Intelligent research and judgment method and device for severity of software defects based on large model

Abstract

The invention provides an intelligent research and judgment method and device for software defect severity based on a large model, belongs to the field of software function testing, and constructs cue words for large model dialogues by taking text information described by software defects as input items. The cue words appoint role information of the large model, instruction information of understanding, studying and judging severity and limit information of an answer result. And performing multiple dialogues with the plurality of large models by using the same prompt word, and collecting answers of the large models. All the answer results are grouped and counted, and the result with the highest count and more than half is selected as the final severity degree judgment result. The accuracy of research and judgment is improved, and the labor cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of software function testing, and in particular to a method and device for intelligently analyzing the severity of software defects based on a large model. Background Art

[0002] Determining the severity of software defects is a common scenario in software testing. However, the severity determination is mostly based on the experience of testers and is often not accurate enough. This will bring input bias to subsequent analysis actions such as quality data analysis and R&D team performance analysis. For example, when analyzing defect severity statistics and evaluating and predicting the quality of the tested software, the evaluation conclusion may be too high or too low. When analyzing defect severity statistics and evaluating the individual capabilities of the R&D team, the evaluation conclusion may be inaccurate, affecting individual performance appraisals, etc.

[0003] The current existing method is to build and train a BP neural network model and define a set of weight values ​​for project information, which is also related to the business of the system under test. By inputting the collected information into the BP neural network model, the weight system is combined to assign weights and scores, and finally the severity is determined. This brings several problems: 1. The collected project information is large, but not comprehensive enough. 2. There is only one set of neural network models, which cannot be adjusted dynamically and the judgment is limited. 3. The weight system of project information is related to the business system and is not universal. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a method for intelligently judging the severity of software defects based on a large model, aiming to avoid judgment bias based on personal experience.

[0005] The technical solution of the present invention is:

[0006] A method for intelligently evaluating the severity of software defects based on a large model, comprising:

[0007] Use the text information of the software defect description as input to construct prompt words for large model dialogue;

[0008] The prompt word stipulates the role information of the big model, the instruction information for understanding and judging the severity, and the limiting information of the answer result;

[0009] Use the same prompt words to talk to several large models and collect their answers;

[0010] All the answers are grouped and counted, and the result with the highest count and more than half is selected as the final severity determination result.

[0011] Furthermore,

[0012] Optional general large models include: Baidu AI and Zhipu AI.

[0013] The present invention also provides a large model-based intelligent software defect severity assessment device, which is used to implement the above-mentioned method, including:

[0014] There are two parts: browser extension and backend service;

[0015] Browser extensions are used to provide human-computer interaction and send requests to backend services;

[0016] The backend service obtains defect description information from the request, constructs prompt words, and initiates multi-threaded dialogue requests to several large models; the answer results of all large models are grouped and counted, and the results with the highest count and more than half are selected as the final severity judgment results, which are returned to the browser extension page for display.

[0017] Furthermore,

[0018] When the extension page is opened, it expands in the form of a sidebar; the page contains at least one multi-line text input box for entering defect description information.

[0019] The target site information is bound to the browser extension. The extension page will only be opened when the browser extension is turned on and the browser visits the target site.

[0020] Going further,

[0021] By entering the description of the software defect in the browser extension page and clicking the smart analysis button, the browser extension will send a request to the backend service.

[0022] The browser extension page provides a copy button to copy the software defect description information with one click, making it easy to paste it elsewhere.

[0023] Furthermore,

[0024] The backend service includes an intelligent question-and-answer module and a result determination module. The intelligent question-and-answer module obtains defect description information from the request and constructs prompt words. Multithreading initiates multiple dialogue requests with the large model. The result determination module counts the multiple answer results in groups, selects the result with the highest count and more than half as the final defect severity determination result, and returns it to the browser extension page for display.

[0025] The beneficial effects of the present invention are

[0026] The present invention assists software testers in determining the severity of defects. It avoids the judgment bias based on personal experience, not only ensures the accuracy of the severity of defects, but also provides a data basis for a more fair subsequent quality data analysis and R&D team performance analysis. In the R&D team, the authority of the research and judgment scheme and conclusion of the present invention can also reduce the communication costs between R&D personnel and testers, and avoid additional communication costs between the two parties due to inconsistent severity judgments. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic diagram of the relationship between the device modules of the present invention;

[0028] Figure 2 It is a workflow diagram of the method of the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0030] The method of the present invention includes taking the text information of the software defect description as an input item and constructing prompt words for the large model dialogue. The prompt words stipulate the role information of the large model, the instruction information for understanding and judging the severity, and the limiting information of the answer result. Example of dialogue prompt words:

[0031]

[0032] Dialogue prompts:

[0033] (1) Role information: You are a professional software testing engineer.

[0034] (2) Instruction information: Please understand the following description of the software defect and assign the appropriate severity to the defect.

[0035] (3) Answer result limitation: There are three options for severity: severe, average, and mild. The answer only requires the conclusion, not the analysis process.

[0036] (4) Sample answer: Serious

[0037] (5) Defect description: The value of the apply_occupy_cost field in the mea_budget table was not operated in the "Sauce Liquor Marketing Activities" section of "Declaration of Cancellation Confirmation or Confirmation Rejection".

[0038] Use the same prompt words to have 7 conversations with multiple general large models of Baidu AI and Zhipu AI, and collect 7 answers from the large models. Optional general large models: Baidu AI, Zhipu AI. Group and count all the answer results, and select the result with the highest count and more than half as the final severity judgment result.

[0039] The device of the present invention comprises two parts: a browser extension program and a backend service.

[0040] (1) Browser extensions are used to provide human-computer interaction. When this extension page is opened, it expands in the form of a sidebar. The page contains at least one multi-line text input box for entering defect description information. It contains an "Intelligent Analysis" button and a "Copy Defect" button. The target site information is bound to this extension. The extension page will only be opened when the browser extension is turned on and the browser visits the target site. When the browser visits other non-target sites, the extension will not be opened regardless of whether the browser extension is turned on. When this extension page is opened, it expands in the form of a sidebar. The page contains a multi-line text input box for entering defect description information, and contains an "Intelligent Analysis" button and a "Copy Defect" button.

[0041] The tester enters the description of the software defect in the browser extension page and clicks the "Smart Analysis" button. The extension sends a request to the backend service. Wait for the backend service to return the response result and display it on the page. The tester clicks the "Copy Defect" button to copy the software defect description information to the clipboard, which is convenient for the tester to paste it into other defect management systems.

[0042] (2) The backend service includes an intelligent question-and-answer module and a result determination module. The intelligent question-and-answer module obtains defect description information from the request and constructs prompt words. Multithreading initiates seven dialogue requests with the large model. The result determination module groups and counts the seven answer results, selects the result with the highest count and more than half as the final defect severity determination result, and returns it to the browser extension page for display.

[0043] When using:

[0044] Step 1: Access a defect management system (for example, http: / / defect.demo.com) through a browser and prepare to register the defects you have found.

[0045] Step 2: In the extension program management function of the browser, the extension program of the present invention is opened. The browser now opens a sidebar on a defect management system page.

[0046] Step 3: Enter the defect description information in the input box of the extension program and click the "Smart Judgment" button. Wait for the judgment result of the backend service. For example, enter: Sauce Wine-Marketing Activities-Declare Cancellation Confirmation or Confirmation Rejection without operating the value of the apply_occupy_cost field of the mea_budget table.

[0047] Step 4: Expand the program page to display the return results of the backend service: Average

[0048] Step 5: Click the "Copy Defect" button to copy the defect description information in the input box to the clipboard.

[0049] Step 6: Register the defect in a defect management system and set the severity to average.

[0050] The above description is only a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, and is not used to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for intelligently judging the severity of software defects based on a large model, characterized in that: include: Use the text information of the software defect description as input to construct prompt words for large model dialogue; The prompt word stipulates the role information of the big model, the instruction information for understanding and judging the severity, and the limiting information of the answer result; Use the same prompt words to talk to several large models and collect their answers; All the answers are grouped and counted, and the result with the highest count and more than half is selected as the final severity determination result.

2. The method according to claim 1, characterized in that Optional general large models include: Baidu AI and Zhipu AI.

3. A software defect severity intelligent analysis device based on a large model, used to implement the method of claim 1, characterized in that: include: There are two parts: browser extension and backend service; Browser extensions are used to provide human-computer interaction and send requests to backend services; The backend service obtains defect description information from the request, constructs prompt words, and initiates multi-threaded dialogue requests to several large models; the answer results of all large models are grouped and counted, and the results with the highest count and more than half are selected as the final severity judgment results, which are returned to the browser extension page for display.

4. The device according to claim 3, characterized in that When the extension page is opened, it expands in the form of a sidebar; the page contains at least one multi-line text input box for entering defect description information.

5. The device according to claim 3, characterized in that The target site information is bound to the browser extension. The extension page will only be opened when the browser extension is turned on and the browser visits the target site.

6. The device according to claim 5, characterized in that By entering the description of the software defect in the browser extension page and clicking the smart analysis button, the browser extension will send a request to the backend service.

7. The device according to claim 6, characterized in that The browser extension page provides a copy button to copy the software defect description information with one click, making it easy to paste it elsewhere.

8. The device according to claim 3, characterized in that The backend services include intelligent question-answering module and result determination module; The intelligent question-answering module obtains defect description information from the request and constructs prompt words; multiple threads initiate multiple dialogue requests with the large model; The result determination module groups and counts the results of several answers, selects the result with the highest count and more than half as the final defect severity determination result, and returns it to the page of the browser extension program for display.

Images