Human resource platform management system and method based on multi-source data analysis

Abstract

The invention discloses a human resource platform management system and method based on multi-source data analysis, and relates to the technical field of human resource management.The human resource platform management method comprises the steps that interview videos and question text data of interaction between an interviewer and a robot are extracted, and a correlation index is established through timestamp alignment and content matching rules; according to historical recruitment data, constructing a post capability assessment dimension library containing first / second-level capability assessment dimensions and basic scores; segmenting the video into segments according to a round of question and answer, and adding dimension association and content description labels; calculating a fragment dimension matching degree and a cosine similarity content matching degree, and screening target video fragments of which the comprehensive matching degree reaches a threshold value; scoring and sorting the target fragments by using a multi-dimensional scoring model to generate a fragment scoring table; calculating an interview comprehensive score of the interviewer and generating an interview comprehensive score table of the associated tags; and the interviewer inputs screening conditions, and the system locks personnel and calls corresponding fragments.

Description

Technical Field

[0001] This invention relates to the field of human resource management technology, specifically to a human resource platform management system and method based on multi-source data analysis. Background Technology

[0002] In today's fiercely competitive talent market, efficient and precise recruitment processes are crucial for companies to gain core competitiveness. Traditional in-person interviews are limited by time and space, costly, and inefficient. To address this challenge, many companies have begun adopting asynchronous video interviews. In this method, job seekers interact with an intelligent chatbot on a platform, answering questions from a question bank. The entire interview process is recorded and generated as a video file, which interviewers can then view and evaluate asynchronously at their convenience. This approach allows interviewers to schedule their time flexibly, avoiding the hassle of coordinating multiple schedules, and has become an important method for modern human resource screening.

[0003] However, this method still has significant limitations. For example, individual interview videos are typically long and vary in length depending on the candidate's responses. When faced with a large number of job seekers, interviewers still face a considerable workload, resulting in significant time investment, visual fatigue, and low screening efficiency. Therefore, there is an urgent need for a method that can automatically parse and evaluate interview videos to solve these problems. Summary of the Invention

[0004] The purpose of this invention is to provide a human resources platform management system and method based on multi-source data analysis to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a human resources platform management method based on multi-source data analysis, the method comprising: Extract all interview video data and question text data from the interactions between interviewees and interview robots stored on the human resources platform; establish a correlation index between the interview video data and the question text data; obtain the core competency requirement parameters for the interview positions and construct a job competency assessment dimension library; divide each interview video data into several video segments and add attribute tags to each video segment; Based on the job competency assessment dimension library, several video segments after the interview video data were segmented were filtered to obtain the target video segment; The preset multi-dimensional scoring model is retrieved to score the behavior of the interviewee in the target video segment, and the segment score of each target video segment is obtained. The segment scores of the same type of target segments are sorted to generate a target segment score table. Extract several target video segment scores for the same interviewee under various job competency dimensions, calculate the interviewee's overall interview score, and generate an overall interview score table; associate the corresponding attribute tags of the target video segments with the overall interview score table; The interviewer inputs screening criteria into the human resources platform. The human resources platform then locks in the interviewees who meet the screening criteria from the comprehensive interview scoring table, and retrieves the corresponding target video clips for review based on the associated attribute tags.

[0006] Furthermore, establishing a correlation index between the interview video data and the question text data includes: An association index is established between the interview video data and the question text data through preset association rules; the association rules include timestamp alignment rules and content matching rules; the timestamp alignment rules are established by extracting the interaction time interval of the interview video data and matching it with the timestamp of the question text data; the content matching rules are established by translating the audio content of the interview video data into text data and calculating the content similarity between the text data and the question text data. The aforementioned question text data refers to the standardized text data corresponding to all questions and standard answers raised by the interview robot to the interviewee during the interaction process. It is also the question carrier that is pre-stored by the human resources platform and output by the robot according to the set logic in the interview process.

[0007] Furthermore, the core competency requirements parameters for the interviewed positions are obtained, and a job competency assessment dimension library is constructed, including: Interview evaluation data of successful candidates is extracted from the historical recruitment evaluation data of the human resources platform. The interview evaluation data includes professional ability data, comprehensive quality data, and job suitability data. Professional ability, comprehensive quality, and job suitability are set as primary ability evaluation dimensions. Each primary ability evaluation dimension includes several secondary ability evaluation sub-items. A corresponding base score is preset for each secondary ability evaluation sub-item. The primary ability evaluation dimensions, secondary ability evaluation sub-items, and the base scores corresponding to the secondary ability sub-items together constitute the job ability evaluation dimension library. The aforementioned base score needs to be anchored to the "performance standards of historically recruited candidates in the corresponding secondary sub-items." That is, by analyzing the evaluation scores of historically recruited candidates in each secondary competency assessment sub-item (such as the average score and minimum passing score of historically recruited Python development engineers in the "Python Syntax Application" sub-item), the score range or benchmark value is preset as the "base score" for the corresponding secondary competency assessment sub-item of the current position, ensuring that the base score meets the "actual recruitment standards of the position."

[0008] Furthermore, each interview video data is divided into several video segments, and attribute tags are added to each video segment, including: The video segmentation rule is based on a single round of question and answer, meaning that the time from when the interview robot asks a question to when the interviewee finishes answering the question is divided into one video segment. The attribute tags include dimension association tags and content description tags; each video segment includes several dimension association tags and several content description tags; the dimension association tags are used to associate the primary ability assessment dimensions and secondary ability assessment sub-items in the video segment, and the content description tags are used to record the text data translated from the video segment.

[0009] Furthermore, based on the job competency assessment dimension library, several video segments after the interview video data were segmented were filtered to obtain the target video segments, including: The matching degree of the attribute tags of each video segment is calculated with the assessment dimensions in the job competency assessment dimension library. The matching degree calculation includes dimension matching degree calculation and content matching degree calculation. The dimension matching degree calculation involves comparing the secondary competency assessment sub-items included in each primary competency assessment dimension of each video segment with the secondary competency assessment sub-items in the job competency assessment dimension library. The base score corresponding to each secondary competency assessment sub-item is extracted from the job competency assessment dimension library. The base scores corresponding to each secondary competency assessment sub-item corresponding to the primary competency assessment dimension are then summed to obtain the primary competency assessment dimension score. Finally, the dimensions matching degree of the video segment is obtained by summing the scores of each primary competency assessment dimension. The content matching degree calculation uses a cosine similarity algorithm to convert the text data recorded in the video segment content description tags into vector form, and extracts the corresponding question text data through the video segment's association index. The question text data is then converted into vector form, and the cosine similarity between the text data in the video segment content description tag field and the question text data is calculated. The calculation formula is as follows: ; Where A represents the vector corresponding to the text data recorded in the video clip content description tag, and B represents the vector corresponding to the question text data. This represents the magnitude of the vector corresponding to the text data recorded in the video clip content description tag. This represents the magnitude of the vector corresponding to the query text data. Cosine similarity represents the content matching degree of video segments; Weights are preset for dimensional matching degree and content matching degree respectively. The overall matching degree of the video segment is obtained by weighted summation of the dimensional matching degree and content matching degree of the video segment. A matching degree threshold is preset based on the historical recruitment evaluation data of the human resources platform. If the overall matching degree of the video segment is greater than or equal to the preset matching degree threshold, the video segment is marked as the target video segment.

[0010] Furthermore, a preset multi-dimensional scoring model is retrieved to score the interviewee's performance in the target video segments, resulting in segment scores for each target video segment. Segment scores for segments of the same category are then sorted to generate a target segment scoring table, including: The multi-dimensional evaluation model includes several primary behavioral evaluation dimensions, each of which includes several secondary behavioral evaluation sub-items. The weights and basic scores of the secondary behavioral evaluation sub-items included in each primary behavioral evaluation dimension are determined by training a gradient boosting tree algorithm using historical recruitment data from the human resources platform. Feature extraction is performed on each target video segment to obtain feature data associated with the secondary behavioral sub-items. The feature extraction includes language behavior feature extraction, non-language behavior feature extraction, and logical thinking and adaptive behavior feature extraction. Language behavior feature extraction involves translating the target video segment into text data and using natural language processing tools to extract speech rate fluctuation values. Non-language behavior feature extraction involves analyzing frame images of the target video segment and using computer vision algorithms to extract facial change amplitude, limb movement frequency, and eye focus duration percentage. Logical thinking and adaptive behavior feature extraction involves analyzing the syntactic structure of the text data translated from the target video segment to obtain subject-verb-object logical completeness and cosine similarity between arguments and viewpoints. The extracted features are matched with the base scores of each secondary behavior evaluation item to obtain the base scores of each secondary behavior evaluation item included in the target video segment. The base scores of all secondary behavior evaluation items under the same primary behavior evaluation dimension are weighted and summed according to the weights corresponding to each secondary behavior evaluation item to obtain the weighted score of the primary behavior evaluation dimension. The weighted scores of each primary behavior evaluation dimension included in the target video segment are accumulated to obtain the total score of the target video segment. Based on the dimensional association tags of the target video segments, the target video segments are categorized into the first-level competency assessment dimension category of the corresponding job competency assessment dimension; for the same category of target video segments, they are sorted from high to low; if there are segments with the same total score, they are sorted again from high to low according to the weighted scores of logical thinking and adaptability; if the weighted scores in the second sort are still the same, they are sorted from shortest to longest according to the duration of the target video segments; Generate a target segment scoring table; the target segment scoring table is generated in a structured format and includes a basic index area, a behavior scoring area, and a related information area; The basic index area includes: interviewee ID, target segment number, job competency assessment dimension, and segment duration; The behavioral scoring areas include: the weighted scores of each primary behavioral evaluation dimension, the basic scores of each secondary behavioral evaluation sub-item, the total score of each target video segment, and the ranking of each target video segment. The associated information area includes: dimension association tags for each target segment, content description tags, and question text data corresponding to each target video segment.

[0011] Furthermore, the scores of several target video segments for the same interviewee across various job competency dimensions are extracted, the interviewee's overall interview score is calculated, and an overall interview score table is generated. The attribute tags associated with the corresponding target video segments in the overall interview score table include: Using the interviewee ID as a unique identifier, the dimensional matching degree of all target video segments of the interviewee is retrieved, and the dimensional matching degree of all the dimensional matching degrees is accumulated to obtain the interviewee's comprehensive ability score; the total segment score of the interviewee's target video segments under all first-level ability assessment dimensions is retrieved from the target segment score table, and the total segment scores are accumulated to obtain the interviewee's comprehensive behavior score. Obtain preset ability weights and behavior weights, and sum the comprehensive ability score and comprehensive behavior score of the interviewee to obtain the comprehensive interview score of the interviewee; sort all the comprehensive interview scores of the interviewees from high to low to generate an interview comprehensive score table.

[0012] Furthermore, the interviewer inputs screening criteria into the human resources platform, which are the scoring ranges for the primary competency assessment dimensions. The human resources platform then locks down the interviewees who meet the screening criteria in the comprehensive interview scoring table based on the primary competency assessment dimensions input by the interviewer, obtains the interviewee's interview ID, and retrieves the corresponding target video clip based on the interviewee's interview ID and the dimension-related tags associated with the primary competency assessment dimensions.

[0013] Furthermore, to better implement the above methods, a human resources platform management system based on multi-source data analysis is also provided. This system includes: a data extraction and association module, a job competency assessment dimension construction module, a video segmentation and tag addition module, a target video segment screening module, a segment rating table generation module, a comprehensive rating table generation module, and a video retrieval module. The data extraction and association module extracts interview video data and question text data, and establishes an association index between the two; The job competency assessment dimension construction module obtains core competency parameters for the job and constructs a job competency assessment dimension library that includes primary competency assessment dimensions, secondary competency assessment sub-items, and corresponding basic scores for secondary competency assessment sub-items. The video segmentation and tagging module segments videos into segments based on a round of question and answer, and adds dimension-related tags and content description tags to each video segment. The target video clip filtering module calculates the matching degree of video clips based on the job competency assessment dimension library and filters out target video clips. The segment rating table generation module uses a multi-dimensional model to score and rank the target segments and generate a segment rating table for the target segments; The comprehensive scoring table generation module calculates and sorts the interviewees' comprehensive interview scores, and generates a comprehensive interview scoring table with associated target segment tags; The video retrieval module receives the filtering criteria input by the interviewer, identifies interviewees who meet the filtering criteria, and retrieves the target video clips based on attribute tags.

[0014] Furthermore, the data extraction and association module includes: a data extraction unit and an association index establishment unit; The data extraction unit extracts interview video data and question text data of the interaction between the interviewee and the interview robot stored in the human resources platform; The association index building unit establishes time association indexes and text association indexes between video data and question text data according to timestamp alignment rules and content matching rules.

[0015] Furthermore, the target video segment filtering module includes: a matching degree calculation unit and a threshold judgment and labeling unit; The matching degree calculation unit calculates the dimensional matching degree and content matching degree of the video segment respectively, and then calculates the weighted sum of the dimensional matching degree and content matching degree to obtain the comprehensive matching degree of the video segment. The threshold judgment and marking unit compares the overall matching degree of video segments with a preset matching degree threshold, and marks video segments with an overall matching degree greater than or equal to the matching degree threshold as target video segments.

[0016] Furthermore, the segment scoring table generation module includes: a feature extraction unit, a segment scoring calculation unit, and a scoring table generation unit; The feature extraction unit extracts the language characteristics, non-language features, and logical thinking and adaptability features of the target segment. The segment scoring calculation unit calculates the weighted sum of the base scores of the secondary behavior evaluation sub-items in the target video segment to obtain the weighted score of the primary behavior evaluation dimension, and accumulates the weighted scores of the primary behavior evaluation dimension of the target video segment to obtain the total segment score of the target video segment. The scoring sheet generation unit sorts the total scores of the segments from highest to lowest and generates a scoring sheet for the target segment.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. It solves the pain point of traditional asynchronous video interviews where interviewers need to watch full long videos and have a heavy workload in screening. By automatically segmenting video clips, screening target clips, and intelligent scoring, it greatly reduces the time investment in manual viewing and evaluation, and significantly improves recruitment screening efficiency.

[0018] 2. Construct a job competency assessment dimension library based on historical recruitment data, and combine it with a scoring model extracted from multi-dimensional features (verbal, non-verbal, and logical thinking) to ensure that the assessment criteria are highly matched with the core requirements of the job, avoid subjective assessment bias, and improve the accuracy of recruitment assessment.

[0019] 3. The structured segment scoring sheet and comprehensive interview scoring sheet, along with associated attribute tags, allow interviewers to quickly locate interviewees who meet the screening criteria and directly retrieve target video segments for the corresponding ability dimensions, simplifying the recruitment decision-making process. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the method of the human resources platform management system and method based on multi-source data analysis of the present invention. Figure 2 This is a schematic diagram of the system structure of the human resources platform management system and method based on multi-source data analysis of the present invention. Detailed Implementation

[0021] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example: Figure 1 As shown, the present invention provides a technical solution, a human resources platform management method based on multi-source data analysis, the method comprising: Extract all interview video data and question text data from the interactions between interviewees and interview robots stored on the human resources platform; establish a correlation index between the interview video data and the question text data; obtain the core competency requirement parameters for the interview positions and construct a job competency assessment dimension library; divide each interview video data into several video segments and add attribute tags to each video segment; Based on the job competency assessment dimension library, several video segments after the interview video data were segmented were filtered to obtain the target video segment; The preset multi-dimensional scoring model is retrieved to score the behavior of the interviewee in the target video segment, and the segment score of each target video segment is obtained. The segment scores of the same type of target segments are sorted to generate a target segment score table. Extract several target video segment scores for the same interviewee under various job competency dimensions, calculate the interviewee's overall interview score, and generate an overall interview score table; associate the corresponding attribute tags of the target video segments with the overall interview score table; The interviewer inputs screening criteria into the human resources platform, which then locks in the interviewees who meet the screening criteria in the comprehensive interview scoring table, and retrieves the corresponding target video clips for review based on the associated attribute tags. Establishing a correlation index between interview video data and question text data includes: An association index is established between the interview video data and the question text data through preset association rules; the association rules include timestamp alignment rules and content matching rules; the timestamp alignment rules are established by extracting the interaction time interval of the interview video data and matching it with the timestamp of the question text data; the content matching rules are established by translating the audio content of the interview video data into text data and calculating the content similarity between the text data and the question text data. This includes obtaining the core competency requirements parameters for the interview position and constructing a job competency assessment dimension library, which includes: Interview evaluation data of successful candidates is extracted from the historical recruitment evaluation data of the human resources platform. The interview evaluation data includes professional ability data, comprehensive quality data, and job suitability data. Professional ability, comprehensive quality, and job suitability are set as primary ability evaluation dimensions. Each primary ability evaluation dimension includes several secondary ability evaluation sub-items. A corresponding base score is preset for each secondary ability evaluation sub-item. The primary ability evaluation dimensions, secondary ability evaluation sub-items, and the base scores corresponding to the secondary ability sub-items together constitute the job ability evaluation dimension library. Each interview video data segment is divided into several video clips, and attribute tags are added to each video clip, including: The video segmentation rule is based on a single round of question and answer, meaning that the time from when the interview robot asks a question to when the interviewee finishes answering the question is divided into one video segment. The attribute tags include dimension association tags and content description tags; each video segment includes several dimension association tags and several content description tags; the dimension association tags are used to associate the primary ability assessment dimensions and secondary ability assessment sub-items in the video segment, and the content description tags are used to record the text data translated from the video segment; Specifically, based on the job competency assessment dimension library, several video segments were selected from the segmented interview video data to obtain the target video segments, including: The matching degree of the attribute tags of each video segment is calculated with the assessment dimensions in the job competency assessment dimension library. The matching degree calculation includes dimension matching degree calculation and content matching degree calculation. The dimension matching degree calculation involves comparing the secondary competency assessment sub-items included in each primary competency assessment dimension of each video segment with the secondary competency assessment sub-items in the job competency assessment dimension library. The base score corresponding to each secondary competency assessment sub-item is extracted from the job competency assessment dimension library. The base scores corresponding to each secondary competency assessment sub-item corresponding to the primary competency assessment dimension are then summed to obtain the primary competency assessment dimension score. Finally, the dimensions matching degree of the video segment is obtained by summing the scores of each primary competency assessment dimension. The content matching degree calculation uses a cosine similarity algorithm to convert the text data recorded in the video segment content description tags into vector form, and extracts the corresponding question text data through the video segment's association index. The question text data is then converted into vector form, and the cosine similarity between the text data in the video segment content description tag field and the question text data is calculated. The calculation formula is as follows: ; Where A represents the vector corresponding to the text data recorded in the video clip content description tag, and B represents the vector corresponding to the question text data. This represents the magnitude of the vector corresponding to the text data recorded in the video clip content description tag. This represents the magnitude of the vector corresponding to the query text data. Cosine similarity represents the content matching degree of video segments; Weights are preset for dimensional matching degree and content matching degree respectively. The overall matching degree of the video segment is obtained by weighted summing of the dimensional matching degree and content matching degree of the video segment. A matching degree threshold is preset based on the historical recruitment evaluation data of the human resources platform. If the overall matching degree of the video segment is greater than or equal to the preset matching degree threshold, the video segment is marked as the target video segment. The process includes retrieving a pre-defined multi-dimensional scoring model to score the interviewee's performance in the target video segments, obtaining segment scores for each target video segment, sorting the segment scores of target segments of the same category, and generating a target segment score table, including: The multi-dimensional evaluation model includes several primary behavioral evaluation dimensions, each of which includes several secondary behavioral evaluation sub-items. The weights and basic scores of the secondary behavioral evaluation sub-items included in each primary behavioral evaluation dimension are determined by training a gradient boosting tree algorithm using historical recruitment data from the human resources platform. Feature extraction is performed on each target video segment to obtain feature data associated with the secondary behavioral sub-items. The feature extraction includes language behavior feature extraction, non-language behavior feature extraction, and logical thinking and adaptive behavior feature extraction. Language behavior feature extraction involves translating the target video segment into text data and using natural language processing tools to extract speech rate fluctuation values. Non-language behavior feature extraction involves analyzing frame images of the target video segment and using computer vision algorithms to extract facial change amplitude, limb movement frequency, and eye focus duration percentage. Logical thinking and adaptive behavior feature extraction involves analyzing the syntactic structure of the text data translated from the target video segment to obtain subject-verb-object logical completeness and cosine similarity between arguments and viewpoints. The extracted features are matched with the base scores of each secondary behavior evaluation item to obtain the base scores of each secondary behavior evaluation item included in the target video segment. The base scores of all secondary behavior evaluation items under the same primary behavior evaluation dimension are weighted and summed according to the weights corresponding to each secondary behavior evaluation item to obtain the weighted score of the primary behavior evaluation dimension. The weighted scores of each primary behavior evaluation dimension included in the target video segment are accumulated to obtain the total score of the target video segment. Based on the dimensional association tags of the target video segments, the target video segments are categorized into the first-level competency assessment dimension category of the corresponding job competency assessment dimension; for the same category of target video segments, they are sorted from high to low; if there are segments with the same total score, they are sorted again from high to low according to the weighted scores of logical thinking and adaptability; if the weighted scores in the second sort are still the same, they are sorted from shortest to longest according to the duration of the target video segments; Generate a target segment scoring table; the target segment scoring table is generated in a structured format and includes a basic index area, a behavior scoring area, and a related information area; The basic index area includes: interviewee ID, target segment number, job competency assessment dimension, and segment duration; The behavioral scoring areas include: the weighted scores of each primary behavioral evaluation dimension, the basic scores of each secondary behavioral evaluation sub-item, the total score of each target video segment, and the ranking of each target video segment. The associated information area includes: dimension association tags for each target segment, content description tags, and question text data corresponding to each target video segment; Specifically, the process involves extracting scores from several target video segments for the same interviewee across various job competency dimensions, calculating the interviewee's overall interview score, and generating an overall interview score table. The attribute tags associated with the corresponding target video segments in the overall interview score table include: Using the interviewee ID as a unique identifier, the dimensional matching degree of all target video segments of the interviewee is retrieved, and the dimensional matching degree of all the dimensional matching degrees is accumulated to obtain the interviewee's comprehensive ability score; the total segment score of the interviewee's target video segments under all first-level ability assessment dimensions is retrieved from the target segment score table, and the total segment scores are accumulated to obtain the interviewee's comprehensive behavior score. Obtain preset ability weights and behavior weights, and sum the comprehensive ability score and comprehensive behavior score of the interviewee to obtain the comprehensive interview score of the interviewee; sort all the comprehensive interview scores of the interviewees from high to low to generate an interview comprehensive score table. In this process, the interviewer inputs screening criteria into the human resources platform, which are the scoring ranges of the first-level competency assessment dimensions. The human resources platform locks the interviewees who meet the screening criteria in the comprehensive interview score table according to the first-level competency assessment dimensions input by the interviewer, obtains the interviewee's interview ID, and retrieves the corresponding target video clip based on the interviewee's interview ID and the dimension-related tags associated with the first-level competency assessment dimensions. In order to better implement the above methods, a human resources platform management system based on multi-source data analysis is also provided. The system includes: data extraction and association module, job competency assessment dimension construction module, video segmentation and tag addition module, target video segment screening module, segment rating table generation module, comprehensive rating table generation module, and video retrieval module. The data extraction and association module extracts interview video data and question text data, and establishes an association index between the two; The job competency assessment dimension construction module obtains core competency parameters for the job and constructs a job competency assessment dimension library that includes primary competency assessment dimensions, secondary competency assessment sub-items, and corresponding basic scores for secondary competency assessment sub-items. The video segmentation and tagging module segments videos into segments based on a round of question and answer, and adds dimension-related tags and content description tags to each video segment. The target video clip filtering module calculates the matching degree of video clips based on the job competency assessment dimension library and filters out target video clips. The segment rating table generation module uses a multi-dimensional model to score and rank the target segments and generate a segment rating table for the target segments; The comprehensive scoring table generation module calculates and sorts the interviewees' comprehensive interview scores, and generates a comprehensive interview scoring table with associated target segment tags; The video retrieval module receives the filtering criteria input by the interviewer, identifies interviewees who meet the filtering criteria, and retrieves the target video clips based on attribute tags. The data extraction and association module includes: a data extraction unit and an association index establishment unit; The data extraction unit extracts interview video data and question text data of the interaction between the interviewee and the interview robot stored in the human resources platform; The association index building unit establishes time association indexes and text association indexes between video data and question text data according to timestamp alignment rules and content matching rules, respectively. The target video segment filtering module includes: a matching degree calculation unit and a threshold judgment and marking unit; The matching degree calculation unit calculates the dimensional matching degree and content matching degree of the video segment respectively, and then calculates the weighted sum of the dimensional matching degree and content matching degree to obtain the comprehensive matching degree of the video segment. The threshold judgment and marking unit compares the overall matching degree of video segments with a preset matching degree threshold, and marks video segments with an overall matching degree greater than or equal to the matching degree threshold as target video segments; The segment scoring table generation module includes: a feature extraction unit, a segment scoring calculation unit, and a scoring table generation unit; The feature extraction unit extracts the language characteristics, non-language features, and logical thinking and adaptability features of the target segment. The segment scoring calculation unit calculates the weighted sum of the base scores of the secondary behavior evaluation sub-items in the target video segment to obtain the weighted score of the primary behavior evaluation dimension, and accumulates the weighted scores of the primary behavior evaluation dimension of the target video segment to obtain the total segment score of the target video segment. The scoring sheet generation unit sorts the total scores of the segments from highest to lowest and generates a scoring sheet for the target segment. In an embodiment of the present invention, taking the recruitment of a Python development engineer by a technology company as an example, firstly, the data extraction and association module extracts interview videos of 50 interviewees and textual questions such as "Python project development" and "exception handling solutions," and establishes a data association index through timestamp matching and audio-to-text content similarity calculation; the job competency assessment dimension construction module constructs the job competency assessment dimension from historical recruitment data, setting professional competency (secondary competency assessment sub-items: Python syntax application, framework usage) and job suitability (secondary competency assessment sub-item: project experience matching degree) as primary competency assessment dimensions and pre-setting base scores. The system employs a capability assessment dimension library; a video segmentation module segments videos by "question-answer" format and adds tags such as "professional ability - Python syntax application"; a target segment selection module calculates segment matching degree and selects target video segments with a comprehensive matching degree ≥ 75; a multi-dimensional model extracts and scores feature dimensions such as the interviewee's code logic completeness and body movement frequency from the segments, generating a segment scoring table; a comprehensive scoring table generation module calculates and sorts the overall interview score based on a 70% capability weight and a 30% behavior weight; when the interviewer inputs "professional ability score 80-100", the system locks 8 interviewees and retrieves their "Python syntax application" related video segments through associated tags.

[0023] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A human resource platform management method based on multi-source data analysis, characterized in that: The method comprises: extracting all interviewee interview video data and question text data in the process of interacting with the interview robot stored in the human resource platform; establishing an association index between the interview video data and the question text data; obtaining core ability requirement parameters of the interview post, and constructing a post ability evaluation dimension library; dividing each piece of interview video data into a plurality of video segments, and adding attribute labels to each video segment; According to the post ability evaluation dimension library, the plurality of video segments after the interview video data is divided are screened to obtain target video segments; retrieve the preset multi-dimensional scoring model, score the behavior performance of the interviewee in the target video segment, obtain the segment score of each target video segment, sort the segment scores of the same type of target segment, and generate a target segment score table; extracting a plurality of target segment scores of the same interviewee under each post ability dimension, calculating the interview comprehensive score of the interviewee, and generating an interview comprehensive score table; in the interview comprehensive score table, the attribute labels of the corresponding target video segments are associated; The interviewer inputs a screening condition to the human resource platform, the human resource platform locks the interviewee in the interview comprehensive score table that meets the screening condition according to the screening condition, and retrieves the corresponding target video segment according to the associated attribute label for review.

2. The human resource platform management method based on multi-source data analysis according to claim 1, characterized in that: the association index between the interview video data and the question text data comprises: establishing the association index between the interview video data and the question text data through a preset association rule; the association rule comprises a timestamp alignment rule and a content matching rule; the timestamp alignment rule matches the interaction time interval of the interview video data with the timestamp of the question text data to establish a time association index; the content matching rule converts the audio content of the interview video data into text data, calculates the content similarity of the text data and the question text data, and establishes a text association index; the core ability requirement parameters of the interview post are obtained, and the post ability evaluation dimension library is constructed, which comprises: extracting the interview evaluation data of the hired person from the historical recruitment evaluation data of the human resource platform, the interview evaluation data comprising professional ability data, comprehensive quality data, and post adaptation degree data; setting professional ability, comprehensive quality, and post adaptation degree as a first-level ability evaluation dimension, each first-level ability evaluation dimension comprising a plurality of second-level ability evaluation sub-items; presetting a corresponding basic score for each second-level ability evaluation sub-item; the first-level ability evaluation dimension, the second-level ability evaluation sub-item, and the corresponding basic score of the second-level ability evaluation sub-item jointly constitute the post ability evaluation dimension library; the interview video data is divided into a plurality of video segments, and an attribute label is added to each video segment, which comprises: the video segment division rule is to divide one round of questioning as a unit, that is, to divide a video segment from the time point when the interview robot asks a question to the time point when the interviewee finishes answering the question; The attribute label includes a dimension association label and a content description label; each video segment includes a plurality of dimension association labels and a plurality of content description labels; the dimension association label is used for associating a primary capability evaluation dimension and a secondary capability evaluation sub-item in the video segment, and the content description label is used for recording text data translated from the video segment.

3. The human resource platform management method based on multi-source data analysis according to claim 1, characterized in that: The target video segment is obtained by screening the plurality of video segments after the interview video data is cut according to the post capability evaluation dimension library, and the screening comprises: The attribute label of each video segment is matched with the evaluation dimension in the post capability evaluation dimension library to perform matching degree calculation, and the matching degree calculation comprises dimension matching degree calculation and content matching degree calculation; The dimension matching degree calculation is to compare the secondary capability evaluation sub-items included in each primary capability evaluation dimension in each video segment with the secondary capability evaluation sub-items in the post capability evaluation dimension library, extract the basic score corresponding to the secondary capability evaluation sub-items in the post capability evaluation dimension library, accumulate the basic scores corresponding to each secondary capability evaluation sub-item corresponding to the primary capability evaluation dimension to obtain the primary capability evaluation dimension score, and accumulate the primary capability evaluation dimension scores to obtain the dimension matching degree of the video segment; The content matching degree calculation converts the text data recorded in the content description label of the video segment into a vector form through a cosine similarity algorithm, extracts the corresponding question text data through the association index of the video segment, converts the question text data into a vector form, calculates the cosine similarity between the text data in the content description label field of the video segment and the question text data, and the calculation formula is: ； wherein A represents a vector corresponding to the text data recorded in the content description label of the video segment, B represents a vector corresponding to the text data of the question, represents a length of the vector corresponding to the text data recorded in the content description label of the video segment, represents a length of the vector corresponding to the text data of the question, represents a cosine similarity, i.e., a content matching degree of the video segment; The weights corresponding to the dimension matching degree and the content matching degree are respectively preset, the comprehensive matching degree of the video segment is obtained by weighted summation of the dimension matching degree and the content matching degree of the video segment, the matching degree threshold is preset according to the historical recruitment evaluation data of the human resource platform, and if the comprehensive matching degree of the video segment is greater than or equal to the preset matching degree threshold, the video segment is marked as a target video segment.

4. The human resource platform management method based on multi-source data analysis according to claim 1, characterized in that: The performance of the interviewee in the target video segment is scored by calling the preset multi-dimensional scoring model to obtain the segment score of each target video segment, the segment scores of the target segments of the same type are sorted to generate a target segment score table, and the target segment score table comprises: The multi-dimensional evaluation model comprises a plurality of primary behavior evaluation dimensions, each primary behavior evaluation dimension comprises a plurality of secondary behavior evaluation sub-items, and the weights and basic scores of the secondary behavior evaluation sub-items included in each primary behavior evaluation dimension are determined by gradient boosting tree algorithm training based on historical recruitment data of the human resource platform. The feature extraction is performed on each target video segment to obtain feature data associated with the secondary behavior sub-items; wherein the feature extraction includes language behavior feature extraction, non-language behavior feature extraction, logical thinking and adaptive behavior feature extraction; the language behavior feature extraction is performed by translating the target video segment into text data, and a natural language processing tool is used to extract a speech speed fluctuation value; the non-language behavior feature extraction is performed by analyzing frame images of the target video segment, and a computer vision algorithm is used to extract a facial change amplitude, a body movement frequency, and an eye focus time length proportion; the logical thinking and adaptive behavior feature extraction is performed according to a syntax structure analysis of the text data translated from the target video segment to obtain a subject-predicate-object logical integrity and a cosine similarity of arguments and viewpoints; The extracted features are matched with the basic scores of the secondary behavior evaluation sub-items to obtain the basic scores of the secondary behavior evaluation sub-items included in the target video segment, and the basic scores of all the secondary behavior evaluation sub-items under the same primary behavior evaluation dimension are weighted and summed according to the weights corresponding to the secondary behavior evaluation sub-items to obtain the weighted score of the primary behavior evaluation dimension; the weighted scores of the primary behavior evaluation dimensions included in the target video segment are accumulated to obtain the segment total score of the target video segment; The target video segment is classified into a primary ability evaluation dimension category of a corresponding post ability evaluation dimension according to the dimension association label of the target video segment; the same type of target video segment is sorted from high to low; if the segment total scores are the same, the weighted scores of the logical thinking and adaptive behavior are sorted from high to low for secondary sorting; if the weighted scores are still the same in the secondary sorting, the target video segments are sorted from short to long according to the time length; A target segment score table is generated; the target segment score table is generated in a structured format and includes a basic index area, a behavior score area, and an association information area; The basic index area includes an interviewee ID, a target segment number, a post ability evaluation dimension, and a segment time length; The behavior score area includes the weighted scores of the primary behavior evaluation dimensions, the basic scores of the secondary behavior evaluation sub-items, the segment total scores of the target video segments, and the ranking positions of the target video segments; The association information area includes the dimension association labels of the target segments, content description labels, and the question text data corresponding to the target video segments.

5. The human resource platform management method based on multi-source data analysis according to claim 1, characterized in that: The scores of the target segments of the same interviewee under each post ability dimension are extracted, the interview comprehensive score of the interviewee is calculated, and an interview comprehensive score table is generated; The attribute labels of the corresponding target video segments are associated in the interview comprehensive score table, and the attribute labels include: The interviewee ID is taken as a unique identifier to call the dimension matching degrees of all target video clips of the interviewee, and the dimension matching degrees are accumulated to obtain a comprehensive ability score of the interviewee; the interviewee's total clip score of the target video clips under all first-level ability evaluation dimensions is called from the target clip score table, and the total clip scores are accumulated to obtain a comprehensive behavior score of the interviewee; The preset ability weight and behavior weight are obtained, and the comprehensive ability score and the comprehensive behavior score of the interviewee are weighted and summed to obtain an interview comprehensive score of the interviewee; the interview comprehensive scores of all interviewees are sorted from high to low to generate an interview comprehensive score table.

6. The human resource platform management method based on multi-source data analysis according to claim 1, characterized in that: The interviewer inputs a screening condition to the human resource platform, the screening condition is a first-level ability evaluation dimension score interval; the human resource platform locks the interviewees meeting the screening condition in the interview comprehensive score table according to the first-level ability evaluation dimension input by the interviewer, obtains the interview ID of the interviewee, and calls the corresponding target video clips according to the interview ID of the interviewee and the dimension association label associated with the first-level ability evaluation dimension.

7. A human resource platform management system based on multi-source data analysis, for performing the human resource platform management method based on multi-source data analysis according to any one of claims 1-6, characterized in that: The system comprises a data extraction and association module, a post ability evaluation dimension construction module, a video cutting and label adding module, a target video clip screening module, a clip score table generation module, a comprehensive score table generation module, and a video calling module. The data extraction and association module extracts interview video data and question text data and establishes an association index therebetween; The post ability evaluation dimension construction module obtains post core ability parameters and constructs a post ability evaluation dimension library comprising a first-level ability evaluation dimension, a second-level ability evaluation sub-item, and a second-level ability evaluation sub-item corresponding basic score; The video cutting and label adding module cuts a video into clips according to a round of questioning, adds a dimension association label and a content description label to each video clip; The target video clip screening module calculates a video clip matching degree according to the post ability evaluation dimension library and screens out target video clips; The clip score table generation module scores, sorts, and generates a clip score table of the target clips using a multi-dimensional model; The comprehensive score table generation module calculates and sorts the interview comprehensive scores of the interviewees and generates an interview comprehensive score table associated with the target clip labels; The video calling module receives a screening condition input by an interviewer, locks the interviewees meeting the screening condition, and calls target video clips according to attribute labels.

8. The human resource platform management system based on multi-source data analysis according to claim 7, characterized in that: The data extraction and association module comprises a data extraction unit and an association index establishment unit. The data extraction unit extracts interview video data and question text data of an interviewee interacting with an interview robot stored in a human resource platform; The association index establishment unit respectively establishes a time association index and a text association index between the video data and the question text data according to a timestamp alignment rule and a content matching rule.

9. The human resource platform management system based on multi-source data analysis of claim 7, wherein: The target video clip screening module comprises a matching degree calculation unit and a threshold judgment and marking unit. The matching degree calculation unit respectively calculates video segment dimension matching degree and content matching degree, and obtains comprehensive matching degree of the video segment by weighted summation of the dimension matching degree and the content matching degree; The threshold judgment and marking unit compares the comprehensive matching degree of the video segment with a preset matching degree threshold, and marks the video segment with a comprehensive matching degree greater than or equal to the matching degree threshold as a target video segment.

10. The human resource platform management system based on multi-source data analysis of claim 7, wherein: The segment score table generation module comprises a feature extraction unit, a segment score calculation unit and a score table generation unit. The feature extraction unit extracts language features, non-language features and logical thinking and adaptability features of the target segment. The segment score calculation unit obtains a weighted score of a primary behavior evaluation dimension by weighted summation of basic scores of secondary behavior evaluation sub-items in the target video segment, accumulates the weighted score of the primary behavior evaluation dimension of the target video segment, and obtains a segment total score of the target video segment. The score table generation unit sorts the segment total scores from high to low, and generates a target segment score table.

Images