Recruitment trend prediction and question answering method based on knowledge graph and time sequence

Abstract

The invention discloses a recruitment trend prediction and question answering method based on a knowledge graph and a time sequence. The method comprises the following steps: S1, constructing a position information data set; s2, representing the position information data set as a knowledge graph through semantic analysis; s3, continuously updating and managing nodes and edges in the knowledge graph based on a new recruitment data source or market change; s4, identifying a periodic mode, a seasonal change and other trends; s5, predicting future recruitment demand changes; s6, receiving a natural language query of a user, performing semantic analysis and intention recognition on the query, retrieving related information from a knowledge graph and a time sequence analysis result based on an analysis result, and generating and outputting a corresponding natural language answer; and S7, performing visualization processing on the prediction result and the question and answer result to generate a visualization report comprising a chart and a trend line. According to the invention, the accuracy and efficiency of recruitment data analysis and prediction are improved.

Description

Technical Field

[0001] The present invention relates to the field of recruitment technology, and in particular to a recruitment trend prediction and question-answering method based on knowledge graph and time series. Background Art

[0002] In the prior art, the analysis and prediction of the recruitment market mainly rely on traditional data processing and analysis methods. Usually, the market demand trend is identified by statistical analysis of structured recruitment data. Simple regression analysis of historical data, linear extrapolation or basic statistical methods are generally used to predict future recruitment needs. Although the above methods can reflect the overall changes in the recruitment market to a certain extent, with the diversification and complexity of data sources, traditional methods are unable to cope with multi-dimensional and unstructured data.

[0003] The information sources of the recruitment market are complex and diverse, involving job descriptions, company information, job seekers' backgrounds, industry trends and other aspects. Existing technologies mainly rely on manual or semi-automatic methods to integrate and analyze information, lacking systematic semantic understanding and data integration capabilities, resulting in low efficiency in information extraction and utilization. In addition, existing recruitment data analysis tools are usually limited to processing structured data and cannot effectively utilize the potential information in natural language texts, which makes the analysis of the recruitment market lack depth and breadth.

[0004] Changes in recruitment demand are often affected by time factors, but existing analysis tools mostly use static data analysis methods, which makes it difficult to capture the dynamic trend of recruitment demand over time. The limitations of static analysis make it difficult for companies to make timely and effective recruitment decisions when facing a rapidly changing market environment, which can easily lead to problems such as insufficient staff or over-recruitment. In recruitment-related question-and-answer systems, existing technologies usually process user queries based on keyword matching or simple rules, lacking in-depth understanding of natural language and semantic analysis capabilities of context. Not only is it difficult to process complex queries and unable to meet users' needs for accurate and timely recruitment information, but it is also easy to make misjudgments or fail to give effective answers when faced with semantically ambiguous or polysemous queries, resulting in a poor user experience. Summary of the invention

[0005] One object of the present invention is to propose a recruitment trend prediction and question-answering method based on knowledge graph and time series, which improves the accuracy and efficiency of recruitment data analysis and prediction.

[0006] A recruitment trend prediction and question-answering method based on knowledge graph and time series according to an embodiment of the present invention includes the following steps:

[0007] S1. Extract job description data, company information data, job seeker background data and industry trend data from various recruitment-related data sources to construct a job information dataset;

[0008] S2, through semantic analysis, the job information dataset is represented as a knowledge graph, and the knowledge graph includes nodes and edges;

[0009] S3. Based on new recruitment data sources or market changes, the nodes and edges in the knowledge graph are continuously updated and managed so that the information in the knowledge graph can reflect the current recruitment market situation in real time;

[0010] S4. Conduct time series analysis on historical job information datasets in the recruitment field, including the time of job postings, job seekers’ application time, and seasonal changes in market demand, to identify cyclical patterns, seasonal changes, and other trends;

[0011] S5. Combine the results of knowledge graph and time series analysis to predict future changes in recruitment demand;

[0012] S6. Receive a natural language query from a user, perform semantic analysis and intent recognition on the query, retrieve relevant information from the knowledge graph and the results of time series analysis based on the analysis results, and generate and output a corresponding natural language answer;

[0013] S7. Visualize the prediction results and question-and-answer results to generate a visualization report including charts and trend lines.

[0014] Optionally, the S1 includes the following steps:

[0015] S11. Extracting job description data from various recruitment-related data sources p 、Company information data D c , Job seeker background data D a and industry trend data i ;

[0016] S12. Extract the job description data D p 、Company information data D c , Job seeker background data D a and industry trend data i Perform preprocessing to remove noisy data and standardize the formats of various types of data;

[0017] S13. Constructing a job information dataset D based on the preprocessed data t :

[0018] D t ={(D′ p ,D′ c ,D′ a,D′ i )};

[0019] Among them, D′ p represents the preprocessed job description dataset, including job titles, job requirements, and required skills; D′ c represents the preprocessed company information dataset, including company name, industry type and company size information; D′ a represents the preprocessed job applicant background dataset, including job applicant name, experience and education background attributes; D′ i Represents the preprocessed industry trend dataset, including industry categories, market trends, and demand fluctuation data.

[0020] Optionally, S2 includes the following steps:

[0021] S21. Position information dataset D t Perform semantic analysis and map the job description data, company information data, job seeker background data, and industry trend data into job node N in the knowledge graph. p 、Company Node N c 、Job seeker node N a and industry node N i , position node N p Including semantic information of job title, job requirements and required skills, company node N c Including company name, industry type and company size information, job seeker node N a The semantic information including the job applicant’s name, experience and education background, industry node N i Including semantic information of industry categories, market trends and demand fluctuations, the final mapped node N x The expression is:

[0022]

[0023] Among them, x∈{p,c,a,i} represents the node type, D′ x is the corresponding job information dataset, is the i-th element in the job information dataset, n x is the number of elements in the job information dataset, φ x N is a function that maps raw data into semantic features. x is the final mapped node;

[0024] S22. Extract the association information between entities from the job information dataset based on the semantic relationship between nodes, and map the association information to the edge E in the knowledge graph. xy :

[0025]

[0026] Where x,y∈{p,c,a,i},E xy Represents node N x With node N y The edge between represents the i-th relation instance extracted from the job information dataset, m xy is the number of relation instances, ψ xy A function that maps a relation instance to a semantic association;

[0027] Edge E pc Represents position node N p With company node N c The relationship between the position and the company, ps Represents position node N p The relationship between the skills required in the job requirements and the ap Represents job seeker node N a With position node N p The matching relationship between the edges E ic Represents industry node N i The impact relationship with the company's recruitment needs;

[0028] S23, the generated knowledge graph G is expressed as:

[0029] G=(N,E)=(U x∈{p,c,a,i} N x ,∪ (x,y)∈{(p,c),(p,s),(a,p),(i,c)} E xy );

[0030] Where N is the node set.

[0031] Optionally, S4 includes the following steps:

[0032] S41. Historical job information dataset D in the recruitment field t Preprocess the time-related data in the job postings, including the time data T p , Job seeker application time data T a and seasonal variation data of market demand T s Get the preprocessed time data;

[0033] S42, perform time series modeling on the preprocessed time data to generate a time series TS x :

[0034]

[0035] Among them, α x , β x , γx ,ω x and φ x are the model parameters for the time data type x, representing the growth rate, amplitude, frequency, and phase of the time series. is the data at the i-th time point, TS x is the generated time series;

[0036] S43. Time series TS x Analyze to identify cyclical patterns, seasonal changes, and other trends:

[0037]

[0038] Among them, λ x (k) and κ x (j) are the weight functions of Fourier transform and ARIMA model, i is the imaginary unit, p, d, q are the parameters of ARIMA model, n x is the length of the time series, m x is the order of the ARIMA model, A x It is the result of time series analysis;

[0039] S44, the result of time series analysis A x Mapping back to the job information dataset D t , identifying cyclical patterns and seasonal changes:

[0040]

[0041] Among them, g x (t) is the mapping function associated with the time data type x, A x is the result of time series analysis, D t is the original job information dataset, D″ t is the job information dataset after mapping, t0 and t n are the starting point and end point of the time interval respectively.

[0042] Optionally, S5 includes the following steps:

[0043] S51, the knowledge graph G and the time series analysis result A x Combined with the input dataset I to generate the prediction model t :

[0044] I t =(G,∑ x∈{p,a,s} A x );

[0045] Among them, I tAn input dataset representing a specific position in a certain period of time in the future. The input dataset includes the semantic relationship between the position and the company, the historical trend of position demand, and other time series features related to prediction.

[0046] S52, according to the input data set I t Constructing demand forecasting model M d The output of the demand forecasting model is the demand for a specific position in a specific time period in the future D f ;

[0047] S53. Verify and adjust the demand for specific positions in the future, and iteratively train and optimize the demand forecasting model based on historical data and market conditions;

[0048] S54, the final generated prediction result D′ f Provided to users as a basis for recruiting decision support.

[0049] Optionally, the construction of the demand forecasting model includes:

[0050] The input data set I t Through the embedding layer E, multi-dimensional feature extraction and mapping are performed to obtain the embedded feature vector v t :

[0051]

[0052] Among them, α i and Respectively represent the weight parameter of the i-th node feature in the knowledge graph G and the time series A x The weight parameter at the jth time point in G i ) represents the i-th node feature extracted from the knowledge graph, Represents the features of the jth time point in the time series, embedded in the feature vector v t It is multi-dimensional, including the fusion representation of position information and time series features;

[0053] Embed the feature vector v t Input to the recurrent neural network layer RNN to capture the time series pattern and model the dynamic changes to generate a time-sensitive feature vector h t :

[0054] h t =RNN(v t )=σ(W h ·v t +U h ·h t-1 +b h );

[0055] Among them, Wh and U h is the weight matrix of the recurrent neural network, b h is the bias term, σ is the activation function, h t-1 is the hidden state of the previous time step, the time-sensitive feature vector h t Including the dynamic characteristics of job requirements changing over time;

[0056] The time-sensitive feature vector h t Input to the fully connected layer FC for nonlinear combination and prediction of job demand, and obtain the future demand D for a specific job f :

[0057] D f =FC(h t )=W f ·tanh(h t )+b f ;

[0058] Among them, W f is the weight matrix of the fully connected layer, b f is the bias term, and tanh is the hyperbolic tangent activation function.

[0059] Optionally, the S6 comprises the following steps:

[0060] S61. Receive user's natural language query Q u , the query is segmented, part-of-speech tagged, and dependency syntactic analysis is performed through the natural language processing module to generate an intermediate representation S of semantic parsing q ;

[0061] S62. Intermediate representation based on semantic analysis q Perform intent recognition through the pre-trained classification model C i Classify the user's query intent and generate the intent label L q ;

[0062] S63, according to the intention label L q From the knowledge graph G and time series analysis results A x Retrieve relevant information from the r ;

[0063] S64. Set the preliminary answer set A r Input to the answer generation module G a Generate structured natural language answers A n .

[0064] The beneficial effects of the present invention are:

[0065] (1) The present invention realizes the semantic integration and management of multi-source recruitment data by constructing a knowledge graph based on the recruitment field. The dynamic update mechanism of nodes and edges in the knowledge graph enables the system to reflect changes in the recruitment market in real time, which not only improves the efficiency of data organization and retrieval, but also converts complex recruitment-related information into practical knowledge through semantic level analysis, thereby supporting more accurate trend prediction and complex query.

[0066] (2) The present invention innovatively combines time series analysis with knowledge graphs, successfully solving the problem of dynamic prediction of recruitment demand over time. The time series analysis model system can identify the cyclical patterns, seasonal changes and other trends of recruitment demand, and combine the semantic information in the knowledge graph to make more accurate demand forecasts. It is not only based on the dynamic changes of historical data, but also comprehensively considers the multi-dimensional information of positions, companies, and industries, making the prediction results more accurate and timely.

[0067] (3) The present invention realizes intelligent analysis and response to users' natural language queries by introducing natural language processing technology. The system can perform semantic understanding and intent recognition on complex recruitment-related questions raised by users, and generate accurate natural language answers based on knowledge graph and time series analysis results. It can not only handle semantically ambiguous or polysemous queries, but also provide more accurate and context-related answers, thereby greatly improving the user experience and meeting users' needs for complex information queries. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0069] Figure 1 A flowchart of a recruitment trend prediction and question-answering method based on knowledge graph and time series proposed by the present invention;

[0070] Figure 2 This is a schematic diagram of constructing a knowledge graph in the recruitment field in a recruitment trend prediction and question-answering method based on a knowledge graph and time series proposed in the present invention. DETAILED DESCRIPTION

[0071] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0072] refer to Figure 1-2 , a recruitment trend prediction and question-answering method based on knowledge graph and time series, including the following steps:

[0073] S1. Extract job description data, company information data, job seeker background data and industry trend data from various recruitment-related data sources to construct a job information dataset;

[0074] S2, through semantic analysis, the job information dataset is represented as a knowledge graph, and the knowledge graph includes nodes and edges;

[0075] S3. Based on new recruitment data sources or market changes, the nodes and edges in the knowledge graph are continuously updated and managed so that the information in the knowledge graph can reflect the current recruitment market situation in real time;

[0076] S4. Conduct time series analysis on historical job information datasets in the recruitment field, including the time of job postings, job seekers’ application time, and seasonal changes in market demand, to identify cyclical patterns, seasonal changes, and other trends;

[0077] S5. Combine the results of knowledge graph and time series analysis to predict future changes in recruitment demand;

[0078] S6. Receive the user's natural language query, perform semantic analysis and intent recognition on the query, retrieve relevant information from the knowledge graph and time series analysis results based on the analysis results, and generate and output the corresponding natural language answer;

[0079] S7. Visualize the prediction results and question-and-answer results to generate a visualization report including charts and trend lines.

[0080] In this implementation, S1 includes the following steps:

[0081] S11. Extracting job description data from various recruitment-related data sources p 、Company information data D c , Job seeker background data D a and industry trend data i ;

[0082] S12. Extract the job description data D p 、Company information data D c , Job seeker background data D a and industry trend data i Perform preprocessing to remove noisy data and standardize the formats of various types of data;

[0083] S13. Constructing a job information dataset D based on the preprocessed data t :

[0084] D t ={(D′ p ,D′ c ,D′ a,D′ i )};

[0085] Among them, D′ p represents the preprocessed job description dataset, including job titles, job requirements, and required skills; D′ c represents the preprocessed company information dataset, including company name, industry type and company size information; D′ a represents the preprocessed job applicant background dataset, including job applicant name, experience and education background attributes; D′ i Represents the preprocessed industry trend dataset, including industry categories, market trends, and demand fluctuation data.

[0086] In this implementation, S2 includes the following steps:

[0087] S21. Position information dataset D t Perform semantic analysis and map the job description data, company information data, job seeker background data, and industry trend data into job node N in the knowledge graph. p 、Company Node N c 、Job seeker node N a and industry node N i , position node N p Including semantic information of job title, job requirements and required skills, company node N c Including company name, industry type and company size information, job seeker node N a The semantic information including the job applicant’s name, experience and education background, industry node N i Including semantic information of industry categories, market trends and demand fluctuations, the final mapped node N x The expression is:

[0088]

[0089] Among them, x∈{p,c,a,i} represents the node type, D′ x is the corresponding job information dataset, is the i-th element in the job information dataset, n x is the number of elements in the job information dataset, φ x N is a function that maps raw data into semantic features. x is the final mapped node;

[0090] S22. Extract the association information between entities from the job information dataset based on the semantic relationship between nodes, and map the association information to the edge E in the knowledge graph. xy :

[0091]

[0092] Where x,y∈{p,c,a,i},E xy Represents node N x With node N y The edge between represents the i-th relation instance extracted from the job information dataset, m xy is the number of relation instances, ψ xy A function that maps a relation instance to a semantic association;

[0093] Edge E pc Represents position node N p With company node N c The relationship between the position and the company, ps Represents position node N p The relationship between the skills required in the job requirements and the ap Represents job seeker node N a With position node N p The matching relationship between the edges E ic Represents industry node N i The impact relationship with the company's recruitment needs;

[0094] S23, the generated knowledge graph G is expressed as:

[0095] G=(N,E)=(∪ x∈{p,c,a,i} N x ,∪ (x,y)∈{(p,c),(p,s),(a,p),(i,c)} E xy );

[0096] Where N is the node set.

[0097] In this implementation, S4 includes the following steps:

[0098] S41. Historical job information dataset D in the recruitment field t Preprocess the time-related data in the job postings, including the time data T p , Job seeker application time data T a and seasonal variation data of market demand T s Get the preprocessed time data;

[0099] S42, perform time series modeling on the preprocessed time data to generate a time series TS x :

[0100]

[0101] Among them, α x , β x, γ x ,ω x and φ x are the model parameters for the time data type x, representing the growth rate, amplitude, frequency, and phase of the time series. is the data at the i-th time point, TS x is the generated time series;

[0102] S43. Time series TS x Analyze to identify cyclical patterns, seasonal changes, and other trends:

[0103]

[0104] Among them, λ x (k) and κ x (j) are the weight functions of Fourier transform and ARIMA model, i is the imaginary unit, p, d, q are the parameters of ARIMA model, n x is the length of the time series, m x is the order of the ARIMA model, A x It is the result of time series analysis;

[0105] S44, the result of time series analysis A x Mapping back to the job information dataset D t , identifying cyclical patterns and seasonal changes:

[0106]

[0107] Among them, g x (t) is the mapping function associated with the time data type x, A x is the result of time series analysis, D t is the original job information dataset, D″ t is the job information dataset after mapping, t0 and t n are the starting point and end point of the time interval respectively.

[0108] In this implementation, S5 includes the following steps:

[0109] S51, the knowledge graph G and the time series analysis result A x Combined with the input dataset I to generate the prediction model t :

[0110] I t =(G,Σ x∈{p,a,s} A x );

[0111] Among them, I tAn input dataset representing a specific position in a certain period of time in the future. The input dataset includes the semantic relationship between the position and the company, the historical trend of position demand, and other time series features related to prediction.

[0112] S52, according to the input data set I t Constructing demand forecasting model M d The output of the demand forecasting model is the demand for a specific position in a specific time period in the future D f ;

[0113] S53. Verify and adjust the demand for specific positions in the future, and iteratively train and optimize the demand forecasting model based on historical data and market conditions;

[0114] S54, the final generated prediction result D′ f Provided to users as a basis for recruiting decision support.

[0115] In this implementation, the construction of the demand forecasting model includes:

[0116] The input data set I t Through the embedding layer E, multi-dimensional feature extraction and mapping are performed to obtain the embedded feature vector v t :

[0117]

[0118] Among them, α i and Respectively represent the weight parameter of the i-th node feature in the knowledge graph G and the time series A x The weight parameter at the jth time point in G i ) represents the i-th node feature extracted from the knowledge graph, Represents the features of the jth time point in the time series, embedded in the feature vector v t It is multi-dimensional, including the fusion representation of position information and time series features;

[0119] Embed the feature vector v t Input to the recurrent neural network layer RNN to capture the time series pattern and model the dynamic changes to generate a time-sensitive feature vector h t :

[0120] h t =RNN(v t )=σ(W h ·v t +U h ·h t-1 +b h );

[0121] Among them, Wh and U h is the weight matrix of the recurrent neural network, b h is the bias term, σ is the activation function, h t-1 is the hidden state of the previous time step, the time-sensitive feature vector h t Including the dynamic characteristics of job requirements changing over time;

[0122] The time-sensitive feature vector h t Input to the fully connected layer FC for nonlinear combination and prediction of job demand, and obtain the future demand D for a specific job f :

[0123] D f =FC(h t )=W f ·tanh(h t )+b f ;

[0124] Among them, W f is the weight matrix of the fully connected layer, b f is the bias term, and tanh is the hyperbolic tangent activation function.

[0125] In this implementation, S6 includes the following steps:

[0126] S61. Receive user's natural language query Q u , the query is segmented, part-of-speech tagged, and dependency syntactic analysis is performed through the natural language processing module to generate an intermediate representation S of semantic parsing q ;

[0127] S62. Intermediate representation based on semantic analysis q Perform intent recognition through the pre-trained classification model C i Classify the user's query intent and generate the intent label L q ;

[0128] S63, according to the intention label L q From the knowledge graph G and time series analysis results A x Retrieve relevant information from the r ;

[0129] S64. Set the preliminary answer set A r Input to the answer generation module G a Generate structured natural language answers A n .

[0130] Embodiment 1:

[0131] A global leading technology company is a high-tech enterprise focusing on 5G communication technology, artificial intelligence and big data analysis. In January 2024, the company plans to launch several important projects in the next 6 months and urgently needs to recruit a group of high-quality technical talents, especially artificial intelligence engineers and big data analysts. However, faced with the rapidly changing market environment and talent competition, the company encountered difficulties in formulating recruitment plans, especially in predicting future recruitment needs and responding quickly to market changes.

[0132] The company used its internal recruitment system to collect recruitment data from the past five years, including job descriptions, internal job requirements, job seekers' resumes, and market recruitment trend reports. After data cleaning and standardization, it constructed a knowledge graph that includes job nodes, company nodes, job seeker nodes, and industry nodes.

[0133] In the process of recruiting demand forecasting, the company conducted a time series analysis of the recruitment data of various technical positions in the past five years. The system identified that in the past three years, the recruitment demand for artificial intelligence engineers reached a peak in February and July each year, and showed an increasing trend year by year. The system also analyzed the company's new project launch plan and predicted that from March to June 2024, the company's demand for senior artificial intelligence engineers will increase significantly. The specific time series analysis results are shown in Table 1 below:

[0134] Table 1 Time series analysis results

[0135]

[0136] By combining the results of time series analysis with knowledge graphs, the company built a forecasting model, which showed that the demand for AI engineers and big data analysts will peak in April and May 2024. The company adjusted its recruitment plan accordingly and started talent reserve in advance.

[0137] In the actual recruitment process, the company's human resources department used the natural language question-answering system to conduct multiple inquiries. In Example 1, on March 10, 2024, Ms. Li, the HR manager of the company, entered the question "How will the demand for big data analysts change in the next two months?" The system performed word segmentation and semantic analysis on the query through the natural language processing module, and generated the following reply based on the knowledge graph and time series analysis results:

[0138] "It is estimated that the demand for big data analysts will be 40 and 42 respectively during April and May 2024, mainly concentrated in the start-up phase of big data projects. It is recommended to start recruitment in mid-March in advance to cope with the peak demand."

[0139] Based on the system's recommendations, the HR team started the recruitment advertising and talent screening work in advance, and finally successfully completed the recruitment goal of big data analysts at the end of May.

[0140] In order to verify the effectiveness of the method of the present invention, the company conducted a detailed comparison of the prediction results and actual recruitment data of the traditional method and the method of the present invention, as shown in Table 2 below:

[0141] Table 2 Implementation effect and data comparison

[0142]

[0143] From the data comparison, it can be seen that the method of the present invention is significantly better than the traditional method in terms of the accuracy of recruitment demand forecasting, especially during peak demand periods. The company can adjust its recruitment strategy in a timely manner, significantly improving the recruitment success rate. At the same time, since the demand forecast is made in advance, the company is more efficient in recruitment advertising and interview arrangements, reducing ineffective expenses.

[0144] Through the application of the present invention, the company not only successfully predicted and responded to the peak of recruitment for key technical positions in the first half of 2024, but also optimized the recruitment process and reduced human resource management costs. This embodiment fully demonstrated the advantages of the present invention in solving problems such as the difficulty of multi-source data integration, inaccurate dynamic trend prediction, and untimely natural language question and answer response, and proved the feasibility and effectiveness of the present invention in an actual recruitment environment.

[0145] The present invention realizes the semantic integration and management of multi-source recruitment data by constructing a knowledge graph based on the recruitment field. The dynamic update mechanism of nodes and edges in the knowledge graph enables the system to reflect changes in the recruitment market in real time, which not only improves the efficiency of data organization and retrieval, but also converts complex recruitment-related information into practical knowledge through semantic level analysis, thereby supporting more accurate trend prediction and complex queries.

[0146] This invention innovatively combines time series analysis with knowledge graphs, successfully solving the problem of dynamic prediction of recruitment demand over time. The time series analysis model system can identify cyclical patterns, seasonal changes and other trends in recruitment demand, and combine the semantic information in the knowledge graph to make more accurate demand forecasts. It is not only based on the dynamic changes of historical data, but also comprehensively considers multi-dimensional information of positions, companies, and industries, making the prediction results more accurate and timely.

[0147] The present invention realizes intelligent parsing and response to users' natural language queries by introducing natural language processing technology. The system can perform semantic understanding and intent recognition on complex recruitment-related questions raised by users, and generate accurate natural language answers based on knowledge graph and time series analysis results. It can not only handle semantically ambiguous or polysemous queries, but also provide more accurate and context-related answers, thereby greatly improving user experience and meeting users' needs for complex information queries.

[0148] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A recruitment trend prediction and question-answering method based on knowledge graph and time series, characterized in that: The steps include: S1. Extract job description data, company information data, job seeker background data and industry trend data from various recruitment-related data sources to construct a job information dataset; S2, through semantic analysis, the job information dataset is represented as a knowledge graph, and the knowledge graph includes nodes and edges; S3. Based on new recruitment data sources or market changes, the nodes and edges in the knowledge graph are continuously updated and managed so that the information in the knowledge graph can reflect the current recruitment market situation in real time; S4. Conduct time series analysis on historical job information datasets in the recruitment field, including the time of job postings, job seekers’ application time, and seasonal changes in market demand, to identify cyclical patterns, seasonal changes, and other trends; S5. Combine the results of knowledge graph and time series analysis to predict future changes in recruitment demand; S6. Receive a natural language query from a user, perform semantic analysis and intent recognition on the query, retrieve relevant information from the knowledge graph and the results of time series analysis based on the analysis results, and generate and output a corresponding natural language answer; S7. Visualize the prediction results and question-and-answer results to generate a visualization report including charts and trend lines.

2. According to claim 1, a recruitment trend prediction and question-answering method based on knowledge graph and time series, characterized in that: The S1 comprises the following steps: S11. Extracting job description data from various recruitment-related data sources p 、Company information data D c , Job seeker background data D a and industry trend data i ; S12. Extract the job description data D p 、Company information data D c , Job seeker background data D a and industry trend data i Perform preprocessing to remove noisy data and standardize the formats of various types of data; S13. Constructing a job information dataset D based on the preprocessed data t : D t ={(D′ p ,D′ c ,D′ a ,D′ i )}; Among them, D′ p represents the preprocessed job description dataset, including job titles, job requirements, and required skills; D′ c represents the preprocessed company information dataset, including company name, industry type and company size information; D′ a represents the preprocessed job applicant background dataset, including job applicant name, experience and education background attributes; D′ i Represents the preprocessed industry trend dataset, including industry categories, market trends, and demand fluctuation data.

3. According to claim 1, a recruitment trend prediction and question-answering method based on knowledge graph and time series, characterized in that: The S2 comprises the following steps: S21. Position information dataset D t Perform semantic analysis and map the job description data, company information data, job seeker background data, and industry trend data into job node N in the knowledge graph. p 、Company Node N c 、Job seeker node N a and industry node N i , position node N p Including semantic information of job title, job requirements and required skills, company node N c Including company name, industry type and company size information, job seeker node N a The semantic information including the job applicant’s name, experience and education background, industry node N i Including semantic information of industry categories, market trends and demand fluctuations, the final mapped node N x The expression is: Among them, x∈{p,c,a,i} represents the node type, D′ x is the corresponding job information dataset, is the i-th element in the job information dataset, n x is the number of elements in the job information dataset, φ x N is a function that maps raw data into semantic features. x is the final mapped node; S22. Extract the association information between entities from the job information dataset based on the semantic relationship between nodes, and map the association information to the edge E in the knowledge graph. xy : Where x,y∈{p,c,a,i},E xy Represents node N x With node N y The edge between xyi represents the i-th relation instance extracted from the job information dataset, m xy is the number of relation instances, ψ xy A function that maps a relation instance to a semantic association; Edge E pc Represents position node N p With company node N c The relationship between the position and the company, ps Represents position node N p The relationship between the skills required in the job requirements and the side ap Represents job seeker node N a With position node N p The matching relationship between the edges E ic Represents industry node N i The impact relationship with the company's recruitment needs; S23, the generated knowledge graph G is expressed as: G=(N,E)=(∪ x∈{p,c,a,i} N x ,∪ (x,y)∈{(p,c),(p,s),(a,p),(i,c)} E xy ); Where N is the node set.

4. According to claim 1, a recruitment trend prediction and question-answering method based on knowledge graph and time series, characterized in that: The S4 comprises the following steps: S41. Historical job information dataset D in the recruitment field t Preprocess the time-related data in the job postings, including the time data T p , Job seeker application time data T a and seasonal variation data of market demand T s Get the preprocessed time data; S42, perform time series modeling on the preprocessed time data to generate a time series TS x : Among them, α x , β x , γ x ,ω x and φ x are the model parameters for the time data type x, representing the growth rate, amplitude, frequency, and phase of the time series. is the data at the i-th time point, TS x is the generated time series; S43. Time series TS x Analyze to identify cyclical patterns, seasonal changes, and other trends: Among them, λ x (k) and κ x (j) are the weight functions of Fourier transform and ARIMA model, i is the imaginary unit, p, d, q are the parameters of ARIMA model, n x is the length of the time series, m x is the order of the ARIMA model, A x It is the result of time series analysis; S44, the result of time series analysis A x Mapping back to the job information dataset D t , identifying cyclical patterns and seasonal changes: Among them, g x (t) is the mapping function associated with the time data type x, A x is the result of time series analysis, D t is the original job information dataset, D″ t is the job information dataset after mapping, t0 and t n are the starting point and end point of the time interval respectively.

5. According to claim 1, a recruitment trend prediction and question-answering method based on knowledge graph and time series, characterized in that: The S5 comprises the following steps: S51, the knowledge graph G and the time series analysis result A x Combined with the input dataset I to generate the prediction model t : I t =(G,∑ x∈{p,a,s} A x ); Among them, I t An input dataset representing a specific position in a certain period of time in the future. The input dataset includes the semantic relationship between the position and the company, the historical trend of position demand, and other time series features related to prediction. S52, according to the input data set I t Constructing demand forecasting model M d The output of the demand forecasting model is the demand for a specific position in a specific time period in the future D f ; S53. Verify and adjust the demand for specific positions in the future, and iteratively train and optimize the demand forecasting model based on historical data and market conditions; S54, the final generated prediction result D′ f Provided to users as a basis for recruiting decision support.

6. The recruitment trend prediction and question-answering method based on knowledge graph and time series according to claim 1 is characterized in that: The construction of the demand forecasting model includes: Input data set I t Through the embedding layer E, multi-dimensional feature extraction and mapping are performed to obtain the embedded feature vector v t : Among them, α i and β xj Respectively represent the weight parameter of the i-th node feature in the knowledge graph G and the time series A x The weight parameter at the jth time point in G i ) represents the i-th node feature extracted from the knowledge graph, ψ(A xj ) represents the feature of the jth time point in the time series, embedded in the feature vector v t It is multi-dimensional, including the fusion representation of position information and time series features; Embed the feature vector v t Input to the recurrent neural network layer RNN to capture the time series pattern and model the dynamic changes to generate a time-sensitive feature vector h t : h t =RNN(v t )=σ(W h ·v t +U h ·h t-1 +b h ); Among them, W h and U h is the weight matrix of the recurrent neural network, b h is the bias term, σ is the activation function, h t-1 is the hidden state of the previous time step, the time-sensitive feature vector h t Including the dynamic characteristics of job requirements changing over time; The time-sensitive feature vector h t Input to the fully connected layer FC for nonlinear combination and prediction of job demand, and obtain the future demand D for a specific job f : D f =FC(h t )=W f ·tanh(h t )+b f ; Among them, W f is the weight matrix of the fully connected layer, b f is the bias term, and tanh is the hyperbolic tangent activation function.

7. The recruitment trend prediction and question-answering method based on knowledge graph and time series according to claim 1, characterized in that: The S6 comprises the following steps: S61. Receive user's natural language query Q u , the query is segmented, part-of-speech tagged, and dependency syntactic analysis is performed through the natural language processing module to generate an intermediate representation S of semantic parsing q ; S62. Intermediate representation based on semantic analysis q Perform intent recognition through the pre-trained classification model C i Classify the user's query intent and generate the intent label L q ; S63, according to the intention label L q From the knowledge graph G and time series analysis results A x Retrieve relevant information from the r ; S64. Set the preliminary answer set A r Input to the answer generation module G a Generate structured natural language answers A n .

Images