Accurate matching method and system for research and development requirements and research and development personnel

Abstract

The invention discloses a research and development demand and research and development personnel accurate matching method and system, and relates to the technical field of task allocation, and the method comprises the steps: obtaining a research and development demand document meeting a description rule; the method comprises the following steps: automatically disassembling an obtained research and development demand document into a research and development task set with minimum granularity through three operations of natural language processing, document pre-training learning and deep learning; key features of each research and development task in the set are extracted, the key features are compared with items in a preset task classification and classification library through a matching algorithm, and the most matched task type and level are found out; and on the basis of the key features, task types and levels of the research and development tasks, the research and development tasks are compared with research and development personnel in a preset research and development personnel comprehensive rating library through a matching algorithm, the most suitable research and development personnel with the shortest prediction time are found out, and the corresponding research and development tasks are distributed. According to the invention, accurate matching between research and development requirements and research and development personnel can be realized, the research and development efficiency and quality are improved, and resource waste is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of task allocation, and in particular to a method and system for accurately matching R&D requirements with R&D personnel. Background Art

[0002] With the continuous development of technologies such as cloud computing, big data, and artificial intelligence, project operation management is gradually moving towards digitalization and intelligence. Digital technology can improve operational efficiency and quality while reducing labor costs, while intelligent technology can improve the accuracy of predictions and decisions.

[0003] In the current traditional R&D team management, the supervision of task distribution is completely based on the subjective judgment of the team leader. Who will be assigned to do the task and how long it will take to complete it are completely determined by the leader's personal experience. This leads to the following problems:

[0004] 1. Inaccurate subjective judgment;

[0005] 2. Insufficient digital control during mission development;

[0006] 3. Insufficient use of data for full-process task control. Summary of the invention

[0007] In response to the needs and shortcomings of current technological development, the present invention provides a method and system for accurately matching R&D needs with R&D personnel. By classifying and grading tasks and rating the capabilities of R&D personnel, the needs can be automatically decomposed into standard R&D tasks, the time consumption of needs can be intelligently predicted, and the best R&D personnel can be intelligently recommended, thereby achieving scientific and efficient R&D management.

[0008] In the first aspect, the present invention provides a method for accurately matching R&D requirements with R&D personnel. The technical solution adopted to solve the above technical problems is as follows:

[0009] A method for accurately matching R&D requirements with R&D personnel comprises the following steps:

[0010] S1. Obtain R&D requirement documents that comply with description rules;

[0011] S2: Through natural language processing, document pre-training learning, and deep learning, the acquired R&D requirement documents are automatically decomposed into a set of R&D tasks with the smallest granularity;

[0012] S3, extract the key features of each R&D task in the set, compare them with the entries in the preset task classification and grading library through a matching algorithm, and find the most matching task type and level;

[0013] S4. Extract the key features, specific task types and levels of each R&D task in the set, compare them with the R&D personnel in the preset R&D personnel comprehensive rating database through a matching algorithm, find the most suitable R&D personnel with the shortest predicted time, and provide them to the R&D manager for viewing through a visual interface;

[0014] S5. According to the output result of the matching algorithm in step S4, the R&D manager assigns the decomposed R&D tasks to the corresponding R&D personnel, and the R&D personnel corresponding to the R&D task set form a R&D team.

[0015] Optionally, the step S2 specifically includes the following process:

[0016] S2.1, Natural Language Processing: Perform text preprocessing on R&D requirement documents by segmenting, removing stop words and part-of-speech tagging; extract entities in the text using named entity recognition technology, and determine the association between entities through relationship extraction; perform dependency syntactic analysis on sentences to identify the dependency relationship between words in the sentence; identify the semantic role in each predicate-argument structure through semantic role labeling, and understand the specific event or action described in the sentence; extract keywords and generate a summary by analyzing and summarizing the text content;

[0017] S2.2. Use a BERT-based pre-trained model; fine-tune the pre-trained model according to the characteristics of the R&D requirements document to better adapt it to the language rules and terminology of the specified field; use the pre-trained model to generate context-related word embeddings based on the summary text to capture the deep semantic information in the summary text;

[0018] S2.3. Build a task-oriented deep learning model. The deep learning model first classifies the summary text into preset categories based on the output of the summary text and the pre-trained model, then labels the words in the summary text through the sequence labeling model, identifies the boundaries and types of tasks, and then calculates the semantic similarity between different parts of the summary text to assist in determining which content belongs to the same task or related tasks, ultimately achieving the goal of breaking down the R&D requirement document into a set of R&D tasks with the smallest granularity.

[0019] Optionally, the matching algorithm selected in step S3 is one of cosine similarity calculation, rule-based matching, decision tree, support vector machine or neural network algorithm;

[0020] The matching algorithm selected in step S4 is one of rule-based matching, multi-factor comprehensive evaluation, optimization algorithm or machine learning model.

[0021] Optionally, the comprehensive rating database of the R&D personnel involved includes the R&D personnel's skill information, work experience, historical performance, preferred fields and ability level, where: the R&D personnel's historical performance includes the on-time completion rate of tasks, code quality, speed of solving technical problems, teamwork and communication skills, innovation ability, learning ability, project management and leadership ability, customer feedback, task complexity, workload and work attitude;

[0022] Based on the historical performance of R&D personnel, a pre-trained quantitative model is used to generate a capability score for each R&D personnel. R&D personnel are divided into different capability levels according to the capability score based on the preset threshold range.

[0023] After extracting the key features, specific task types and levels of each R&D task in the collection, they are compared with the R&D personnel in the preset R&D personnel comprehensive rating library through a matching algorithm. First, the R&D personnel who match the R&D task are found, and then the time consumption of the R&D task is predicted based on the historical performance of the R&D personnel. Finally, the R&D personnel with the shortest time consumption are selected.

[0024] Optionally, the pre-training process of the quantization model involved is as follows:

[0025] Collect historical performance data of R&D personnel and convert the collected data into quantitative features as historical data sets;

[0026] Use the R&D personnel’s historical data set to train the selected quantitative model to ensure that the quantitative model can accurately map the R&D personnel’s characteristics to their ability scores;

[0027] Through the trained quantitative model, a comprehensive capability score is generated for each R&D personnel, which reflects the performance of the R&D personnel in various evaluation dimensions;

[0028] A pre-trained quantitative model is used to generate a capability score for each R&D personnel, and the R&D personnel are divided into different capability levels according to the capability score based on the preset threshold range.

[0029] In the second aspect, the present invention provides a precise matching system between R&D requirements and R&D personnel. The technical solutions adopted to solve the above technical problems are as follows:

[0030] A precise matching system between R&D requirements and R&D personnel, comprising:

[0031] Requirements acquisition module, used to obtain R&D requirements documents that meet the description rules;

[0032] The document parsing module is used to automatically decompose the acquired R&D requirement documents into a set of R&D tasks with the smallest granularity through three operations: natural language processing, document pre-training learning, and deep learning.

[0033] The task matching module is used to extract the key features of each R&D task in the set, compare them with the entries in the preset task classification and grading library through the matching algorithm, and find the most matching task type and level;

[0034] The personnel matching module is used to extract the key features, specific task types and levels of each R&D task in the set, and compare them with the R&D personnel in the preset R&D personnel comprehensive rating database through the matching algorithm to find the most suitable R&D personnel with the shortest predicted time.

[0035] The task management module is used to assign the R&D tasks decomposed by the document parsing module to the corresponding R&D personnel according to the output results of the personnel matching module. The R&D personnel corresponding to the R&D task set form a R&D team.

[0036] Optionally, the document parsing module involved automatically decomposes the acquired R&D requirement documents into a minimum-granularity R&D task set through three operations: natural language processing, document pre-training learning, and deep learning. The specific process includes:

[0037] (1) Natural language processing: Perform text preprocessing on R&D requirement documents by segmenting, removing stop words, and tagging parts of speech; extract entities in the text using named entity recognition technology, and determine the relationship between entities through relationship extraction; perform dependency syntactic analysis on sentences to identify the dependency relationship between words in the sentence; identify the semantic role in each predicate-argument structure through semantic role labeling to understand the specific event or action described in the sentence; extract keywords and generate a summary by analyzing and summarizing the text content;

[0038] (2) Document pre-training learning: Use a BERT-based pre-training model; fine-tune the pre-training model based on the characteristics of the R&D requirement document to make it better adapt to the language rules and terminology of the specified field; use the pre-training model to generate context-related word embeddings based on the summary text to capture the deep semantic information in the summary text;

[0039] (3) Deep learning: Build a task-oriented deep learning model. The deep learning model first classifies the summary text into preset categories based on the output of the summary text and the pre-trained model. Then, the sequence labeling model is used to label the words in the summary text, identify the boundaries and types of tasks, and then calculate the semantic similarity between different parts of the summary text to assist in determining which content belongs to the same task or related tasks. Ultimately, the goal of breaking down the R&D requirement document into a set of R&D tasks with the smallest granularity is achieved.

[0040] Optionally, the matching algorithm selected by the task matching module involved is one of cosine similarity calculation, rule-based matching, decision tree, support vector machine or neural network algorithm;

[0041] The matching algorithm selected by the personnel matching module involved is one of rule-based matching, multi-factor comprehensive evaluation, optimization algorithm or machine learning model.

[0042] Optionally, the comprehensive rating database of the R&D personnel involved includes the R&D personnel's skill information, work experience, historical performance, preferred fields and ability level, where: the R&D personnel's historical performance includes the on-time completion rate of tasks, code quality, speed of solving technical problems, teamwork and communication skills, innovation ability, learning ability, project management and leadership ability, customer feedback, task complexity, workload and work attitude;

[0043] Based on the historical performance of R&D personnel, a pre-trained quantitative model is used to generate a capability score for each R&D personnel. R&D personnel are divided into different capability levels according to the capability score based on the preset threshold range.

[0044] The personnel matching module extracts the key features, specific task types and levels of each R&D task in the collection, and compares them with the R&D personnel in the preset R&D personnel comprehensive rating library through a matching algorithm. It first finds the R&D personnel who match the R&D task, then predicts the time consumption of the R&D task based on the R&D personnel's historical performance, and finally selects the R&D personnel with the shortest time consumption.

[0045] Optionally, the pre-training process of the quantization model involved is as follows:

[0046] Collect historical performance data of R&D personnel and convert the collected data into quantitative features as historical data sets;

[0047] Use the R&D personnel’s historical data set to train the selected quantitative model to ensure that the quantitative model can accurately map the R&D personnel’s characteristics to their ability scores;

[0048] Through the trained quantitative model, a comprehensive capability score is generated for each R&D personnel, which reflects the performance of the R&D personnel in various evaluation dimensions;

[0049] A pre-trained quantitative model is used to generate a capability score for each R&D personnel, and the R&D personnel are divided into different capability levels according to the capability score based on the preset threshold range.

[0050] The invention provides a method and system for accurately matching R&D requirements with R&D personnel, which has the following beneficial effects compared with the prior art:

[0051] 1. The present invention achieves accurate matching of R&D requirements and R&D personnel by obtaining R&D requirement documents, breaking them down into R&D task sets of the smallest granularity, extracting key features and comparing them with a preset library, and finding the most suitable R&D personnel, thereby improving R&D efficiency and quality and reducing resource waste;

[0052] 2. In the process of matching R&D needs with R&D personnel, the present invention first classifies and grades R&D tasks, and makes digital portraits of R&D personnel, so as to realize scientific, efficient and intelligent distribution of R&D tasks, achieve the most reasonable use of R&D personnel capabilities, and maximize the effectiveness of the R&D team. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Attached Figure 1 is a flow chart of a method according to Embodiment 1 of the present invention;

[0054] Attached Figure 2 This is a module connection block diagram of the second embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the technical solution, the technical problem solved and the technical effect of the present invention more clearly understood, the technical solution of the present invention is clearly and completely described below in conjunction with specific embodiments.

[0056] Embodiment 1:

[0057] Combined with Figure 1 This embodiment proposes a method for accurately matching R&D requirements with R&D personnel, which includes the following steps:

[0058] S1. Obtain R&D requirement documents that comply with description rules.

[0059] Since the writing format of the R&D requirement document is relatively casual, the specific format and description rules of the R&D requirement document must be specified in advance to obtain an R&D requirement document that meets the standards, so as to quickly execute subsequent steps.

[0060] S2: Through natural language processing, document pre-training learning, and deep learning, the acquired R&D requirement documents are automatically decomposed into a set of R&D tasks with the smallest granularity. The specific process includes the following:

[0061] S2.1, Natural Language Processing: Perform text preprocessing on R&D requirement documents by segmenting, removing stop words and part-of-speech tagging; extract entities in the text using named entity recognition technology, and determine the association between entities through relationship extraction; perform dependency syntactic analysis on sentences to identify the dependency relationship between words in the sentence; identify the semantic role in each predicate-argument structure through semantic role labeling, and understand the specific event or action described in the sentence; extract keywords and generate a summary by analyzing and summarizing the text content;

[0062] S2.2. Use a BERT-based pre-trained model; fine-tune the pre-trained model according to the characteristics of the R&D requirements document to better adapt it to the language rules and terminology of the specified field; use the pre-trained model to generate context-related word embeddings based on the summary text to capture the deep semantic information in the summary text;

[0063] S2.3. Build a task-oriented deep learning model. The deep learning model first classifies the summary text into preset categories based on the output of the summary text and the pre-trained model, then labels the words in the summary text through the sequence labeling model, identifies the boundaries and types of tasks, and then calculates the semantic similarity between different parts of the summary text to assist in determining which content belongs to the same task or related tasks, ultimately achieving the goal of breaking down the R&D requirement document into a set of R&D tasks with the smallest granularity.

[0064] S3. Extract the key features of each R&D task in the collection, compare them with the entries in the preset task classification and grading library through a matching algorithm, and find the most matching task type and level.

[0065] The matching algorithm selected in this step is one of cosine similarity calculation, rule-based matching, decision tree, support vector machine or neural network algorithm.

[0066] The preset task classification and grading library includes the following features:

[0067] Task type: including different types of programming tasks, such as front-end development, back-end development, database management, etc.

[0068] Task level: Classified according to the complexity of the task and the skill level required, such as beginner, intermediate and advanced;

[0069] Technical requirements: These include specific technologies or tools needed to complete the task, such as a specific programming language, framework, or software;

[0070] Industry field: includes specific industries or fields related to the task, such as finance, medical care, education, etc.

[0071] Other features may include:

[0072] Experience requirements: including work experience or project experience required to complete the task;

[0073] Educational background: Include the minimum educational level or professional background required to complete the task;

[0074] Skills and abilities: including specific skills and abilities, such as problem solving, teamwork, etc.

[0075] Working hours and availability: including the estimated time required to complete the task and the availability of the R&D staff;

[0076] The above features are only possible features. The actual preset task classification and grading library may be adjusted according to specific R&D needs and organizational needs.

[0077] S4. Extract the key features, specific task types and levels of each R&D task in the set, compare them with the R&D personnel in the preset R&D personnel comprehensive rating library through a matching algorithm, find the most suitable R&D personnel with the shortest predicted time, and provide them to the R&D manager for viewing through a visual interface.

[0078] The matching algorithm selected in this step is one of rule-based matching, multi-factor comprehensive evaluation, optimization algorithm or machine learning model.

[0079] The preset comprehensive rating database for R&D personnel contains the R&D personnel's skill information, work experience, historical performance, preferred areas and ability level, among which: the R&D personnel's historical performance includes the punctuality of completing tasks, code quality, speed of solving technical problems, teamwork and communication skills, innovation ability, learning ability, project management and leadership ability, customer feedback, task complexity, workload and work attitude.

[0080] Based on the historical performance of R&D personnel, a pre-trained quantitative model is used to generate an ability score for each R&D personnel. Through the preset threshold range, R&D personnel are divided into different ability levels according to the ability score, such as level one, level two, level three, or junior, intermediate, and senior.

[0081] This step extracts the key features, specific task types and levels of each R&D task in the collection, and then compares them with the R&D personnel in the preset R&D personnel comprehensive rating library through a matching algorithm. First, the R&D personnel who match the R&D task are found, and then the time consumption of the R&D task is predicted based on the historical performance of the R&D personnel. Finally, the R&D personnel with the shortest time consumption are selected.

[0082] It should be added that the pre-training process of the quantization model is as follows:

[0083] Collect historical performance data of R&D personnel and convert the collected data into quantitative features as historical data sets;

[0084] The selected quantitative model is trained using the historical data set of the R&D personnel to ensure that the quantitative model can accurately map the characteristics of the R&D personnel to their ability scores; the quantitative model uses one of logistic regression, decision tree, random forest or neural network;

[0085] Through the trained quantitative model, a comprehensive capability score is generated for each R&D personnel, which reflects the performance of the R&D personnel in various evaluation dimensions;

[0086] A pre-trained quantitative model is used to generate a capability score for each R&D personnel, and the R&D personnel are divided into different capability levels according to the capability score based on the preset threshold range.

[0087] S5. According to the output result of the matching algorithm in step S4, the R&D manager assigns the decomposed R&D tasks to the corresponding R&D personnel, and the R&D personnel corresponding to the R&D task set form a R&D team.

[0088] Embodiment 2:

[0089] Combined with Figure 2 This embodiment proposes a precise matching system between R&D requirements and R&D personnel, which includes:

[0090] Requirements acquisition module, used to obtain R&D requirements documents that meet the description rules;

[0091] The document parsing module is used to automatically decompose the acquired R&D requirement documents into a set of R&D tasks with the smallest granularity through three operations: natural language processing, document pre-training learning, and deep learning.

[0092] The task matching module is used to extract the key features of each R&D task in the set, compare them with the entries in the preset task classification and grading library through the matching algorithm, and find the most matching task type and level;

[0093] The personnel matching module is used to extract the key features, specific task types and levels of each R&D task in the set, and compare them with the R&D personnel in the preset R&D personnel comprehensive rating database through the matching algorithm to find the most suitable R&D personnel with the shortest predicted time.

[0094] The task management module is used to assign the R&D tasks decomposed by the document parsing module to the corresponding R&D personnel according to the output results of the personnel matching module. The R&D personnel corresponding to the R&D task set form a R&D team.

[0095] In this embodiment, the document parsing module involved automatically decomposes the acquired R&D requirement documents into a minimum-granularity R&D task set through three operations: natural language processing, document pre-training learning, and deep learning. The specific process includes:

[0096] (1) Natural language processing: Perform text preprocessing on R&D requirement documents by segmenting, removing stop words, and tagging parts of speech; extract entities in the text using named entity recognition technology, and determine the relationship between entities through relationship extraction; perform dependency syntactic analysis on sentences to identify the dependency relationship between words in the sentence; identify the semantic role in each predicate-argument structure through semantic role labeling to understand the specific event or action described in the sentence; extract keywords and generate a summary by analyzing and summarizing the text content;

[0097] (2) Document pre-training learning: Use a BERT-based pre-training model; fine-tune the pre-training model based on the characteristics of the R&D requirement document to make it better adapt to the language rules and terminology of the specified field; use the pre-training model to generate context-related word embeddings based on the summary text to capture the deep semantic information in the summary text;

[0098] (3) Deep learning: Build a task-oriented deep learning model. The deep learning model first classifies the summary text into preset categories based on the output of the summary text and the pre-trained model. Then, the sequence labeling model is used to label the words in the summary text, identify the boundaries and types of tasks, and then calculate the semantic similarity between different parts of the summary text to assist in determining which content belongs to the same task or related tasks. Ultimately, the goal of breaking down the R&D requirement document into a set of R&D tasks with the smallest granularity is achieved.

[0099] In this embodiment, the matching algorithm selected by the task matching module is one of cosine similarity calculation, rule-based matching, decision tree, support vector machine or neural network algorithm; the matching algorithm selected by the personnel matching module is one of rule-based matching, multi-factor comprehensive evaluation, optimization algorithm or machine learning model.

[0100] In this embodiment, the R&D personnel comprehensive rating database includes the R&D personnel's skill information, work experience, historical performance, preferred fields and ability level, where: the R&D personnel's historical performance includes the punctuality of completing tasks, code quality, speed of solving technical problems, teamwork and communication skills, innovation ability, learning ability, project management and leadership ability, customer feedback, task complexity, workload and work attitude;

[0101] Based on the historical performance of R&D personnel, a pre-trained quantitative model is used to generate a capability score for each R&D personnel. R&D personnel are divided into different capability levels according to the capability score based on the preset threshold range.

[0102] The personnel matching module extracts the key features, specific task types and levels of each R&D task in the collection, and compares them with the R&D personnel in the preset R&D personnel comprehensive rating library through a matching algorithm. It first finds the R&D personnel who match the R&D task, then predicts the time consumption of the R&D task based on the R&D personnel's historical performance, and finally selects the R&D personnel with the shortest time consumption.

[0103] It should be added that the pre-training process of the quantization model is as follows:

[0104] Collect historical performance data of R&D personnel and convert the collected data into quantitative features as historical data sets;

[0105] The selected quantitative model is trained using the historical data set of the R&D personnel to ensure that the quantitative model can accurately map the characteristics of the R&D personnel to their ability scores; the quantitative model uses one of logistic regression, decision tree, random forest or neural network;

[0106] Through the trained quantitative model, a comprehensive capability score is generated for each R&D personnel, which reflects the performance of the R&D personnel in various evaluation dimensions;

[0107] A pre-trained quantitative model is used to generate a capability score for each R&D personnel, and the R&D personnel are divided into different capability levels according to the capability score based on the preset threshold range.

[0108] In summary, the method and system for accurately matching R&D needs with R&D personnel of the present invention can achieve accurate matching of R&D needs with R&D personnel, improve R&D efficiency and quality, and reduce resource waste.

[0109] The above specific examples are used to explain the principles and implementation methods of the present invention in detail. These examples are only used to help understand the core technical content of the present invention. Based on the above specific embodiments of the present invention, any improvements and modifications made by technicians in this technical field without departing from the principles of the present invention should fall within the scope of patent protection of the present invention.

Claims

1. A method for accurately matching R&D needs with R&D personnel, characterized in that , including the following steps: S1. Obtain R&D requirement documents that comply with description rules; S2: Through natural language processing, document pre-training learning, and deep learning, the acquired R&D requirement documents are automatically decomposed into a set of R&D tasks with the smallest granularity; S3, extract the key features of each R&D task in the set, compare them with the entries in the preset task classification and grading library through a matching algorithm, and find the most matching task type and level; S4. Extract the key features, specific task types and levels of each R&D task in the set, compare them with the R&D personnel in the preset R&D personnel comprehensive rating database through a matching algorithm, find the most suitable R&D personnel with the shortest predicted time, and provide them to the R&D manager through a visual interface; S5. According to the output result of the matching algorithm in step S4, the R&D manager assigns the decomposed R&D tasks to the corresponding R&D personnel, and the R&D personnel corresponding to the R&D task set form a R&D team.

2. A method for accurately matching R&D requirements with R&D personnel according to claim 1, characterized in that: The step S2 specifically includes the following process: S2.1, Natural Language Processing: Perform text preprocessing on R&D requirement documents by segmenting, removing stop words and part-of-speech tagging; extract entities in the text using named entity recognition technology, and determine the association between entities through relationship extraction; perform dependency syntactic analysis on sentences to identify the dependency relationship between words in the sentence; identify the semantic role in each predicate-argument structure through semantic role labeling, and understand the specific event or action described in the sentence; extract keywords and generate a summary by analyzing and summarizing the text content; S2.

2. Use a BERT-based pre-trained model; fine-tune the pre-trained model according to the characteristics of the R&D requirements document to better adapt it to the language rules and terminology of the specified field; use the pre-trained model to generate context-related word embeddings based on the summary text to capture the deep semantic information in the summary text; S2.

3. Build a task-oriented deep learning model. The deep learning model first classifies the summary text into preset categories based on the output of the summary text and the pre-trained model, then labels the words in the summary text through the sequence labeling model, identifies the boundaries and types of tasks, and then calculates the semantic similarity between different parts of the summary text to assist in determining which content belongs to the same task or related tasks, ultimately achieving the goal of breaking down the R&D requirement document into a set of R&D tasks with the smallest granularity.

3. The method for accurately matching R&D requirements with R&D personnel according to claim 1, characterized in that: The matching algorithm selected in step S3 is one of cosine similarity calculation, rule-based matching, decision tree, support vector machine or neural network algorithm; The matching algorithm selected in step S4 is one of rule-based matching, multi-factor comprehensive evaluation, optimization algorithm or machine learning model.

4. The method for accurately matching R&D requirements with R&D personnel according to claim 1, characterized in that: The comprehensive rating database for R&D personnel includes the R&D personnel's skill information, work experience, historical performance, preferred areas and ability level, among which: the R&D personnel's historical performance includes the on-time completion rate of tasks, code quality, speed of solving technical problems, teamwork and communication skills, innovation ability, learning ability, project management and leadership ability, customer feedback, task complexity, workload and work attitude; Based on the historical performance of R&D personnel, a pre-trained quantitative model is used to generate a capability score for each R&D personnel. R&D personnel are divided into different capability levels according to the capability score based on the preset threshold range. After extracting the key features, specific task types and levels of each R&D task in the collection, they are compared with the R&D personnel in the preset R&D personnel comprehensive rating library through a matching algorithm. First, the R&D personnel who match the R&D task are found, and then the time consumption of the R&D task is predicted based on the historical performance of the R&D personnel. Finally, the R&D personnel with the shortest time consumption are selected.

5. The method for accurately matching R&D requirements with R&D personnel according to claim 4, characterized in that: The pre-training process of the quantized model is as follows: Collect historical performance data of R&D personnel and convert the collected data into quantitative features as historical data sets; Use the R&D personnel’s historical data set to train the selected quantitative model to ensure that the quantitative model can accurately map the R&D personnel’s characteristics to their ability scores; Through the trained quantitative model, a comprehensive capability score is generated for each R&D personnel, which reflects the performance of the R&D personnel in various evaluation dimensions; A pre-trained quantitative model is used to generate a capability score for each R&D personnel, and the R&D personnel are divided into different capability levels according to the capability score based on the preset threshold range.

6. A precise matching system between R&D requirements and R&D personnel, characterized in that: It includes: Requirements acquisition module, used to obtain R&D requirements documents that meet the description rules; The document parsing module is used to automatically decompose the acquired R&D requirement documents into a set of R&D tasks with the smallest granularity through three operations: natural language processing, document pre-training learning, and deep learning. The task matching module is used to extract the key features of each R&D task in the set, compare them with the entries in the preset task classification and grading library through the matching algorithm, and find the most matching task type and level; The personnel matching module is used to extract the key features, specific task types and levels of each R&D task in the set, and compare them with the R&D personnel in the preset R&D personnel comprehensive rating database through the matching algorithm to find the most suitable R&D personnel with the shortest predicted time. The task management module is used to assign the R&D tasks decomposed by the document parsing module to the corresponding R&D personnel according to the output results of the personnel matching module. The R&D personnel corresponding to the R&D task set form a R&D team.

7. A precise matching system between R&D requirements and R&D personnel according to claim 6, characterized in that: The document parsing module automatically decomposes the acquired R&D requirement documents into a minimum-granularity R&D task set through three operations: natural language processing, document pre-training learning, and deep learning. The specific process includes: (1) Natural language processing: Perform text preprocessing on R&D requirement documents by segmenting, removing stop words, and tagging parts of speech; extract entities in the text using named entity recognition technology, and determine the relationship between entities through relationship extraction; perform dependency syntactic analysis on sentences to identify the dependency relationship between words in the sentence; identify the semantic role in each predicate-argument structure through semantic role labeling to understand the specific event or action described in the sentence; extract keywords and generate a summary by analyzing and summarizing the text content; (2) Document pre-training learning: Use a BERT-based pre-training model; fine-tune the pre-training model based on the characteristics of the R&D requirement document to make it better adapt to the language rules and terminology of the specified field; use the pre-training model to generate context-related word embeddings based on the summary text to capture the deep semantic information in the summary text; (3) Deep learning: Build a task-oriented deep learning model. The deep learning model first classifies the summary text into preset categories based on the output of the summary text and the pre-trained model. Then, the sequence labeling model is used to label the words in the summary text, identify the boundaries and types of tasks, and then calculate the semantic similarity between different parts of the summary text to assist in determining which content belongs to the same task or related tasks. Ultimately, the goal of breaking down the R&D requirement document into a set of R&D tasks with the smallest granularity is achieved.

8. The precise matching system between R&D requirements and R&D personnel according to claim 6, characterized in that: The matching algorithm selected by the task matching module is one of cosine similarity calculation, rule-based matching, decision tree, support vector machine or neural network algorithm; The matching algorithm selected by the personnel matching module is one of rule-based matching, multi-factor comprehensive evaluation, optimization algorithm or machine learning model.

9. The precise matching system between R&D requirements and R&D personnel according to claim 6, characterized in that: The comprehensive rating database for R&D personnel includes the R&D personnel's skill information, work experience, historical performance, preferred areas and ability level, among which: the R&D personnel's historical performance includes the on-time completion rate of tasks, code quality, speed of solving technical problems, teamwork and communication skills, innovation ability, learning ability, project management and leadership ability, customer feedback, task complexity, workload and work attitude; Based on the historical performance of R&D personnel, a pre-trained quantitative model is used to generate a capability score for each R&D personnel. R&D personnel are divided into different capability levels according to the capability score based on the preset threshold range. The personnel matching module extracts the key features, specific task types and levels of each R&D task in the set, and compares them with the R&D personnel in the preset R&D personnel comprehensive rating library through a matching algorithm. First, the R&D personnel matching the R&D task are found, and then the time consumption of the R&D task is predicted based on the historical performance of the R&D personnel. Finally, the R&D personnel with the shortest time consumption are selected.

10. A precise matching system between R&D requirements and R&D personnel according to claim 9, characterized in that: The pre-training process of the quantized model is as follows: Collect historical performance data of R&D personnel and convert the collected data into quantitative features as historical data sets; Use the R&D personnel’s historical data set to train the selected quantitative model to ensure that the quantitative model can accurately map the R&D personnel’s characteristics to their ability scores; Through the trained quantitative model, a comprehensive capability score is generated for each R&D personnel, which reflects the performance of the R&D personnel in various evaluation dimensions; A pre-trained quantitative model is used to generate a capability score for each R&D personnel, and the R&D personnel are divided into different capability levels according to the capability score based on the preset threshold range.

Images