Problem generation method and device based on multi-agent collaboration and storage medium

Abstract

The invention discloses a question generation method and device based on multi-agent collaboration and a storage medium, and the method comprises the steps: obtaining a post description text, and segmenting the post description text to obtain a plurality of text paragraphs; respectively inputting the plurality of text paragraphs into a plurality of label extraction agents to obtain a plurality of label sets; inputting the plurality of tag sets into an optimization agent to obtain a target tag set; and generating a target question pair according to the target label set, the target question pair comprising a question and an answer. By constructing a multi-agent collaborative mechanism, effective segmentation, parallel processing and result fusion of the post description text are realized, and the accuracy of tag generation is improved, so that the accuracy of problem generation is improved, and the problems that key information is easy to miss and tag coverage is incomplete in the processing process of the post description text are solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly relates to a problem generation method, device, and storage medium based on multi-agent collaboration. Background Art

[0002] With the rapid development of large language models, leveraging their capabilities in generating and expanding tags has become an effective method for solving the bottleneck of resume analysis in the recruitment field. Compared with the traditional manual annotation method, although manual annotation has high accuracy, its process is time-consuming and costly, making it difficult to meet the requirements of constructing large-scale datasets. Existing automatic tag generation methods mainly use carefully designed prompt engineering to guide large language models to automatically generate high-quality tags according to the context semantics.

[0003] In the existing art, the tag generation method of large language models, although alleviating the problems of high cost and low efficiency of manual annotation to a certain extent, still has significant defects in practical applications. For example, when dealing with long texts (job description texts), large language models will face the problem of scattered context attention. Especially when facing recruitment texts containing complex job descriptions and technical details, it is easy to have situations such as missing key information and incomplete tag coverage, making it difficult to meet the requirements of accuracy and professionalism of tags for generating job interview questions, resulting in inaccurate tag generation and inaccurate problem generation.

[0004] Therefore, there is still an urgent need for a problem generation method that can improve the accuracy of tag generation and problem generation. Summary of the Invention

[0005] The main objective of the present invention is to propose a problem generation method, device, and storage medium based on multi-agent collaboration to solve the problem that existing defective large language models are prone to inaccurate tag generation when processing job description texts, thereby causing inaccurate problem generation.

[0006] To achieve the above objective, the present invention proposes a problem generation method based on multi-agent collaboration, and the problem generation method based on multi-agent collaboration includes: Obtain a job description text, and segment the job description text to obtain multiple text paragraphs; Input the multiple text paragraphs into multiple tag extraction agents respectively to obtain multiple tag sets; Input the multiple tag sets into an optimization agent to obtain a target tag set; Generate a target question pair according to the target tag set, where the target question pair includes a question and an answer.

[0007] In some embodiments, the plurality of tag extraction agents include a first tag extraction agent and other tag extraction agents; and the step of inputting the plurality of text paragraphs into the plurality of tag extraction agents to obtain a plurality of tag sets includes: For each of the text paragraphs, input the text paragraph into a first label extraction agent to obtain a first predicted label set; Input the other predicted label sets obtained by other label extraction agents into the first label extraction agent; Obtaining the label set obtained by the first label extraction agent modifying the first predicted label set based on other predicted label sets; The tag sets obtained by each tag extraction agent are aggregated to obtain a plurality of tag sets.

[0008] In some embodiments, inputting the plurality of tag sets into the optimization agent to obtain a target tag set comprises: Inputting a plurality of said label sets into an optimization agent; Obtaining a deduplicated label set obtained by performing deduplication processing on the plurality of label sets by the optimization agent; The deduplication tag set is screened according to a preset tag library to obtain the target tag set.

[0009] In some embodiments, before obtaining the job description text, the method further includes: Crawl Chinese articles from the preset database and obtain original article data; Segmenting the original article data into natural paragraphs to obtain multiple text units; Perform keyword extraction on each of the plurality of text units to obtain keywords corresponding to each of the text units; The keywords corresponding to each of the text units are associated with the tags corresponding to the preset tag library.

[0010] In some embodiments, generating a target question pair according to the target tag set includes: Acquire associated keywords according to each tag in the target tag set; Acquire text units corresponding to the keywords based on the keywords; splicing the text units corresponding to the keywords to obtain a recombined segment; The reorganized segment and the prompt template are input into a preset question generation model to obtain the target question pair.

[0011] In some embodiments, before obtaining the job description text, the method further includes: Collecting multiple original question pairs generated by the general large model according to the preset tag library; Filter multiple of the original question pairs according to preset conditions to obtain a number of question pairs; Save the number of the question pairs to a question library.

[0012] In some embodiments, the generating of the target question pairs according to the target tag set includes: Retrieve the question library according to the target tag set; Determine the question pairs corresponding to the target tag set in the question library as the target question pairs.

[0013] In some embodiments, after the generating of the target question pairs according to the target tag set, it further includes: Evaluate the target question pairs to obtain an evaluation result; Optimize the target question pairs according to the evaluation result.

[0014] The present invention also provides a question generation device based on multi-agent collaboration, including: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the multi-agent collaboration-based question generation method described in any one of the above.

[0015] The present invention also provides a storage medium storing a computer program, the computer program including program instructions, and the program instructions, when executed by a processor, enable the processor to execute the multi-agent collaboration-based question generation method described in any one of the above.

[0016] The present invention processes the job description text through the collaboration of multiple agents. First, the job description text is segmented into multiple text paragraphs, and then multiple label extraction agents respectively extract labels from the multiple text paragraphs to obtain multiple tag sets. Then, the optimization agent optimizes the multiple tag sets to obtain the target tag set. Finally, the target question pairs are generated according to the target tag set. By constructing a multi-agent collaboration mechanism, the effective segmentation, parallel processing, and result fusion of the job description text are realized, the accuracy of label generation is improved, and thus the accuracy of question generation is improved, solving the problems of key information omission and incomplete tag coverage that are prone to occur during the processing of job description text. Description of the Drawings

[0017] Figure 1 It is a schematic flowchart of the multi-agent collaboration-based question generation method in an embodiment of the present invention; Figure 2Another schematic flowchart of the problem generation method based on multi-agent collaboration in an embodiment of the present invention; Figure 3 Another schematic flowchart of the problem generation method based on multi-agent collaboration in an embodiment of the present invention; Figure 4 Another schematic flowchart of the problem generation method based on multi-agent collaboration in an embodiment of the present invention; Figure 5 Another schematic flowchart of the problem generation method based on multi-agent collaboration in an embodiment of the present invention; Figure 6 Another schematic flowchart of the problem generation method based on multi-agent collaboration in an embodiment of the present invention; Figure 7 Another schematic flowchart of the problem generation method based on multi-agent collaboration in an embodiment of the present invention; Figure 8 Another schematic flowchart of the problem generation method based on multi-agent collaboration in an embodiment of the present invention; Figure 9 Schematic structural diagram of the problem generation device based on multi-agent collaboration involved in the embodiment solution of the present invention.

[0018] The realization, functional characteristics and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Detailed implementation manners

[0019] The following will clearly and completely describe the solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts fall within the scope of protection of the present invention.

[0020] It should be noted that all directional indications (such as up, down, left, right, front, back...) in the embodiments of the present invention are only used to explain the relative positional relationship and movement conditions between components in a specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indications will also change accordingly.

[0021] It should also be noted that when an element is referred to as "fixed to" or "disposed on" another element, it can be directly on the other element or there may be an intermediate element at the same time. When an element is referred to as "connected" to another element, it can be directly connected to the other element or there may be an intermediate element at the same time.

[0022] In addition, the descriptions involving "first", "second", etc. in the present invention are for descriptive purposes only, and should not be construed as indicating or implying their relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In addition, the technical solutions between various embodiments may be combined with each other, but it must be based on the fact that those of ordinary skill in the art can implement them. When the combination of technical solutions results in contradictions or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0023] To achieve the above object, the present invention proposes a problem generation method based on multi-agent collaboration. The problem generation method based on multi-agent collaboration includes: Step S110, obtaining a job description text and segmenting the job description text to obtain a plurality of text paragraphs; Step S120, respectively inputting the plurality of text paragraphs into a plurality of tag extraction agents to obtain a plurality of tag sets; Step S130, inputting the plurality of tag sets into an optimization agent to obtain a target tag set; Step S140, generating a target question pair according to the target tag set, where the target question pair includes a question and an answer.

[0024] In this embodiment, referring to Figure 1 and Figure 9 , the problem generation method based on multi-agent collaboration can be applied to a problem generation device based on multi-agent collaboration. The problem generation device based on multi-agent collaboration includes at least one processor and a memory communicatively connected to at least one processor. The problem generation method based on multi-agent collaboration can be stored in the memory; the processor can call the problem generation method based on multi-agent collaboration in the memory, so as to execute the problem generation method based on multi-agent collaboration. In this embodiment, the execution subject of the method steps is the processor.

[0025] It can be understood that the processor is configured with a plurality of agents, where the plurality of agents includes a plurality of tag extraction agents and an optimization agent. The number of tag extraction agents needs to be greater than or equal to the number of text paragraphs, so that each text paragraph can at least correspond to one tag extraction agent.

[0026] When a user needs to use the problem generation device based on multi-agent collaboration to generate job interview questions, the user can input a job description text to the problem generation device based on multi-agent collaboration. For example: the user can be a recruiter, and the job description text input by the recruiter can be about the job content of the position, the responsibilities of the position, the skill requirements of the position, etc. At this time, the processor can obtain the job description text.

[0027] After the processor obtains the job description text, it will split the job description text to split it into multiple text paragraphs. For example: The job description text can be split into equal tokens, and each text paragraph is controlled to not exceed 512 tokens. For example, the total number of tokens in the job description text can be divided by 512 tokens to determine how many text paragraphs the job description text can be divided into.

[0028] After the processor splits to obtain multiple text paragraphs, it can input the multiple text paragraphs into multiple label extraction agents respectively, that is, one label extraction agent is assigned to each text paragraph. Each label extraction agent performs label extraction based on the text paragraph assigned to it to obtain the label set extracted by each label extraction agent, and then each label extraction agent outputs the extracted label set. At this time, the processor can obtain multiple label sets.

[0029] After the processor obtains multiple label sets, it can input the multiple label sets into the optimization agent together. The optimization agent can optimize the multiple label sets. For example: The optimization agent can perform duplicate removal processing on the multiple label sets to eliminate duplicate items. After the optimization agent optimizes the multiple label sets, it can obtain the target label set, and then the optimization agent outputs the optimized target label set. At this time, the processor can obtain the target label set.

[0030] After the processor obtains the target label set, it can generate target question pairs according to the target label set, where the target question pairs include questions and answers.

[0031] In this embodiment, multiple agents cooperate to process the job description text. First, the job description text is split into multiple text paragraphs, then multiple label extraction agents are used to perform label extraction on the multiple text paragraphs respectively to obtain multiple label sets, then the optimization agent is used to optimize the multiple label sets to obtain the target label set, and finally the target question pairs are generated according to the target label set; by constructing a multi-agent cooperation mechanism, effective splitting, parallel processing and result fusion of the job description text are realized, the accuracy of label generation is improved, thereby improving the accuracy of question generation, and solving the problems of easy omission of key information and incomplete label coverage in the process of processing the job description text.

[0032] In some embodiments, the multiple label extraction agents include a first label extraction agent and other label extraction agents; the foregoing inputting the multiple text paragraphs into the multiple label extraction agents respectively to obtain multiple label sets includes: Step S150, for each text paragraph, input the text paragraph into the first label extraction agent to obtain a first predicted label set; Step S151: Input the other predicted label sets obtained by other label extraction agents into the first label extraction agent. Step S152: Obtain the label set obtained by the first label extraction agent to correct the first predicted label set according to the other predicted label sets. Step S153: Aggregate the label sets obtained by each label extraction agent to obtain multiple label sets.

[0033] In this embodiment, referring to Figure 2 , when the processor executes step S120, it can exchange the label sets of each label extraction agent. The multiple label extraction agents may include a first label extraction agent and other label extraction agents, where the first label extraction agent and other label extraction agents are the same label extraction agent. The first label extraction agent and other label extraction agents are relative to the text paragraphs. For each text paragraph, the label extraction agent that the text paragraph is input to is the first label extraction agent for that text paragraph; the label extraction agents that other text paragraphs are input to are other label extraction agents for that text paragraph.

[0034] For example: There are three text paragraphs, namely text paragraph a, text paragraph b, and text paragraph c; there are three label extraction agents, namely label extraction agent a, label extraction agent b, and label extraction agent c; among them, text paragraph a is input to label extraction agent a, text paragraph b is input to label extraction agent b, and text paragraph c is input to label extraction agent c. Then for text paragraph a, label extraction agent a is the first label extraction agent, and label extraction agents b and c are other label extraction agents. Similarly, for text paragraph b, label extraction agent b is the first label extraction agent, and label extraction agents a and c are other label extraction agents.

[0035] For each text paragraph, the processor inputs the text paragraph into the first label extraction agent. The first label extraction agent performs label extraction based on the text paragraph to obtain the first predicted label set of the text paragraph. Then the first label extraction agent will first save the first predicted label set and then output the first predicted label set. At this time, the processor can obtain the first predicted label set. Of course, inputting multiple text paragraphs into multiple label extraction agents is carried out simultaneously. At this time, the processor can obtain the other predicted label sets output by other label extraction agents.

[0036] After the processor obtains the first prediction label set and other prediction label sets, it will input the first prediction label set into other label extraction agents, and also input other prediction label sets into the first label extraction agent. Then, the first label extraction agent can supplement or correct the first prediction label set according to other prediction label sets, so as to obtain a label set, and finally output the label set. At this time, the processor can obtain the label set obtained by the first label extraction agent supplementing or correcting the first prediction label set according to other prediction label sets.

[0037] The processor can obtain multiple label sets by aggregating the label sets obtained by each label extraction agent.

[0038] For example: When text paragraph a is input into label extraction agent a, label extraction agent a can obtain prediction label set a by extracting text paragraph a. When text paragraph b is input into label extraction agent b, label extraction agent b can obtain prediction label set b by extracting text paragraph b. When text paragraph c is input into label extraction agent c, label extraction agent c can obtain prediction label set c by extracting text paragraph c.

[0039] Then, for text paragraph a, label extraction agent a is the first label extraction agent, and label extraction agents b and c are other label extraction agents; prediction label set a is the first prediction label set, and prediction label sets b and c are other prediction label sets. At this time, prediction label sets b and c will be input into label extraction agent a, and prediction label set a will also be input into label extraction agents b and c. Then, label extraction agent a can supplement or correct prediction label set a according to prediction label sets b and c, so as to obtain a label set.

[0040] Similarly, label extraction agent b can supplement or correct prediction label set b according to prediction label sets a and c, so as to obtain a label set. Label extraction agent c can supplement or correct prediction label set c according to prediction label sets a and b, so as to obtain a label set. Finally, the processor aggregates the label sets obtained by each label extraction agent to obtain multiple label sets.

[0041] In this embodiment, each label extraction agent can refer to the output of other label extraction agents to further supplement or correct the prediction label set generated by itself, so as to obtain a label set, thereby enhancing the integrity and consistency of the label set.

[0042] In some embodiments, the foregoing step of inputting multiple label sets into an optimization agent to obtain a target label set includes: Step S160: Input multiple tag sets into the optimization agent; Step S161: Obtain the deduplicated tag set obtained by the optimization agent after deduplicating the multiple tag sets; Step S162: Screen the deduplicated tag set according to the preset tag library to obtain the target tag set.

[0043] In this embodiment, referring to Figure 3 , when the processor executes step S130, it can deduplicate the multiple tag sets. The processor first inputs the obtained multiple tag sets into the optimization agent. Then the optimization agent will deduplicate the multiple tag sets. For example, the optimization agent will integrate the multiple tag sets into a large tag set, and then deduplicate the same tags in this large tag set to obtain the deduplicated tag set. Finally, the optimization agent outputs the obtained deduplicated tag set. At this time, the processor can obtain the deduplicated tag set.

[0044] After the processor obtains the deduplicated tag set, it can screen the deduplicated tag set according to the preset tag library to obtain the target tag set. The preset tag library can be user-defined or generated from the historical tags output by the optimization agent. The processor can compare the deduplicated tag set with the preset tag library and screen out multiple tags that meet the specifications to obtain the target tag set.

[0045] In some embodiments, before obtaining the job description text described above, it further includes: Step S170: Crawl Chinese articles in the preset database to obtain the original article data; Step S171: Perform natural paragraph segmentation on the original article data to obtain multiple text units; Step S172: Extract keywords from each of the multiple text units to obtain the keywords corresponding to each text unit; Step S173: Associate the keywords corresponding to each text unit with the tags corresponding to the preset tag library.

[0046] In this embodiment, referring to Figure 4 , before the processor executes step S110, it also needs to first crawl Chinese articles in the preset database to obtain the original article data. The processor can access the preset database according to the preset interface to crawl the Chinese articles in the preset database. Among them, the Chinese text can be Chinese articles related to the job, such as articles on the technical details of the job. Then these Chinese articles related to the job are combined to obtain the original article data. Then perform natural paragraph segmentation on the original article data to obtain multiple text units.

[0047] After the processor obtains multiple text units, it can perform keyword extraction on the multiple text units, and respectively extract the keywords corresponding to each text unit. Then, the keywords corresponding to each text unit are associated with the tags corresponding to the preset tag library. For example: when the keyword is the same as the tag, the keyword and the tag can be associated; or when the meaning of the keyword is the same as the tag, the keyword and the tag can be associated.

[0048] For example: The processor first crawls Chinese articles in the preset database, cleans the Chinese articles, then uses the cleaned Chinese articles as the original article data, and then splits the original article data into multiple independent text units according to natural paragraphs. Then, keyword extraction is performed on each text unit, the first 3 high-frequency keywords of each text unit are extracted, and they are associated with the tags corresponding to the preset tag library through regular expressions.

[0049] In some embodiments, the foregoing generation of the target question pair according to the target tag set includes: Step S180, obtaining the associated keywords according to each tag in the target tag set; Step S181, obtaining the text units corresponding to the keywords based on the keywords; Step S182, splicing the text units corresponding to the keywords to obtain a recombined segment; Step S183, inputting the recombined segment and the prompt template into a preset question generation model to obtain the target question pair.

[0050] In this embodiment, referring to Figure 5 , when the processor executes step S140, it can generate a target question pair according to the Chinese articles in the preset database. The processor first obtains the associated keywords according to each tag in the target tag set; then obtains the text units corresponding to the keywords according to the keywords. Then, the text units corresponding to the keywords are spliced to obtain a recombined segment; finally, the recombined segment and the prompt template are input into the preset question generation model, and the preset question generation model outputs the target question pair according to the recombined segment and the prompt template; at this time, the processor can obtain the target question pair. When the user needs multiple target question pairs to form a test paper, the above steps can be repeated to obtain more target question pairs.

[0051] Among them, the preset question generation model has a limit on the length of character input (for example: the character length limit is 4096 characters), and this character length limit can determine the preset length. For example: when the processor sequentially splices the text units corresponding to each keyword obtained to get a recombined segment, if the total length of the recombined segment reaches or approaches the preset length, it will not continue splicing; instead, it will splice the text units corresponding to the remaining keywords for the next recombined segment. The processor can assign the same prompt template to each recombined segment, where the prompt template can be used to guide the preset question generation model to generate a structured target question pair. The prompt template can include clear instructions (for example: according to the content in the recombined segment, please generate a question and answer with technical depth).

[0052] In some embodiments, before obtaining the job description text mentioned above, it further includes: Step S190, collecting a plurality of original question pairs generated by the general large model according to the preset tag library; Step S191, screening a plurality of original question pairs according to preset conditions to obtain several question pairs; Step S192, saving the several question pairs to the question library.

[0053] In this embodiment, referring to Figure 6 , before the processor executes step S110, it can also collect a plurality of original question pairs generated by the general large model according to the preset tag library. The processor first collects a plurality of original question pairs generated by the general large model according to the preset tag library, then screens the plurality of original question pairs according to preset conditions, so as to screen out several question pairs that meet the preset conditions, and then saves the several question pairs to the question library.

[0054] For example: the processor first collects a large number of original question pairs generated by the general large model based on the tags in the preset tag library, then screens out the parts with higher scores, standardized formats and accurate contents from the large number of original question pairs to obtain several question pairs; then saves the several question pairs to the question library. Among them, the question pairs in the question library are all associated with the tags that generated the question pairs.

[0055] In some embodiments, the foregoing generating the target question pair according to the target tag set includes: Step S200, retrieving the question library according to the target tag set; Step S201, determining the question pairs corresponding to the target tag set in the question library as the target question pairs.

[0056] In this embodiment, referring to Figure 7, when the processor executes step S140, it can also obtain the target question pair from the question library. The processor can retrieve the question library according to the tags in the target tag set, and then determine the question pair corresponding to the tags in the target tag set in the question library as the target question pair. Among them, when the target tag set includes multiple tags, a corresponding question pair can be found for each tag. At this time, each question pair can be the target question pair.

[0057] In some embodiments, after generating the target question pair according to the target tag set as described above, it further includes: Step S210, evaluating the target question pair to obtain an evaluation result; Step S211, optimizing the target question pair according to the evaluation result.

[0058] In this embodiment, referring to Figure 8 , after the processor executes step S140, it will also evaluate and optimize the target question pair. The processor can also be configured with three types of agents: an evaluation agent, an optimization agent, and a control agent. The processor uses the control agent to call the evaluation agent to evaluate the target question pair to obtain an evaluation result; and then calls the optimization agent to optimize the target question pair according to the evaluation result.

[0059] For example: The evaluation agent is mainly responsible for comprehensively analyzing the quality of the target question pair from multiple dimensions. Its evaluation content covers aspects such as difficulty level, practical value, answer correctness, and question rationality. Among them, the difficulty evaluation adopts a three-level classification system, which divides the questions in the target question pair into three levels: high, medium, and low, to facilitate subsequent screening according to the needs of different job levels. The practicality evaluation focuses on whether the question can truly reflect the abilities of the applicant and is closely related to the core responsibilities of the position. The correctness evaluation ensures that the question semantics is clear and unambiguous, and the reference answer accurately corresponds to the requirements of the question stem. The rationality evaluation is used to judge whether the question setting fits the actual work scenario and avoid fictional content that is divorced from the business background.

[0060] The optimization agent performs targeted reconstruction on the target question pair according to the evaluation result of the evaluation agent. The optimization agent not only corrects grammar errors and logical loopholes, but also polishes the expression of the target question pair to make it more in line with the industry term specifications and expression habits. At the same time, during the optimization process, the key features of the original target question pair are retained to maintain the consistency of the intention of the target question pair and ensure the effectiveness of information transmission between agents.

[0061] The control agent serves as the process coordination center for the evaluation agent and the optimization agent, responsible for managing the communication sequence and interaction rhythm among the agents. It adopts a polling scheduling strategy to control the message flow and sets an interaction limit of up to six rounds to control the consumption of computing resources while ensuring the optimization effect. In each round of interaction, the evaluation agent provides new evaluation results, and the optimization agent conducts iterative optimization based on the evaluation results until the preset quality standard is reached or the maximum round limit is reached.

[0062] Among them, the present invention is not limited to the generation of target question pairs corresponding to the job description text. It can also adjust the input description text. The present invention is also applicable to other scenarios that require automatic generation of structured question pairs based on description texts, such as education assessment, technical training, and vocational ability assessment. For example, in the education industry, the job description text can be replaced with a curriculum syllabus, and exercise questions can be generated at this time.

[0063] The present invention processes the job description text through the cooperation of multiple agents. First, the job description text is segmented into multiple text paragraphs, and then multiple tag extraction agents are used to extract tags from the multiple text paragraphs respectively to obtain multiple tag sets. Then, the optimization agent optimizes the multiple tag sets to obtain the target tag set. Finally, the target question pair is generated based on the target tag set. By constructing a multi-agent cooperation mechanism, the effective segmentation, parallel processing, and result fusion of the job description text are realized, the accuracy of tag generation is improved, and thus the accuracy of question generation is improved, solving the problems of easy omission of key information and incomplete tag coverage in the process of processing the job description text.

[0064] The problem generation device based on multi-agent cooperation in the embodiment of the present invention can be a processor capable of running the problem generation method based on multi-agent cooperation; there is at least one processor. As Figure 9 shown, the problem generation device based on multi-agent cooperation may include: a processor 1001 (such as a CPU), a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. Among them, the communication bus 1002 is used to realize the connection and communication among these components. The user interface 1003 may include a display screen (Display) and an input unit, such as a keyboard (Keyboard). Optionally, the user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0065] Those skilled in the art can understand,Figure 9 The structure of the problem generation device based on multi-agent collaboration shown does not limit the problem generation device based on multi-agent collaboration, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0066] As Figure 9 shown, the memory 1005, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a computer program.

[0067] In Figure 9 the problem generation device based on multi-agent collaboration shown, the network interface 1004 is mainly used to connect to the background server and communicate data with the background server; the user interface 1003 is mainly used to connect to the client (user side) and communicate data with the client; and the processor 1001 can be used to call the computer program stored in the memory 1005, and when the computer program is called and executed by the processor 1001, the steps of the above-mentioned problem generation method based on multi-agent collaboration are implemented.

[0068] The present invention also proposes a computer device, which includes a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the computer program, it can execute the problem generation method based on multi-agent collaboration described in any one of the above.

[0069] The present invention also proposes a storage medium, which stores a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor, the processor can execute the problem generation method based on multi-agent collaboration described in any one of the above.

[0070] The above are only partial or preferred embodiments of the present invention. Neither the text nor the drawings can limit the scope of protection of the present invention. Any equivalent structural transformation made by using the content of the specification and drawings of the present invention under the overall concept of the present invention, or direct / indirect application in other related technical fields, is included in the scope of protection of the present invention.

Claims

1. A problem generation method based on multi-agent collaboration, characterized in that The problem generation method based on multi-agent collaboration includes: Obtain the job description text and segment the job description text to obtain multiple text paragraphs; Input the multiple text paragraphs into multiple tag extraction agents respectively to obtain multiple tag sets; Input the multiple tag sets into an optimization agent to obtain a target tag set; Generate a target question pair according to the target tag set, where the target question pair includes a question and an answer.

2. The problem generation method based on multi-agent collaboration according to claim 1, wherein The multiple tag extraction agents include a first tag extraction agent and other tag extraction agents; The step of inputting the multiple text paragraphs into multiple tag extraction agents respectively to obtain multiple tag sets includes: For each text paragraph, input the text paragraph into the first tag extraction agent to obtain a first predicted tag set; Input the other predicted tag sets obtained by the other tag extraction agents into the first tag extraction agent; Obtain the tag set obtained by the first tag extraction agent by correcting the first predicted tag set according to the other predicted tag sets; Aggregate the tag sets obtained by each tag extraction agent to obtain multiple tag sets.

3. The problem generation method based on multi-agent collaboration according to claim 2, wherein The step of inputting the multiple tag sets into an optimization agent to obtain a target tag set includes: Input the multiple tag sets into an optimization agent; Obtain the deduplicated tag set obtained by the optimization agent by performing deduplication processing on the multiple tag sets; Screen the deduplicated tag set according to a preset tag library to obtain the target tag set.

4. The problem generation method based on multi-agent collaboration according to claim 3, characterized in that Before obtaining the job description text, it further includes: Crawl Chinese articles in a preset database to obtain original article data; Perform natural paragraph segmentation on the original article data to obtain multiple text units; Extract keywords from each of the multiple text units respectively to obtain the keywords corresponding to each text unit; Associate the keywords corresponding to each text unit with the tags corresponding to the preset tag library.

5. The problem generation method based on multi-agent collaboration according to claim 4, characterized in that The step of generating a target question pair according to the target tag set includes: Obtain the associated keywords according to each tag in the target tag set; Obtain the text units corresponding to each keyword based on each keyword; Concatenate the text units corresponding to each keyword to obtain a recombined segment; Input the recombined segment and a prompt template into a preset question generation model to obtain the target question pair.

6. The problem generation method based on multi-agent collaboration according to claim 3, wherein Before obtaining the job description text, it further includes: Collect multiple original question pairs generated by a general large model according to the preset tag library; Screen the multiple original question pairs according to preset conditions to obtain several question pairs; Save the several question pairs to a question library.

7. The problem generation method based on multi-agent collaboration according to claim 6, wherein The step of generating a target question pair according to the target tag set includes: Retrieve the question library according to the target tag set; Determine the question pair corresponding to the target tag set in the question library as the target question pair.

8. The problem generation method based on multi-agent collaboration according to any one of claims 5 or 7, characterized in that, After generating the target question pair according to the target tag set, it further includes: Evaluate the target question pair to obtain an evaluation result; Optimize the target question pair according to the evaluation result.

9. A problem generation device based on multi-agent collaboration, characterized in that, It includes: At least one processor; And, A memory communicatively connected to the at least one processor; where, The memory stores instructions that are executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the multi-agent collaborative-based problem generation method according to any one of claims 1 to 8.

10. A storage medium, characterized in that, The storage medium stores a computer program, the computer program includes program instructions, and the program instructions, when executed by a processor, enable the processor to execute the multi-agent collaborative-based problem generation method according to any one of claims 1 to 8.

Images