automatically learning high-order interaction features between a user and an entity of the job site knowledge graph; the cross compression unit is two CCs v Position and position for automatically learning user preferencesAnd the higher-order interaction characteristics among the position entities of the position knowledge graph.
The cross compression unit adopts a MKR (Multi-Task Feature Learning for Knowledge Graph Enhanced Recommendation) model to propose that the cross compression unit automatically learns high-order characteristic interaction characteristics between items in a recommendation system and entities in KG, and the structure is shown in figure 6.
Here by cross compression of units two CCs v To give an example, a cross compression unit CC u Similar thereto.
For a recruitment position v and an entity h corresponding to the position v in a knowledge graph, firstly constructing a cross matrix
Where d is the dimension of v and h.
Recruitment position v and entity h pass through cross compression unit two CC v The outputs of (2) are:
wherein, the liquid crystal display device comprises a liquid crystal display device,
is the cross-cell weight and bias parameter. Adjusting the parameters can learn the two tasks of knowledge graph semantic matching and recommending at the same time. Also for CC u As does the feature vector calculation for user u.
2. Position-map entity interaction module
The job position-map entity interaction module adopts a deep semantic matching mechanism to triage recruitment positions<h v ,r v ,t v >R in (2) v And t v Respectively extracting information through a deep neural network to respectively obtain t v ″、r v "C"; will h v And the job is presentedPerforming cross compression information extraction on a position set v in the position knowledge graph to obtain h respectively v 'and v'; and r is set to v "AND h v "input deep neural network to obtain predicted value of tail entity
Finally, the predicted value is evaluated by using a similarity function>
Is a predicted result of (a).
h′ v ＝CC v (…CC v (v,h v ))[h]
r′ v ＝MLP(…MLP(r v ))
t′ v ＝MLP(…MLP(t v ))
Wherein CC v Representing a compressed cross unit two, MLP is a multi-layer perceptron, MLP (x) =σ (wx+b), where w is a weight parameter and σ is a nonlinear activation sigmoid function. The similarity function is:
3. user-map entity interaction module
The user-map entity interaction module learns the relationship between the user and the knowledge map entity by using t u 、r u Respectively extracting information through a deep neural network to respectively obtain t u ′、r u 'A'; will h u Cross-compressing information extraction is carried out on the user set u in the job position knowledge graph, and h is obtained respectively u 'and u'; and r is set to u ' and h u ' input deep neural network, obtain predicted value of tail entity
Finally, the predicted value is evaluated by using a similarity function>
Is a predicted result of (a).
h′ u ＝CC u (…CC u (u,h u ))[h]
r′ u ＝MLP(…MLP(r u ))
t′ u ＝MLP(…MLP(t u ))
Wherein CC u Representing the compressed cross unit, 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:
4. recommendation module
(1) User interest preference acquisition
In the recommendation module, a bidirectional LSTM network plus a attentiveness mechanism is used to optimize user preference learning capabilities, network inputs interact behavior data series (seq uv ). LSTM solves the problems of long-term memory and gradient in back propagation, and adopts gating mechanism to control information such as input, memory and the like and make predictions in the current time step. The preference of the user can be extracted from the interaction sequence of the user and the positions, and the weight of each position is adjusted by using the attention mechanism, so that the interest preference of the user on the positions can be extracted more accurately, and the network structure is shown in fig. 7. The calculation formula is as follows:
h out ＝BILSTM(seq uv )
α＝sofrmax(w T tanh(c n ,h out ))
r＝h out α T
uip＝MLP(r)
wherein h is out Outputting the result of the hidden layer for the bi-directional LSTM, alpha nj Is h n ,h j Attention weight, w α And w T And alpha is attention weight, r is LSTM output after adding weight, and the UIP of the user interest preference (user interest preference) is obtained through MLP. The acquisition of user interest preferences is abbreviated as: uip=attention lstm (seq uv )。
(2) Recommendation
The input of the recommendation module is a user interaction matrix Y and the historical behavior data series (User interact term sequence) of the job-seeking user are recorded as Seq uv . The recommendation module inputs the job position data set preferred by the user into a bidirectional LSTM network based on an attention mechanism to obtain a user interest preference vector; and fusing the user interest preference vectors, u ', v'; obtaining the interaction probability of the user and all positions
Wherein, given user characteristic vector u, a CC is passed through cross compression unit u And u' obtained after MLP treatment is:
u cc ＝CC u (u,h u )[u]
u′＝MLP(MLP(…MLP(u cc )))
similarly, v after treatment is:
v cc ＝CC v (v,h v )[v]
v′＝MLP(MLP(…MLP(v cc )))
the recommendation formula after fusing the user interest preference is as follows:
where λ is the weight of the user interest preferences, attenionLSTM (seq uv ) Is an interest preference of the user.
The loss function of the multitask recommendation model based on the knowledge graph and fused with the user preference is as follows:
where θ is the cross entropy loss function, L rs In order to recommend the loss of the module,
loss value of triplet fit for user and job map entity interactions (user, relationship, entity), ->
Loss values fitted for the knowledge-graph entity-entity interaction (position, relation, entity) triples, lambda being regularization coefficient,>
is a regularization term.
After the model is trained, the user feature vector, the feature vector of the position and the interest preference vector of the user are stored.
Step five: and obtaining a job ranking list recommended to the user according to the scores of the user and the job interactions.
In this step, corresponding user feature vectors and user interest vectors are indexed according to the user ID; calculating the similarity between the user and all positions by using a recommendation formula after fusing the user interest preference; and sorting the similarity, and selecting n positions before the maximum similarity to recommend to the user.
In order to realize the personalized matching recommendation method based on the knowledge-graph sentry, the invention also discloses a personalized matching recommendation system based on the knowledge-graph sentry, which comprises a user-graph entity interaction module, a recommendation module, a position-graph entity interaction module and a cross compression module;
the cross compression module comprises a cross compression unit I and a cross compression unit II; the cross compression unit is used for automatically learning high-order interaction characteristics between a user and the entity of the job position knowledge graph; the second cross compression unit is used for automatically learning high-order interaction characteristics between positions preferred by the user and position entities of the position knowledge graph;
the user-map entity interaction module is used for learning a triplet relation between a user and the entity of the job position knowledge map;
the position-map entity interaction module is used for learning a triplet relation between positions preferred by the user and position entities of the position knowledge map;
the recommendation module is used for fusing high-order interaction characteristics between the user and the entity of the job position knowledge graph, high-order interaction characteristics between the job position preferred by the user and the job position entity of the job position knowledge graph, and user preference, and predicting interaction scores between the user and all job positions of the job position knowledge graph.
The embodiment of the personalized matching recommendation system based on the knowledge graph sentry can be applied to any device with data processing capability, and the device with the data processing capability can be a device or a device such as a computer. The apparatus embodiments may be implemented by software, or may be implemented by hardware or a combination of hardware and software. Taking software implementation as an example, the device in a logic sense is formed by reading corresponding computer program instructions in a nonvolatile memory into a memory by a processor of any device with data processing capability. In terms of hardware, in addition to the processor, the memory, the network interface, and the nonvolatile memory, any device with data processing capability in the embodiments of the present invention generally may further include other hardware according to the actual function of the any device with data processing capability, which will not be described herein.
The implementation process of the functions and roles of each unit in the above device is specifically shown in the implementation process of the corresponding steps in the above method, and will not be described herein again.
For the device embodiments, reference is made to the description of the method embodiments for the relevant points, since they essentially correspond to the method embodiments. The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purposes of the present invention. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
The embodiment of the invention also provides a computer readable storage medium, and a program is stored on the computer readable storage medium, and when the program is executed by a processor, the personalized matching recommendation method based on the knowledge graph sentry is realized.
The computer readable storage medium may be an internal storage unit, such as a hard disk or a memory, of any of the data processing enabled devices described in any of the previous embodiments. The computer readable storage medium may also be an external storage device, such as a plug-in hard disk, a Smart Media Card (SMC), an SD card, a Flash memory card (Flash card), or the like, provided on the device. Further, the computer readable storage medium may include both internal storage units and external storage devices of any data processing device. The computer readable storage medium is used to store the program of the computing and other programs and data required by any of the data processing devices, and may also be used to temporarily store data that has been or is to be output.
It will be appreciated by persons skilled in the art that the foregoing description is a preferred embodiment of the invention, and is not intended to limit the invention, but rather to limit the invention to the specific embodiments described, and that modifications may be made to the technical solutions described in the foregoing embodiments, or equivalents may be substituted for elements thereof, for the purposes of those skilled in the art. Modifications, equivalents, and alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (10)
1. The personalized matching recommendation method based on the knowledge graph sentry is characterized by comprising the following steps of:
step one: collecting recruitment information issued by a recruitment enterprise, performing data processing, extracting through a relationship to obtain a recruitment triad, and performing knowledge fusion to obtain a job knowledge graph;
step two: collecting historical job hunting behavior data of a user including click browsing, comment and collection, and obtaining a job data set preferred by the user;
step three: taking the position data set preferred by the user as a seed, and applying a RippleNet algorithm to acquire multi-hop position information and the relation thereof from the position knowledge graph to acquire a triplet of entity interaction between the user and the position knowledge graph;
step four: the method comprises the steps of constructing and training a multitask recommendation model based on a knowledge graph and fusing user preferences, wherein the multitask recommendation model comprises a user-graph entity interaction module, a recommendation module and a position-graph entity interaction module;
obtaining a relationship between a user and the knowledge graph through a user-graph entity interaction module;
obtaining a similarity relation between the position and the map entity through a position-map entity interaction module;
obtaining the score of the interaction between the user and the position through a recommendation module;
step five: and obtaining a job ranking list recommended to the user according to the scores of the user and the job interactions.
2. The knowledge-graph-based individual matching recommendation method for sentry, according to claim 1, characterized in thatIn the third step, the triples of the user interaction with the entity of the job position knowledge graph are expressed as<h u ,r u ,t u >Wherein h is u Is a header entity representing a user; r is (r) u Is a relationship or attribute; t is t u Is a tail entity;
obtaining a relationship between a user and the knowledge graph through a user-graph entity interaction module, wherein the relationship comprises the following steps:
in the user-map entity interaction module, t is set u 、r u Respectively extracting information through a deep neural network to respectively obtain t u ′、r u 'A'; will h u Cross-compressing information extraction is carried out on the user set u in the job position knowledge graph, and h is obtained respectively u 'and u'; and r is set to u ' and h u ' input deep neural network, obtain predicted value of tail entity
Finally, the predicted value is evaluated by using a similarity function>
Is a predicted result of (a).
3. The knowledge-based personal post personalized matching recommendation method of claim 2, wherein the recruitment position triad is represented as<h v ,r v ,t v >Wherein h is v A head entity representing a position; r is (r) v Is the relationship from the head entity to the tail entity; t is t v Is a tail entity;
obtaining a similarity relation between the position and the map entity through a position-map entity interaction module, wherein the similarity relation comprises the following specific steps:
in the position-map entity interaction module, r is calculated v And t v Respectively extracting information through a deep neural network to respectively obtain t v ′、r v 'A'; will h v Performing cross compression information extraction with the position set v in the position knowledge graph to obtain respectivelyTo h v 'and v'; and r is set to v ' and h v ' input deep neural network, obtain predicted value of tail entity
Finally, the predicted value is evaluated by using a similarity function>
Is a predicted result of (a).
4. The personalized matching recommendation method based on knowledge-graph sentry according to claim 3, wherein,
in a recommendation module, inputting the job position data set preferred by the user into a bidirectional LSTM network based on an attention mechanism to obtain a user interest preference vector; and fusing the user interest preference vectors, u ', v'; obtaining the interaction probability of the user and all positions
5. The personalized matching recommendation method based on knowledge-graph and post according to claim 4, wherein after the training of the multitask recommendation model based on knowledge-graph and fused with user preference is completed, feature vectors of users, feature vectors of positions and interest preference vectors of users are stored.
6. The personalized matching recommendation method based on knowledge-graph sentry of claim 5, wherein the fifth step is implemented by the following sub-steps:
indexing corresponding user feature vectors and user interest vectors according to the user ID; calculating the similarity between the user and all positions by using a recommendation formula after fusing the user interest preference; and sorting the similarity, and selecting n positions before the maximum similarity to recommend to the user.
7. The knowledge-graph-based person post personalized matching recommendation method according to claim 1, wherein the deep neural network is a multi-layer perceptron.
8. The personalized matching recommendation system based on the knowledge graph person post is characterized by comprising a user-graph entity interaction module, a recommendation module, a position-graph entity interaction module and a cross compression module;
the cross compression module comprises a cross compression unit I and a cross compression unit II; the cross compression unit is used for automatically learning high-order interaction characteristics between a user and the entity of the job position knowledge graph; the second cross compression unit is used for automatically learning high-order interaction characteristics between positions preferred by the user and position entities of the position knowledge graph;
the user-map entity interaction module is used for learning a triplet relation between a user and the entity of the job position knowledge map;
the position-map entity interaction module is used for learning a triplet relation between positions preferred by the user and position entities of the position knowledge map;
the recommendation module is used for fusing high-order interaction characteristics between the user and the entity of the job position knowledge graph, high-order interaction characteristics between the job position preferred by the user and the job position entity of the job position knowledge graph, and user preference, and predicting interaction scores between the user and all job positions of the job position knowledge graph.
9. An electronic device, comprising:
one or more processors;
storage means for storing one or more programs which, when executed by the electronic device, cause the electronic device to implement the knowledge-graph-based person post personalized matching recommendation method of any one of claims 1 to 7.
10. A computer readable medium having stored thereon a computer program, which when executed by a processor implements the knowledge-based person post personalized matching recommendation method according to any of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211583266.4A CN116127186A (en) | 2022-12-09 | 2022-12-09 | Knowledge-graph-based individual matching recommendation method and system for person sentry |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211583266.4A CN116127186A (en) | 2022-12-09 | 2022-12-09 | Knowledge-graph-based individual matching recommendation method and system for person sentry |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116127186A true CN116127186A (en) | 2023-05-16 |
Family
ID=86309021
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211583266.4A Pending CN116127186A (en) | 2022-12-09 | 2022-12-09 | Knowledge-graph-based individual matching recommendation method and system for person sentry |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116127186A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116452169A (en) * | 2023-06-14 | 2023-07-18 | 北京华品博睿网络技术有限公司 | Online recruitment generation type recommendation system and method |
| CN117056458A (en) * | 2023-08-04 | 2023-11-14 | 北京网聘信息技术有限公司 | Method for carrying out front-end retrieval based on vector space algorithm |
| CN117112116A (en) * | 2023-10-16 | 2023-11-24 | 成都市蓉通数智信息技术有限公司 | User management system based on digital government affairs |
| CN117171428A (en) * | 2023-08-04 | 2023-12-05 | 北京网聘信息技术有限公司 | Method for improving accuracy of search and recommendation results |
| CN118132616A (en) * | 2024-05-08 | 2024-06-04 | 江西工业贸易职业技术学院(江西省粮食干部学校、江西省粮食职工中等专业学校) | Auxiliary analysis method and system for professional planning |
-
2022
- 2022-12-09 CN CN202211583266.4A patent/CN116127186A/en active Pending
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116452169A (en) * | 2023-06-14 | 2023-07-18 | 北京华品博睿网络技术有限公司 | Online recruitment generation type recommendation system and method |
| CN116452169B (en) * | 2023-06-14 | 2023-11-24 | 北京华品博睿网络技术有限公司 | Online recruitment generation type recommendation system and method |
| CN117056458A (en) * | 2023-08-04 | 2023-11-14 | 北京网聘信息技术有限公司 | Method for carrying out front-end retrieval based on vector space algorithm |
| CN117171428A (en) * | 2023-08-04 | 2023-12-05 | 北京网聘信息技术有限公司 | Method for improving accuracy of search and recommendation results |
| CN117171428B (en) * | 2023-08-04 | 2024-04-05 | 北京网聘信息技术有限公司 | Method for improving accuracy of search and recommendation results |
| CN117056458B (en) * | 2023-08-04 | 2024-04-19 | 北京网聘信息技术有限公司 | Method for carrying out front-end retrieval based on vector space algorithm |
| CN117112116A (en) * | 2023-10-16 | 2023-11-24 | 成都市蓉通数智信息技术有限公司 | User management system based on digital government affairs |
| CN117112116B (en) * | 2023-10-16 | 2024-02-02 | 成都市蓉通数智信息技术有限公司 | User management system based on digital government affairs |
| CN118132616A (en) * | 2024-05-08 | 2024-06-04 | 江西工业贸易职业技术学院(江西省粮食干部学校、江西省粮食职工中等专业学校) | Auxiliary analysis method and system for professional planning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Exploration of social media for sentiment analysis using deep learning | |
| CN111931062B (en) | Training method and related device of information recommendation model | |
| CN111241311B (en) | Media information recommendation method and device, electronic equipment and storage medium | |
| CN116127186A (en) | Knowledge-graph-based individual matching recommendation method and system for person sentry | |
| Guo et al. | Efficient hyperparameter optimization for convolution neural networks in deep learning: A distributed particle swarm optimization approach | |
| Kavasidis et al. | An innovative web-based collaborative platform for video annotation | |
| CN110442790A (en) | Recommend method, apparatus, server and the storage medium of multi-medium data | |
| US20220171760A1 (en) | Data processing method and apparatus, computer-readable storage medium, and electronic device | |
| CN111400603A (en) | Information pushing method, device and equipment and computer readable storage medium | |
| Wang et al. | Deep user modeling for content-based event recommendation in event-based social networks | |
| Wang et al. | MFGNet: Dynamic modality-aware filter generation for RGB-T tracking | |
| CN114201516B (en) | User portrait construction method, information recommendation method and related devices | |
| CN112699310A (en) | Cold start cross-domain hybrid recommendation method and system based on deep neural network | |
| Ni et al. | Comparative convolutional dynamic multi-attention recommendation model | |
| Zhu et al. | Multimodal emotion classification with multi-level semantic reasoning network | |
| Sun et al. | Multi-interest sequence modeling for recommendation with causal embedding | |
| Liang et al. | Graph path fusion and reinforcement reasoning for recommendation in MOOCs | |
| Long et al. | Cross-domain personalized image captioning | |
| Zhang et al. | Hybrid structural graph attention network for POI recommendation | |
| Zhang et al. | Probabilistic matrix factorization recommendation of self-attention mechanism convolutional neural networks with item auxiliary information | |
| CN113407704A (en) | Text matching method, device and equipment and computer readable storage medium | |
| Gao et al. | Sequential label enhancement | |
| Li et al. | Adversarial learning of transitive semantic features for cross-domain recommendation | |
| Wan et al. | A learner’s role-based multi dimensional collaborative recommendation (LRMDCR) for group learning support | |
| Ramos et al. | Natural Language User Profiles for Transparent and Scrutable Recommendations |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |