CN116362699A

CN116362699A - Post matching report generation method

Info

Publication number: CN116362699A
Application number: CN202310250101.3A
Authority: CN
Inventors: 郑未; 张航
Original assignee: Guoxin Blue Bridge Education Technology Co ltd
Current assignee: Guoxin Blue Bridge Education Technology Co ltd
Priority date: 2023-03-15
Filing date: 2023-03-15
Publication date: 2023-06-30

Abstract

The application provides a post matching report generation method, which comprises the following steps: acquiring talent data and post data; screening talents meeting the post requirements based on the knowledge graph; acquiring a work experience text and a self-evaluation text in talent resume, and calculating post matching degree; acquiring talent interview conditions with post matching degree meeting recruitment requirements, and automatically generating a post matching report; analyzing the variable area in the post matching report, and detecting the total number of the change of the recruiters of each post; calculating post matching degree of the changed recruiters, sequencing talents, and determining the prepared recruiter talents; adjusting the content of the post matching report according to the post personnel change condition and the list of the prepared recruiters, and outputting a new post report; and (5) automatically re-layout design and adjustment are carried out on the post matching report.

Description

Post matching report generation method

Technical Field

The invention relates to the technical field of information, in particular to a post matching report generation method.

Background

After talent application, the recruiter needs to evaluate talents according to talent data and post data, and selects talents suitable for posts; the post report is a text report which is convenient for recruiters to quickly know information of the recruiters and know the conditions of the recruitment of each post; after the post report is generated, the post report needs to be seen by the interview related personnel; however, the interviewee has a long time difference from interview to job entry, and the situation of personnel may change; the change of talent information can be caused by the change of the recruiter, and the information is inaccurate if the report is displayed in the original mode, so that the judgment of the interview related personnel is affected, and the generated post report needs to be adjusted so as to more clearly define the recruiter list according to the delivery change condition. But post reports are generally not easily modified after generation; therefore, there is a need for a way to ensure that the post report is beautiful enough and reasonable enough when the post report is generated, so that the post report is reserved with areas that change due to talent variation and delivery resume conditions or need to modify the report, and the report content can be better laid out and cut.

Disclosure of Invention

The invention provides a post matching report generation method, which mainly comprises the following steps:

the method for acquiring talent data and post data specifically comprises the following steps: extracting entities in post job-free requirement texts based on a named entity recognition model of a rule base; screening talents meeting the post requirements based on the knowledge graph; acquiring a work experience text and a self-evaluation text in talent resume, and calculating post matching degree; the method comprises the steps of acquiring talent interview conditions with the post matching degree meeting recruitment requirements, automatically generating a post matching report, and automatically generating the post matching report according to the talent interview conditions with the post matching degree meeting recruitment requirements, wherein the method specifically comprises the following steps: acquiring talent interview audio, predicting job entering willingness, acquiring a talent interview condition scoring table, automatically generating a talent information abstract, acquiring comprehensive evaluation of interview staff on talent interviews, extracting a second-level evaluation index, establishing an emotion corpus based on the comprehensive evaluation of talent interview, and classifying according to emotion intensity; analyzing the variable area in the post matching report, and detecting the total number of the changes of the recruiters of each post, wherein the method specifically comprises the following steps: detecting the change condition of the recruiter; calculating post matching degree of the changed recruiters, sequencing talents, and determining the prepared recruiter talents; according to the change condition of the post personnel and the list of the reserved recruiters, adjusting the content of the post matching report and outputting a new post report, wherein according to the change condition of the post personnel and the list of the reserved recruiters, adjusting the content of the post matching report and outputting the new post report comprises the following steps: updating a talent interview dividing table according to the change condition of post personnel, regenerating a talent information abstract, carrying out key annotation on the report according to the change condition of personnel, and outputting a new post report; and (5) automatically re-layout design and adjustment are carried out on the post matching report.

Further optionally, the acquiring talent data and post data includes:

acquiring talent information and post information; the talent data comprises talent basic information, academic, professional names, project experiences, professional skills and variability judgment; wherein, the basic information comprises names, sexes, birth months, native places, contact modes and graduation institutions of talents; the post data comprises post job allowance text; the variability determination is determined by a variability determination value; the variability character judgment value is the on-duty rate after talent interview passes; if the variable character judging value is not higher than the preset threshold value, judging the variable character, and if the variable character judging value is higher than the preset threshold value, judging the variable character as a non-variable character; processing the content of the post job requiring text, and completing entity extraction tasks in the post job requiring text by adopting a named entity recognition model based on a rule base; the corresponding entities in the output post requirement text include: the academic, the professional name, the professional class and the academic gate class; comprising the following steps: extracting entities in post job-free requirement texts based on a named entity recognition model of a rule base;

The named entity recognition model based on the rule base extracts entities in post job-free requirement texts, and specifically comprises the following steps:

constructing a rule base according to a professional catalog of a common higher school published by the national education department, wherein the rule base comprises an academic library, a professional name library, a professional class library and a academic class library; then embedding the rule base into a HanLP natural language processing tool to construct a named entity recognition model based on the rule base; inputting a post job meeting requirement text, and firstly preprocessing; performing word segmentation operation by using a word segmentation interface in the HanLP natural language processing tool, and removing stop words; then, carrying out regular matching through a HanLP natural language processing tool, and identifying an entity corresponding to the rule base from the post job-qualified requirement text after preprocessing: the academic, the professional name, the professional class and the academic gate class.

Further optionally, the screening talents meeting the post requirement based on the knowledge graph includes:

constructing a subject expert knowledge graph according to a professional catalog of a common higher school published by the national education department, wherein the subject expert knowledge graph comprises a professional name, a professional class, a degree gate class, a relationship between the professional name and the professional class and a relationship between the professional name and the degree gate class, and storing the knowledge graph into a Neo4j database of a background; acquiring corresponding entities in the post requirement text: the academic, the professional name, the professional class and the academic gate class; firstly, detecting whether the professional class and the academic gate class are NULL, if not, matching in a Neo4j database, and outputting the matched professional name; then, combining the entity professional names in the matched professional names and post requirement texts to obtain a final professional name; outputting the final professional name and entity academy in the post requirement text as talent primary screening standard; finally, judging and screening talents meeting the post requirements through matching the final professional names, the academic and the variability.

Further optionally, the step of obtaining the work experience text and the self-evaluation text in the talent resume includes:

the method comprises the steps of obtaining a work experience text and a self-evaluation text in a talent resume meeting the post requirements, preprocessing the texts, dividing words of the work experience text and the self-evaluation text in the talent resume by utilizing a jieba toolkit, removing stop words, and storing the last effective words into a txt document; extracting post matching keywords from the work experience text and the self-evaluation text in the talent resume after pretreatment, firstly screening high-frequency words, converting the high-frequency words into word vectors, and extracting the post matching keywords through clustering; calculating word frequency of each word, and arranging the words in sequence from big to small; selecting the first 30 high-frequency words, calling a Word2vec function to vectorize the high-frequency words, and obtaining a Word vector data table; then inputting the word vector data table into a K-means clustering model, wherein the obtained clustering result is a post matching keyword; and calculating word frequency of the post matching keywords in the post requirement text, and taking the word frequency as the post matching degree.

Further optionally, the obtaining the talent interview condition that the post matching degree meets the recruitment requirement, and automatically generating the post matching report includes:

automatically generating a post matching report according to talent interview conditions with post matching degree meeting recruitment requirements, wherein the post matching report comprises talent basic information, post names, matching degree, post arrival time, job entering willingness, interview results and talent information abstracts; the job entering willingness is predicted and available according to talent interview audio, and the talent information abstract is automatically generated based on a talent interview situation scoring table; setting a data format, generating a post matching report according to the sequence of talent basic information, post names, matching degree, post arrival time, job arrival willingness, interview results and talent information abstract, and outputting the post matching report; comprising the following steps: acquiring talent interview audio, and predicting job entering willingness; acquiring a talent interview condition scoring table and automatically generating a talent information abstract; acquiring comprehensive evaluation of interviewees on talents by interviewees, and extracting a second-level evaluation index; establishing an emotion corpus based on comprehensive evaluation of talent interviews, and dividing the emotion corpus into levels according to emotion intensity;

the method for obtaining talent interview audio and predicting job-entering willingness specifically comprises the following steps:

Acquiring talent interview audio, and analyzing the emotion state of the talents in the interview process according to the audio so as to predict job entering willingness; using the NLPCC2013 audio dataset as a training set and labeling emotional states, including positive and negative; first training LSTM neural network, the method is as follows: firstly, extracting a mel cepstrum coefficient as an audio feature vector of an NLPCC2013 audio data set, inputting the audio feature vector into an LSTM model, and adopting a softmax classification function to realize emotion state classification, wherein the softmax classification function can output a probability value of each emotion state, represents the possibility of belonging to each emotion state category, and the probability value is the emotion state classification result with the maximum probability value; repeating the training process until the output result is consistent with the marked emotion state before training; preprocessing talent interview audio, eliminating noise, dividing the talent interview audio into audio segments with fixed duration, inputting the audio segments into openSMILE to obtain emotion characteristics of talents in the interview process, inputting an LSTM model, and outputting probabilities of different emotion states through a softmax classification function. Taking the emotion state corresponding to the person with high probability as an emotion recognition result of the audio segment, wherein the emotion state is positively marked as 1, and the emotion state is negatively marked as 0; and summing the emotion recognition results of all the audio segments in the talent interview audio to positive and negative results respectively, wherein the job-entering will = positive sum/(positive sum + negative sum) > 1/2.

The step of obtaining the talent interview situation scoring table and automatically generating the talent information abstract specifically comprises the following steps:

acquiring comprehensive evaluation of interviewees on talents by interviewees, and extracting a second-level evaluation index; obtaining the scoring condition of the interviewee on talents, including the scores of various secondary evaluation indexes; and extracting emotion words corresponding to the second-level evaluation indexes according to the scores of the interviewees on talents to generate talent information abstract.

The method for obtaining comprehensive evaluation of interviewee on talents and extracting the second-level evaluation index specifically comprises the following steps:

collecting comprehensive evaluation of interviewees on talents as original data, and using an NLPIR word segmentation algorithm to segment the original data and label the parts of speech. Taking appearance instruments, talkback, stability, compressive capacity and development potential as first-level evaluation indexes, and then extracting candidate second-level evaluation indexes from the original data based on an Apriori algorithm. The comprehensive evaluation text needs to be preprocessed before the Apriori algorithm is used. Firstly, establishing a transaction file, extracting noun phrases or guest phrases in a comprehensive evaluation text, arranging the noun phrases or guest phrases in rows by sentence units, and storing the noun phrases or guest phrases or the guest phrases in a transaction file. And then extracting a candidate comprehensive evaluation index set from the transaction file by using an Apriori algorithm. And calculating the support degree of each phrase, and sequentially judging whether the support degree of the phrase is greater than the minimum support degree, and if so, taking the phrase as a candidate comprehensive evaluation index. And finally, selecting a secondary evaluation index from the candidate comprehensive evaluation index set, and summarizing the final evaluation secondary price index.

The comprehensive evaluation based on talent interview establishes an emotion corpus, and classifies the emotion corpus according to emotion intensity, and specifically comprises the following steps:

establishing an emotion corpus, and conveniently extracting modified emotion words corresponding to each secondary index according to each secondary index; aiming at the establishment of an emotion corpus, firstly, extracting a set without the secondary adverbs from a secondary evaluation index set to be used as an emotion extraction set; and obtaining emotion modifier words corresponding to different nouns or guest phrases according to emotion word extraction rules, and classifying according to emotion intensity.

Further optionally, the analyzing the variable area in the post-match report, and detecting the total number of recruiter variations for each post comprises:

analyzing a variable area in the post matching report, wherein the variable area comprises the time to post and the job entering intention; the talents passing the interview are interviewed every two months, and the post time and the job entering wish are collected again; the job entering willingness adopts a telephone return visit mode to acquire call audio of talents, then emotion recognition is carried out on all audio segments in the talent audio, and positive and negative results are summed up respectively to calculate the job entering willingness; uploading the updated data to a database; comprising the following steps: detecting the change condition of the recruiter;

The detecting the change condition of the recruiter specifically includes:

detecting a change in the recruiter, including a reduced number of people and an increased number of people; judging whether the reduced number of people belongs to the low-willingness crowd or not through job entering willingness; counting a low willingness crowd list of each post and outputting the list; the increased crowd is the number of the talents newly increased in two months from the last time of generating the report, the talents list of the pass interviews newly increased in each post is counted and output; and respectively counting the total number of the reduced and increased positions and the total number of the changed positions and outputting the total number.

Further optionally, the calculating the post matching degree of the recruiter after the change, ordering the talents, and determining the prepared talents includes:

the method comprises the steps of calling talent data which pass the interview result from an updated database, wherein the talent data comprises talent basic information, post names, matching degree, time to post, job entering willingness and talent information abstract; acquiring a post recruitment planned number n and planned arrival time; firstly, screening talent data of which talents arrive at the post time and are earlier than or equal to the scheduled talent data of which the talents arrive at the post time, then re-judging the matching degree, sorting the talent data from big to small according to the new matching degree, counting the talent number and judging; if talent number > =post recruitment plan number, taking the first n talents as the prepared talents; if the talent quantity is less than the post recruitment planned number, calculating a talent difference value s, and sorting talents which are later than the planned on-post time from big to small according to the new matching degree, screening the first s talents, and taking talents with the post time being earlier than or equal to the planned on-post time and the first s talents which are later than the planned on-post time as the preliminary recruitment talents.

Further optionally, the adjusting the post matching report content and outputting a new post report based on the post personnel change and the list of prepared recruiters includes:

acquiring talent data of a talent list and data of the prepared recruitment; reporting the adjustment content including update items and add items; the update items comprise time to post, job entering intention, matching degree and talent abstract; the added items comprise whether the recruiter is a prepared recruiter or not, and each post recruiter reduces and increases the total number and the total number after the recruiter changes; firstly, updating the arrival time, the job entering willingness, the matching degree and the talent abstract in a report according to talent data with changed data; then adjusting the added items according to the change condition of the recruiter and the list of the prepared recruiters; comprising the following steps: updating the talent interview scoring table according to the change condition of post personnel, and regenerating a talent information abstract; the report is marked with emphasis according to personnel change conditions, and a new post report is output;

updating a talent interview scoring table according to the change condition of post personnel, and regenerating a talent information abstract, which comprises the following steps:

calling talent interview condition scoring tables with interview results passing from the updated database; taking the secondary indexes in the talent interview situation classifying table as key words, and calculating word frequency of the key words in the post requirement text; outputting the keyword with the highest word frequency, and endowing the keyword with the highest word frequency with a score importance weight, and distributing the score importance weights of the other keywords according to an average number; re-calculating talent interview scoring tables according to the score importance weights; and regenerating the talent information abstract according to the new talent interview scoring table.

The method for marking the report with emphasis according to the personnel change condition and outputting a new post report specifically comprises the following steps:

acquiring a low willingness crowd list of each post and a newly added talent list passing the interview; firstly, carrying out key annotation on low willingness crowd at each post, wherein the method is as follows; setting constraint on the post report generation process, and sequentially detecting whether each person belongs to a low-willingness crowd list; if the character belongs to the talents, the operation is not performed, and if the character belongs to the talents, the bold fonts are used for marking the talents' job-entering willingness and the matching degree; then, the newly added talents passing the interview are marked with emphasis; marking a basic information column of the talents by using red fonts according to the newly added talent list passing the interview; the new post report is output, including talent basic information, post name, degree of matching, time to post, willingness to enter, interview results, talent information summary, whether it is a prepared recruiter, and reporting the final total number of individual post applications reduced and increased and total number of applied applications changed.

Further optionally, the automatically re-laying out and adjusting the post matching report includes:

acquiring all text box elements in the post matching report by using OpenXMLSDK, and simultaneously acquiring the sizes and positions of the text box elements; the size is the horizontal length and the vertical width of the text box element, and the position is the horizontal coordinate and the vertical coordinate of the upper left corner point of the text box element in the page; aiming at text box elements in the post matching report, firstly, strong fonts are optimized, important contents are highlighted, interface disorder caused by excessive font styles is avoided, then, the font styles are optimized, and the legibility of the text is enhanced; firstly, adjusting the positions of text boxes to enable horizontal coordinates of upper left corner points of all the text boxes in a page to be consistent; traversing all the text box elements, and sequentially judging whether each text box element meets the constraint: if the horizontal length is equal to the sum of the horizontal coordinates of the upper left corner point of the text box element in the page, not operating, otherwise changing the horizontal length of the text box element until the constraint is met; then optimizing the emphasized fonts; the accentuated fonts include bold, slant, underlined style fonts; counting the number of the emphasized fonts in all text box elements, and calculating the emphasis font duty ratio; presetting a first threshold value, and judging whether the emphasis font duty ratio exceeds the first threshold value; if the text length is smaller than or equal to the first threshold value, not operating, otherwise, sequencing the text box elements according to the text length, and preferentially taking out the emphasized fonts in the disarmed text; finally, adjusting the font style of the text box element; counting the frequency of the occurrence of fonts of different styles in all text box elements, and calculating the style ratio of the fonts; and taking the style with the highest font style ratio as a main body style, and changing the font style in all text box elements except the emphasized font into the main body style.

The technical scheme provided by the embodiment of the invention can have the following beneficial effects:

the invention can match talent data with post data, generate post matching reports according to the post matching degree, enable recruiters to know basic information of talents, applied post names, matching degree, time to post, job entering willingness, interview results and talent information abstract, and generate post matching reports. Meanwhile, the post matching report can be modified according to the change condition of talents, so that the post matching report is more suitable for new adjustment of the information of the period from the time of the application of the recruiter to the time of the job, and the attractiveness and the information accuracy of the report are maintained.

Drawings

FIG. 1 is a flow chart of a post match report generation method of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and specific embodiments.

The post matching report generation method in this embodiment specifically may include:

step 101, talent data and post data are obtained.

Acquiring talent information and post information; the talent data comprises talent basic information, academic, professional names, project experiences, professional skills and variability judgment; wherein, the basic information comprises names, sexes, birth months, native places, contact modes and graduation institutions of talents; the post data comprises post job allowance text; the variability determination is determined by a variability determination value; the variability character judgment value is the on-duty rate after talent interview passes; if the variable character judging value is not higher than the preset threshold value, judging the variable character, and if the variable character judging value is higher than the preset threshold value, judging the variable character as a non-variable character; processing the content of the post job requiring text, and completing entity extraction tasks in the post job requiring text by adopting a named entity recognition model based on a rule base; the corresponding entities in the output post requirement text include: the academic, the professional name, the professional class and the academic gate class. If the preset character judgment value threshold is 60%, and the character judgment value of Zhang three is 45%, the variable character of Zhang three is judged as the variable character; for example, the acquired post job assignment text is ' financial, economic, management, financial accounting and other related professional priority ', 2020, 2021 at home and abroad, known college students, and major study students '. The corresponding entity in the named entity recognition model output post requirement based on the rule base is { academy: family, major study, professional name: NULL, specialty class: finance class, economics class, management class, finance class, academic category: economics, management }. Named entity recognition, also known as nomination, is commonly used to identify entities contained in text that have a particular meaning or meaning. Named entity recognition serves to automatically identify entities from a given post job title text and tag the corresponding entity type. The step automatically identifies and extracts the related information of talents' academic, professional names, professions and academic categories from post job requirement texts.

And extracting the entity in the post job completion requirement text based on a named entity identification model of the rule base.

Constructing a rule base according to a professional catalog of a common higher school published by the national education department, wherein the rule base comprises an academic library, a professional name library, a professional class library and a academic class library; then embedding the rule base into a HanLP natural language processing tool to construct a named entity recognition model based on the rule base; inputting a post job meeting requirement text, and firstly preprocessing; performing word segmentation operation by using a word segmentation interface in the HanLP natural language processing tool, and removing stop words; then, carrying out regular matching through a HanLP natural language processing tool, and identifying an entity corresponding to the rule base from the post job-qualified requirement text after preprocessing: the academic, the professional name, the professional class and the academic gate class. For example, job designation text is 'family and master reading students, and profession is not limited'. The text is first entered into the word segmentation interface in the HanLP natural language processing tool to obtain a preprocessed text: the method is applicable to the fields of the family, the reach, the filling, the reading, the students, the profession and the limitation. The entity identified from the rule base is { academy: family, major name: NULL, specialty class: NULL, degree gate class: NULL). The academic library in the rule library comprises special departments, family departments, master study groups and doctor study groups, and the professional name library, the professional class library and the academic department library comprise different professional names, professional classes and academic department classes extracted from professional catalogues of common higher schools published by the national education department.

Step 102, screening talents meeting the post requirements based on the knowledge graph.

Constructing a subject expert knowledge graph according to a professional catalog of a common higher school published by the national education department, wherein the subject expert knowledge graph comprises a professional name, a professional class, a degree gate class, a relationship between the professional name and the professional class and a relationship between the professional name and the degree gate class, and storing the knowledge graph into a Neo4j database of a background; acquiring corresponding entities in the post requirement text: the academic, the professional name, the professional class and the academic gate class; firstly, detecting whether the professional class and the academic gate class are NULL, if not, matching in a Neo4j database, and outputting the matched professional name; then, combining the entity professional names in the matched professional names and post requirement texts to obtain a final professional name; outputting the final professional name and entity academy in the post requirement text as talent primary screening standard; finally, judging and screening talents meeting the post requirements through matching the final professional names, the academic and the variability. For example, the post requires the corresponding entity { school } in the text: master study, professional name: business management, specialty: agricultural economy management class, academic department class: management }. And (5) developing agriculture and forestry economic management and rural areas according to the matching result of the professions and the degree gate. The entity names in the matched name and post requirement text are then summed, so the final name is: business management, agriculture and forestry economic management and rural area development. The talent primary screening standard is as follows: { academy: master study, professional name: the discipline expert knowledge map for the development of the rural area comprises 703 professional names, 92 professional classes, 12 degree categories, and 2165 relationship between the professional names and the professional classes. The Neo4j database is a graph database, and can store more data relationships than other databases, and the Neo4j database can be used for convenient storage and query because the discipline professional knowledge graph comprises a large amount of relationship data.

And step 103, acquiring a work experience text and a self-evaluation text in the talent resume, and calculating the post matching degree.

The method comprises the steps of obtaining a work experience text and a self-evaluation text in a talent resume meeting the post requirements, preprocessing the texts, dividing words of the work experience text and the self-evaluation text in the talent resume by utilizing a jieba toolkit, removing stop words, and storing the last effective words into a txt document; the post matching keywords are extracted from the work experience text and the self-evaluation text in the talent resume after pretreatment, the high-frequency vocabulary is firstly needed to be screened out, then the high-frequency vocabulary is converted into word vectors, and then the post matching keywords are extracted through clustering treatment. The word frequency of each word is calculated and arranged in order from large to small. And selecting the first 30 high-frequency words, and calling a Word2vec function to vectorize the high-frequency words to obtain a Word vector data table. And then inputting the word vector data table into a K-means clustering model, wherein the obtained clustering result is the post matching keyword. And calculating word frequency of the post matching keywords in the post requirement text, and taking the word frequency as the post matching degree. For example, "master office software" is divided into words, "master/office/software". The K-means clustering model procedure is as follows: firstly, randomly selecting K word vector data as a clustering center, then sequentially calculating cosine similarity between the rest word vectors and the clustering center, and classifying the clustering center with the minimum cosine similarity and the corresponding word vector into one type. And finally, taking the average value of the various Chinese word vectors as a new clustering center. And repeating the above process until the clustering center is not changed any more, and outputting the post matching keywords. Examples of post matching keywords are as follows { Category one: communication ability, office software, planning ability, leading ability interpersonal interaction, data analysis, writing, and industry background; category two: policy regulations, performance management, cost accounting, enterprise culture, contract management, employee relationships, and employee training; category three: motivating rules, management, strategic thinking, achievement guidance, solving problems, decision making and reform; category four: team spirit, self-driving, compression resistance, saluting, proactive, coordination, responsibility, and petty, ken. For example, post requirement text a is: the self-evaluation text B of talents using office software is: and (5) mastering office software. The word frequency of the office software in A is 1, and if the post requirement text and talent self-evaluation have no other contents, the 1 is the post matching degree.

Step 104, acquiring talent interview conditions with post matching degree meeting recruitment requirements, and automatically generating a post matching report.

Automatically generating a post matching report according to talent interview conditions with post matching degree meeting recruitment requirements, wherein the post matching report comprises talent basic information, post names, matching degree, post arrival time, job entering willingness, interview results and talent information abstracts; the job entering willingness is predicted and available according to talent interview audio, and the talent information abstract is automatically generated based on a talent interview situation scoring table; setting a data format, generating a post matching report according to the sequence of talent basic information, post names, matching degree, post arrival time, job arrival willingness, interview results and talent information abstract, and outputting the post matching report. For example, talent basic information: zhang san, man, 1990.12.12, sichuan Cheng Du, 18711110000, qinghua university. Post name: a blockchain engineer. Matching degree: 0.85. on duty: 2023.06.10. job entering willingness: 0.6. interview results: through the device. Talent information abstract the personnel instrument is correct and holds state, the language expression capability is strong, the post working experience is 3 years, the post matching degree is better, the psychology is mature, the personnel instrument has certain pressure resistance capability, and the personnel instrument can reach the company standard through training.

And acquiring talent interview audio, and predicting job entering willingness.

Acquiring talent interview audio, and analyzing the emotion state of the talents in the interview process according to the audio so as to predict job entering willingness; using the NLPCC2013 audio dataset as a training set and labeling emotional states, including positive and negative; first training LSTM neural network, the method is as follows: firstly, extracting a mel cepstrum coefficient as an audio feature vector of an NLPCC2013 audio data set, inputting the audio feature vector into an LSTM model, and adopting a softmax classification function to realize emotion state classification, wherein the softmax classification function can output a probability value of each emotion state, represents the possibility of belonging to each emotion state category, and the probability value is the emotion state classification result with the maximum probability value; repeating the training process until the output result is consistent with the marked emotion state before training; preprocessing talent interview audio, eliminating noise, dividing the talent interview audio into audio segments with fixed duration, inputting the audio segments into openSMILE to obtain emotion characteristics of talents in the interview process, inputting an LSTM model, and outputting probabilities of different emotion states through a softmax classification function. Taking the emotion state corresponding to the person with high probability as an emotion recognition result of the audio segment, wherein the emotion state is positively marked as 1, and the emotion state is negatively marked as 0; and summing the emotion recognition results of all the audio segments in the talent interview audio to positive and negative results respectively, wherein the job-entering will = positive sum/(positive sum + negative sum) > 1/2. For example, if the total length of the three interview audio is 5 minutes, the interview audio is divided into audio segments of a fixed length of 15 seconds, and a total of 20 audio segments. If the probabilities of outputting different emotion states by the softmax classification function are respectively: positive 0.3456, negative: 0.6544, the piece of audio belongs to negative emotion. Assume that the result after recognition by the LSTM neural network is: positive 12, negative 8, then the job-entering willingness of sheet three is 12/20=0.6. The NLPCC2013 audio dataset comes from a microblog and includes more than 4 thousand chinese voices. Mel-frequency cepstrum coefficients are widely used for a language recognition function in the field of sound processing, and the extraction method is as follows: a section of audio is decomposed into a plurality of frames, and the audio signal is strengthened by a high-pass filter and then is subjected to Fourier transform so as to transform the voice signal into a frequency domain. The spectrum obtained by each frame is passed through a mel filter to obtain mel scale, logarithm is taken, and then inverse discrete fourier transform is carried out to obtain cepstrum domain, and the mel cepstrum coefficient is the amplitude of the cepstrum domain. LSTM models are commonly used for language emotion recognition, and the LSTM model is the dominant translation and speech recognition technology for many large companies. After the audio feature vector is input into the LSTM model, the feature vector is spliced through a fully connected network layer, then a real vector between 0 and 1 is output through a sigmoid function, and the real vector between 0 and 1 is input into a softmax classification function to obtain probabilities of different emotion states.

And acquiring a talent interview condition scoring table and automatically generating a talent information abstract.

Acquiring comprehensive evaluation of interviewees on talents by interviewees, and extracting a second-level evaluation index; obtaining the scoring condition of the interviewee on talents, including the scores of various secondary evaluation indexes; and extracting emotion words corresponding to the second-level evaluation indexes according to the scores of the interviewees on talents to generate talent information abstract. For example, the Zhang three-level index score is { image gas: 10, interview attitudes: 9, working experience: 8, emotional manifestations: 7, mental maturity: and 6, generating talent information abstract, wherein the generated talent information abstract is good in image and gas quality, serious in interview attitude, sufficient in working experience, good in emotion expression and general in mental maturity.

And (5) acquiring comprehensive evaluation of interviewee on talents by interviewee and extracting a second-level evaluation index.

Collecting comprehensive evaluation of interviewees on talents as original data, and using an NLPIR word segmentation algorithm to segment the original data and label the parts of speech. Taking appearance instruments, talkback, stability, compressive capacity and development potential as first-level evaluation indexes, and then extracting candidate second-level evaluation indexes from the original data based on an Apriori algorithm. The comprehensive evaluation text needs to be preprocessed before the Apriori algorithm is used. Firstly, establishing a transaction file, extracting noun phrases or guest phrases in a comprehensive evaluation text, arranging the noun phrases or guest phrases in rows by sentence units, and storing the noun phrases or guest phrases or the guest phrases in a transaction file. And then extracting a candidate comprehensive evaluation index set from the transaction file by using an Apriori algorithm. And calculating the support degree of each phrase, and sequentially judging whether the support degree of the phrase is greater than the minimum support degree, and if so, taking the phrase as a candidate comprehensive evaluation index. And finally, selecting a secondary evaluation index from the candidate comprehensive evaluation index set, and summarizing the final evaluation secondary price index. For example, the NLPIR word segmentation algorithm is a chinese word segmentation algorithm proposed by the chinese academy of sciences. The word segmentation result of 'Zhang Sanyi Liang Zhuang' is 'Zhang Sanyi Zhuang/n Liang/a Zhuang/a,/wd work/vn experience/n Fu/a'. Where nr is a person name, n is a noun, a is an adjective, wd is a comma, and vn is a named verb. The transaction file is a text file. Evaluation text: "frequently attend team games, there is a strong sense of collective reputation". The "transaction file contents are" group/n match/vn, collective/n honor sense/n ", and these contents are stored in the same line. The support degree=the frequency of phrase occurrence/total number of lines, the minimum support degree adopted according to the common processing method in the literature is 0.1, and if the transaction file only has 'group/n match/vn and group/n honor sense/n', the support degree of the 'group/n match/vn' is 1/2>0.1, and the candidate comprehensive evaluation index can be obtained. And when the secondary evaluation indexes are selected from the candidate comprehensive evaluation index set, candidate secondary evaluation indexes with similar expression meanings are removed. Examples of final evaluation indexes are as follows: { instrumentation: skin-care appearance, wearing and making up, and the appearance and the quality of gas; speaking to mention: interview attitudes, term specifications and behavior; stability degree: job motivation, job planning and working experience; pressure resistance capability: emotional manifestations, look manifestations, mood manifestations; development potential: the heart intelligence maturity, the intention definition, the skill proficiency, and the actual secondary index is determined by the original data.

And establishing an emotion corpus based on comprehensive evaluation of talent interviews, and classifying according to emotion intensity.

Establishing an emotion corpus, and conveniently extracting modified emotion words corresponding to each secondary index according to each secondary index; aiming at the establishment of an emotion corpus, firstly, extracting a set without the secondary adverbs from a secondary evaluation index set to be used as an emotion extraction set; according to emotion word extraction rules, emotion modifier words corresponding to different nouns or guest phrases are obtained, emotion strength classification levels are divided into 5 levels, for example, the emotion strength level of the emotion modifier words is divided into 5 levels, the 5-level strongest score interval is 9-10 and 4 levels, the score interval is 8-9, the 3 level is 7-8, the 2 level is 7-6, and the 1 level is less than 6. Different words correspond to different emotion levels. The emotion word extraction rules of the emotion corpus are as follows.

The example of the finally obtained second-level index corresponding emotion words is as follows: { image gas texture: good, excellent, general, good … …, in-plane attitude: righting, positive, serious … … }.

Step 105, analyzing the variable area in the post matching report, and detecting the total number of the applied personnel changes of each post.

Analyzing a variable area in the post matching report, wherein the variable area comprises the time to post and the job entering intention; the talents passing the interview are interviewed every two months, and the post time and the job entering wish are collected again; the job entering willingness adopts a telephone return visit mode to acquire call audio of talents, then emotion recognition is carried out on all audio segments in the talent audio, and positive and negative results are summed up respectively to calculate the job entering willingness; and uploading the updated data to a database. For example, the time between Zhang three and Shift is still 2023.06.10, but according to the call-back audio, the job-entering intention is 0.2, and the time will be on Shift: 2023.06.10 and job entering willingness: 0.2 is re-uploaded to the database.

And detecting the change condition of the recruiter.

Detecting a change in the recruiter, including a reduced number of people and an increased number of people; judging whether the reduced number of people belongs to the low-willingness crowd or not through job entering willingness; counting a low willingness crowd list of each post and outputting the list; the increased crowd is the number of the talents newly increased in two months from the last time of generating the report, the talents list of the pass interviews newly increased in each post is counted and output; and respectively counting the total number of the reduced and increased positions and the total number of the changed positions and outputting the total number. For example, assuming the first threshold is 0.5, if the recruiter's intent to engage <0.5 indicates that he is not coming, the recruiter's intent to engage > = 0.5 indicates that he may come. Assuming that the last report was generated at a time of 2022.9.9, the total number of talents passing the interview during the period 2022.9.9-2022.11.9 is an increased number of people. The first threshold value averages all of the recruiter's willingness to engage.

And 106, calculating post matching degree after the change of the recruiter, sequencing talents, and determining the prepared recruiter talents.

The method comprises the steps of calling talent data which pass the interview result from an updated database, wherein the talent data comprises talent basic information, post names, matching degree, time to post, job entering willingness and talent information abstract; acquiring a post recruitment planned number n and planned arrival time; firstly, screening talent data of which talents arrive at the post time and are earlier than or equal to the scheduled talent data of which the talents arrive at the post time, then re-judging the matching degree, sorting the talent data from big to small according to the new matching degree, counting the talent number and judging; if talent number > =post recruitment plan number, taking the first n talents as the prepared talents; if the talent quantity is less than the post recruitment planned number, calculating a talent difference value s, and sorting talents which are later than the planned on-post time from big to small according to the new matching degree, screening the first s talents, and taking talents with the post time being earlier than or equal to the planned on-post time and the first s talents which are later than the planned on-post time as the preliminary recruitment talents. Since there are fewer people and fewer people to the application, it is necessary to recalculate the degree of matching based on the changed situation in order to select the most appropriate talent. New match = (4 x match +4 x job entering intent +2 x on post day)/10, on post day = | plan on post time-on post time|. Talent difference = job recruitment plan number-talent number. For example, the job recruitment schedule is 3 and the schedule to job time is 2023.06.10. The original talent list is Zhang Sanyi > Liqu, the return visit shows that Zhang Sanzhu is declined, two talents Liu Liu and seven are newly added to the post, at the moment, the new matching degree is recalculated by combining the original matching degree of talents, the job entering intention and the number of days on post, and the order is that Liqu is four > Liu Liu > Liqu five > seven > three. And then, firstly screening out talents whose talents arrive at the post time earlier than or equal to the planned arrival time (assuming that the seven-seven arrival time is later than the planned arrival time), and sorting the talents from big to small according to the new matching degree, wherein the result is that Li four > Liu Liu > Wang five > Zhang three, and Li four, liu Liu and Wang five are taken as the preliminary recruitment talents. If the number of post recruitment plans is 5, the talents later than the planned time to post are ordered according to the new matching degree, the first position is taken as the preliminary recruiter talents, and the four, liu Liu, five, three, seven and seven of the first position are taken as the preliminary recruiter talents.

Step 107, adjusting the post matching report content according to the post personnel change condition and the list of the prepared recruiters, and outputting a new post report.

Acquiring talent data of a talent list and data of the prepared recruitment; reporting the adjustment content including update items and add items; the update items comprise time to post, job entering intention, matching degree and talent abstract; the added items comprise whether the recruiter is a prepared recruiter or not, and each post recruiter reduces and increases the total number and the total number after the recruiter changes; firstly, updating the arrival time, the job entering willingness, the matching degree and the talent abstract in a report according to talent data with changed data; and then adjusting the added items according to the change condition of the recruiter and the list of the prepared recruiters. For example, the adjusted post report should be presented in tabular form, with the tabular contents including talent base information, post name, degree of matching, time to post, willingness to enter job, talent information summary, whether it is a prepared recruiter talent, the total number of individual post takers being reduced and increased, and the total number of changed takers. The time to post, the job intention and the matching degree in the report are all updated data.

Updating the talent interview scoring table according to the change condition of post personnel, and regenerating the talent information abstract.

Calling talent interview condition scoring tables with interview results passing from the updated database; taking the secondary indexes in the talent interview situation classifying table as key words, and calculating word frequency of the key words in the post requirement text; outputting the keyword with the highest word frequency, and endowing the keyword with the highest word frequency with a score importance weight, and distributing the score importance weights of the other keywords according to an average number; re-calculating talent interview scoring tables according to the score importance weights; and regenerating the talent information abstract according to the new talent interview scoring table. The score importance weight = 1-total number of talent matching ranks/recruiters after change. For example, the original scoring table of king five has scores of 5, liu Liu each score was 4, 6, 4. And if the word frequency in the post requirement text is highest, assigning a score importance weight of 1-2/5=3/5 to the image air quality item. The score importance weights of the remaining four indices are (2/5)/4=1/10. The newly generated score table is: royal five 0.5, 3, 0.5, liu Liu 0.4, 3.6, 0.4. The method for obtaining the new score list and automatically generating the new talent information abstract is the same as the step 3-2.

And (5) marking the report with emphasis according to personnel change conditions, and outputting a new post report.

Acquiring a low willingness crowd list of each post and a newly added talent list passing the interview; firstly, carrying out key annotation on low willingness crowd at each post, wherein the method is as follows; setting constraint on the post report generation process, and sequentially detecting whether each person belongs to a low-willingness crowd list; if the character belongs to the talents, the operation is not performed, and if the character belongs to the talents, the bold fonts are used for marking the talents' job-entering willingness and the matching degree; then, the newly added talents passing the interview are marked with emphasis; marking a basic information column of the talents by using red fonts according to the newly added talent list passing the interview; the new post report is output, including talent basic information, post name, degree of matching, time to post, willingness to enter, interview results, talent information summary, whether it is a prepared recruiter, and reporting the final total number of individual post applications reduced and increased and total number of applied applications changed. For example, zhang three is marked with bold fonts because his job-in intent is judged as a low intent crowd. Liu Liu is a newly added talent that passes the interview, so his basic information fields are bolded with red words. Because the low willingness crowd changes the personnel because of the change of the job-entering willingness, the interviewer needs to pay attention to the job-entering willingness of the part of people and the matching degree influenced by the job-entering willingness. The newly added talents passing the interview are likely to become ready-to-record talents, and therefore need to be marked with a more striking red color.

Step 108, automatically re-layout design and adjustment is performed on the post matching report.

And acquiring all text box elements in the post matching report by using the OpenXMLSDK, and simultaneously acquiring the sizes and the positions of the text box elements. The size is the horizontal length and the vertical width of the text box element, and the position is the horizontal coordinate and the vertical coordinate of the upper left corner point of the text box element in the page. Aiming at text box elements in the post matching report, firstly, strong fonts are optimized, important contents are highlighted, interface disorder caused by excessive font styles is avoided, then, the font styles are optimized, and the legibility of the text is enhanced. Firstly, the positions of the text boxes are adjusted, so that the horizontal coordinates of the upper left corner points of the text boxes in the page are consistent. Traversing all the text box elements, and sequentially judging whether each text box element meets the constraint: and if the horizontal length is equal to the sum of the horizontal coordinates of the upper left corner point of the text box element in the page, not operating, otherwise, changing the horizontal length of the text box element until the constraint is met. The emphasized fonts are then optimized. The accentuated fonts include bold, slant, underlined style fonts. And counting the number of the emphasized fonts in all text box elements, and calculating the emphasis font duty ratio. And presetting a first threshold value, and judging whether the emphasis font duty ratio exceeds the first threshold value. And if the text length is smaller than or equal to the first threshold value, not operating, otherwise, sequencing the text box elements according to the text length, and preferentially removing the emphasized fonts in the disarmed text. Finally, the font style of the text box element is adjusted. And counting the frequency of the occurrence of fonts of different styles in all text box elements, and calculating the font style ratio. And taking the style with the highest font style ratio as a main body style, and changing the font style in all text box elements except the emphasized font into the main body style. For example, openXMLSDK can parse element objects in a post match report. OpenXMLSDK is an open source class library provided by Microsoft and can analyze XML documents such as docx/xlsx/pptx. Each element in the post matching report belongs to an object, and the process of acquiring the objects or operating the objects is the process of analyzing the document. Emphasis font duty = emphasis font number/total number of words in the post match report. The first threshold is set to 10%, and according to the design experience of the mainstream document format, when the content emphasized by the document accounts for 10% or less of the whole page content, the reader can be effectively attracted, and the word emphasis effect can be better exerted. Font style duty = number of fonts/number of total words in the post match report for a font style.

The preferred embodiments of the invention disclosed above are intended only to assist in the explanation of the invention. The preferred embodiments are not intended to be exhaustive or to limit the invention to the precise form disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best understand and utilize the invention. The invention is limited only by the claims and the full scope and equivalents thereof.

Claims

1. A post matching report generation method, the method comprising:

2. The method of claim 1, wherein the acquiring talent and post information comprises:

3. The method of claim 1, wherein the screening talents meeting post requirements based on the knowledge-graph comprises:

4. The method of claim 1, wherein the obtaining the work experience text and the self-evaluation text in the talent resume, calculating the post matching degree, comprises:

5. The method of claim 1, wherein the obtaining talent interviews with post matching degree meeting recruitment requirements automatically generates a post matching report, comprising:

Acquiring talent interview audio, and analyzing the emotion state of the talents in the interview process according to the audio so as to predict job entering willingness; using the NLPCC2013 audio dataset as a training set and labeling emotional states, including positive and negative; first training LSTM neural network, the method is as follows: firstly, extracting a mel cepstrum coefficient as an audio feature vector of an NLPCC2013 audio data set, inputting the audio feature vector into an LSTM model, and adopting a softmax classification function to realize emotion state classification, wherein the softmax classification function can output a probability value of each emotion state, represents the possibility of belonging to each emotion state category, and the probability value is the emotion state classification result with the maximum probability value; repeating the training process until the output result is consistent with the marked emotion state before training; preprocessing talent interview audio, eliminating noise, dividing the talent interview audio into audio segments with fixed duration, inputting the audio segments into openSMILE to obtain emotion characteristics of talents in the interview process, inputting an LSTM model, and outputting probabilities of different emotion states through a softmax classification function; taking the emotion state corresponding to the person with high probability as an emotion recognition result of the audio segment, wherein the emotion state is positively marked as 1, and the emotion state is negatively marked as 0; the emotion recognition results of all the audio segments in the talent interview audio are respectively summed up as positive and negative results, and the job-entering will = positive sum/(positive sum + negative sum) > 1/2;

acquiring comprehensive evaluation of interviewees on talents by interviewees, and extracting a second-level evaluation index; obtaining the scoring condition of the interviewee on talents, including the scores of various secondary evaluation indexes; extracting emotion words corresponding to the second-level evaluation indexes according to the scores of the interviewees on talents to generate talent information abstract;

collecting comprehensive evaluation of interviewees on talents as original data, and using an NLPIR word segmentation algorithm to segment the original data and label the parts of speech; taking appearance instruments, talkback, stability, compressive capacity and development potential as first-level evaluation indexes, and then extracting candidate second-level evaluation indexes from the original data based on an Apriori algorithm; preprocessing the comprehensive evaluation text before using the Apriori algorithm; firstly, establishing a transaction file, extracting noun phrases or guest phrases in a comprehensive evaluation text, arranging the noun phrases or guest phrases in rows by sentence units, and storing the noun phrases or guest phrases or the guest phrases in a transaction file; then extracting a candidate comprehensive evaluation index set from the transaction file by using an Apriori algorithm; calculating the support degree of each phrase, and sequentially judging whether the support degree of the phrase is greater than the minimum support degree, if so, taking the phrase as a candidate comprehensive evaluation index; finally, selecting a secondary evaluation index from the candidate comprehensive evaluation index set, and summarizing the final evaluation secondary price index;

6. The method of claim 1, wherein analyzing the variable area in the post-match report, detecting a total number of applied member changes for each post, comprises:

The detecting the change condition of the recruiter specifically includes:

7. The method of claim 1, wherein the calculating the post-change degree of match for the recruiter, ordering talents, and determining the prepared talents comprises:

8. The method of claim 1, wherein said adjusting the post match report content and outputting a new post report based on post personnel change conditions and a list of prepared recruiters, comprises:

Calling talent interview condition scoring tables with interview results passing from the updated database; taking the secondary indexes in the talent interview situation classifying table as key words, and calculating word frequency of the key words in the post requirement text; outputting the keyword with the highest word frequency, and endowing the keyword with the highest word frequency with a score importance weight, and distributing the score importance weights of the other keywords according to an average number; re-calculating talent interview scoring tables according to the score importance weights; regenerating talent information abstract according to the new talent interview scoring table;

9. The method of claim 1, wherein the automatically re-laying out and adjusting of the post-match report comprises: