Device for controlling results of model questions and answers through role definition

Abstract

The invention discloses a device for controlling a model question and answer result through role definition, and the device sets different roles through a role model setting module, and carries out the output limitation of the roles. The role automatic verification module verifies the content output by the role model by using a verification rule; a model result output comparison module performs similarity verification on files according to output results of different roles, compares similar text contents in different roles one by one, and confirms role model versions; performing question-answering by using a role model question-answering module; meanwhile, the model optimization module can configure output elements and form a specific role model according to the current role of the user, and defines a content generation method taking figures, things, events, targets and scenery as the center by performing condition combination setting on the model, so that the accuracy is greatly improved; meanwhile, sharing among different role versions is achieved through the model sharing module, and continuous optimization learning is conducted on the role template through the model continuous learning module.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence, and particularly relates to a device for controlling the result of model question and answer through role definition. BACKGROUND

[0002] Different people obtain different reply effects in the process of question and answer of an existing large model, a general data model can only satisfy part of 'high-end users', and cannot obtain answers to problems from different angles according to different user identities or post requirements, and cannot provide professional model experience for different users. SUMMARY

[0003] The application aims to provide a device for controlling the result of model question and answer through role definition to solve the problems in the background art.

[0004] To achieve the above-mentioned purpose, the application provides the following technical scheme: a device for controlling the result of model question and answer through role definition, comprising a role model setting module, a role automatic verification module, a model result output module, a role model question and answer module, a model optimization module, a model sharing module and a model continuous learning module, The role model setting module is used for setting different roles and output limiting of different model roles. The role automatic verification module is used for verifying and scoring the content output by the role module according to the role model setting and verification rules. The model result output comparison module is used for file similarity checking according to different role version models and different output results, and one-to-one comparison of similar text content in different role templates to finally confirm the model version. The role model question and answer module is used for knowledge question and answer according to user roles. The model optimization module is used for configuring the elements of model output according to the current role of the user to form a specific role model. The model sharing module is used for sharing between different role versions. The model continuous learning module is used for continuous optimization and learning of the role template.

[0005] Preferably, in the role automatic verification module, problem checking rules corresponding to different roles are set, in the checking process, the structure of model output is compared with the set template content item by item according to the role setting in the role model setting module, and calculation is performed according to different dimensions.

[0006] Preferably, the score rule of the design model calculation is: score = chapter * k + output limit * M + text format * N + content accuracy * w + sensitive word filtering * x + other, wherein k, M, N, w are percentage numbers, and the percentage values of k, M, N, w change according to different roles; if the score range is not less than 90, it is determined to be adopted; if the score range is less than 90, it is determined not to be used; and according to the score result, each version is adjusted and each historical version is saved.

[0007] Preferably, when the model result output comparison module compares, different versions of role models are selected, and different output results are obtained by inputting problems into the models; for the same kind of role, different output structures are used to check the similarity of the files, and similar text contents in different role templates are compared one by one; when the content of a role version is clicked, the output content of another role version is synchronously displayed, and the final role version is determined.

[0008] Preferably, in the model optimization module, if the role model rule is not applicable to the current user, the user adopts an interface mode to customize the selection elements and perform parameterized configuration, so as to form a role model suitable for the current user.

[0009] Preferably, the role model question and answer module first determines the current role of the system after the user logs in the system, obtains the role assigned by the system, and if the user has multiple roles, the user is switched to different roles for question and answer, and the user can score the corresponding results.

[0010] Preferably, the sharing mode in the model sharing module includes role version sharing and multi-role model template automatic matching; the premise of the multi-role model template automatic matching is that the model is described in a tagged manner and is disclosed.

[0011] Preferably, the sharing mode in the model sharing module includes role version sharing and multi-role model template automatic matching; the premise of the multi-role model template automatic matching is that the model is described in a tagged manner and is disclosed.

[0012] Preferably, the main mode of continuous learning in the model continuous learning module is to collect the required contents in the template, and 'optimization suggestions' are proposed for the role templates without the same constraint conditions, so as to further optimize the role templates.

[0013] Compared with the prior art, the present application has the following advantages: (1) The present application defines a content generation method centered on characters, things, events, targets and visions by setting conditions for the model, and greatly improves the accuracy.

[0014] (2) Users of the present invention can construct different model data specifications according to different job requirements, optimize and test the role model, and directly share it with others online, thereby facilitating the sharing and reuse of models and avoiding duplicate investment and waste of resources. DETAILED DESCRIPTION

[0015] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. Example

[0016] The present invention provides a device for controlling the results of model question and answer through role definition, including a role model setting module, a role automatic verification module, a model result output module, a role model question and answer module, a model optimization module, a model sharing module, and a model continuous learning module. The role model setting module creates different roles, including master control roles and application roles. The master control role can be used to limit the overall output of the model, such as blocking sensitive words at the beginning or taking measures for sensitive results. The application role can further restrict the output content of the current role model based on the current role type. For example, due to different business scopes, different model roles can be set for different business scopes, such as document clerks, reviewers, or other roles, with different chapter formats, different output restriction specifications, different output text formats, content accuracy, sensitive word filtering, and other restrictions. This allows the output of the question and answer content to be output in a certain chapter structure, with a certain number of words, and in what form: text or a combination of text and images. The images can be in the form of mind maps or other modes, or based on whether the content is accurate, etc. The role object range includes people, things, events, goals, and visions. When setting roles in the model, the specific meaning of the role is described by using extended conditions to set the model's learning of the specific deep meaning of the role, facilitating the model's understanding of the role. Common expansion criteria include: naming, connotation, conditions, attributes, time, location, space (scope of influencing roles), words, sentence composition, large model generation keywords, vectors, and model content thresholds. Roles include the following: high-quality development, socialism, civilized society, good people and good deeds, and private enterprise roles.

[0017] After the role model is set, the role automatic verification module verifies according to the role category and the role setting, and outputs corresponding reasoning content according to the role setting and the corresponding rules from different dimensions. If the role is a letter receiver, the reasoning content is output according to the role of the letter receiver. If the role is an auditor, the reasoning content is output according to the role of the auditor. Therefore, according to different roles, the content output by the role module is verified and scored. In the verification process, according to the role setting in the role model setting module, the problem is automatically verified, the structure of the model output is compared with the template content, and the matching degree is associated. Whether the model output result meets the standard is verified, for example, when the role is an auditor, the output problem of the auditor is compared with the template content structure, and the two are compared item by item. According to different dimensions, the score is calculated. The calculation rule is: score = chapter * k + output limit * M + text format * N + content accuracy * w + sensitive word filtering * x + other, wherein k, M, N, and w are percentage numbers, and the percentage values of k, M, N, and w change according to different roles. If it is an auditor type problem, the value of k can be 30%, the value of M can be 20%, the value of N can be 10%, the value of w can be 20%, the value of x can be 15%, and the value of other restrictions can be defined by the user, which can be 5%. Then, according to the actual output situation, the auditor's problem is scored. If the score range is not less than 90, it is determined to be adopted. If the score range is less than 90, it is not determined to be used. According to the score result, each version is adjusted and saved.

[0018] After the role automatic verification module has been verified and scored, it is determined whether to use the version according to the score. If it is determined to use the version, the model result output comparison module verifies the similarity of the file according to different role version models combined with different output results, and compares the similar text content in different role templates one by one to finally confirm the model version. For the same kind of role, the similarity of the file is verified by different output structures, and the similar text content in different role templates is compared one by one. When a role version content is clicked, the output content of another role version is synchronously displayed, and the final role version is determined. For example, when A is clicked, the output content of B is synchronously displayed, and the final role model version is selected and confirmed.

[0019] The role model question and answer module is actually used in the question and answer module. After the user logs in the system, the current role of the system is determined, and the role assigned by the system is obtained. If the user has multiple roles, for example, when working part-time, the user can switch different role modes, ask knowledge, and support the user to score the corresponding results. The corresponding scoring dimension can be from one star to five stars.

[0020] The model optimization module configures the elements of the model output according to the current role of the user when in use, and the element content can be chapter, output limit, text limit and the like, and then a specific role model is formed; in the model optimization module, if the role model rule is not applicable to the current user, the user can use the interface to customize the elements of the selection interface, such as chapter, output limit, text format, accuracy of content and the like, to perform parameterized configuration, form a role model suitable for the current user, and some users can directly reference the limit module rule of his current role when the role model rule given by him is not suitable or not particularly satisfactory, and use the interface to select and parameterize the configuration of the elements that may affect the output of the model, to form a unique role model of his own; The model sharing module is shared between different role versions, and the user can select to share the role version for others to use; or the model is labeled and described, the characteristic information of the model is described and classified, and is selected to be public, and other users can automatically match and select the use of the role template; The model continuous learning module continuously optimizes and learns the role template, and can collect the common requirements in the template and give suggestive 'optimization suggestions' for the role templates that do not involve such constraints, so as to promote the continuous upgrading and persistent learning of the role model.

[0021] It is apparent to those skilled in the art that the present application is not limited to the details of the foregoing exemplary embodiments, and can be implemented in other concrete forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be considered in all respects as illustrative and not restrictive, and the scope of the present application is defined by the appended claims rather than the above description, and it is intended to encompass all changes falling within the meaning and scope of the equivalent elements of the claims.

Claims

1. A device for controlling the results of model question-answering through role definition, characterized by: Including role model setting module, role automatic verification module, model result output module, role model question and answer module, model optimization module, model sharing module, model continuous learning module, The role model setting module is used to set different roles and perform output restrictions on different model roles; The role automatic verification module is used to verify and score the content output by the role module according to the role model settings and verification rules; The model result output comparison module is used to perform file similarity verification based on different role version models and different output results, and compare similar text contents in different role templates one by one to finally confirm the model version; The role model question-answering module is used to conduct knowledge questions and answers based on the user role; The model optimization module is used to configure the elements of the model output according to the user's current role to form a specific role model; The model sharing module is used for sharing between different role versions; The model continuous learning module is used to continuously optimize and learn the role template.

2. The device for controlling the results of model question-answering through role definition according to claim 1, characterized in that: The role categories in the role model setting module include master control roles and application roles; the master control role is used to limit the overall output of the model; the application role refers to further restrictions on the output content of the current role model based on the current role type; the restriction range includes chapter format, output specifications, text format, content accuracy, and sensitive word filtering; the role object range includes people, things, events, goals, and visions; when setting roles in the model, the extension conditions are used to describe the specific meaning of the role, which is used to set the model's learning of the specific deep meaning of the role; the role extension condition range includes naming, connotation, conditions, attributes, time, place, spatial words, sentence composition, large model generation keywords, vectors, and thresholds that affect model content; role content includes high-quality development, socialism, civilized society, good people and good deeds, and private enterprises.

3. The device for controlling the results of model question-answering through role definition according to claim 1, characterized in that: In the role automatic verification module, corresponding question verification rules are set according to different roles. During the verification process, according to the role settings in the role model setting module, automatic question verification is used to compare the structure of the model output with the set template content item by item, and calculations are performed according to different dimensions.

4. The device for controlling the results of model question-answering through role definition according to claim 3, characterized in that: The scoring rule for design model calculation is: score = chapter*k+output restriction*M+text format*N+content accuracy*w+sensitive word filtering*x+others, where k, M, N, and w are percentages, and the percentage values ​​of k, M, N, and w vary according to different roles; if the score range is not less than 90, it will be adopted; if the score range is less than 90, it will not be used; at the same time, according to the scoring results, each version will be adjusted and each historical version will be saved.

5. The device for controlling the results of model question-answering through role definition according to claim 1, characterized in that: The model result output comparison module selects different versions of role models when performing comparison, and obtains different output results by inputting questions into the model; For the same role, different output structures are used to verify the similarity of files, and similar text contents in different role templates are compared one by one. When you click on the content of one role version, the output content of another role version is displayed synchronously, and the final role version is determined.

6. The device for controlling the results of model question-answering through role definition according to claim 1, characterized in that: In the model optimization module, if the role model rules are not applicable to the current user, the user uses an interface-based approach to customize selection elements and perform parameterized configuration to form a role model that suits the current user.

7. The device for controlling the results of model question-answering through role definition according to claim 1, characterized in that: After the user logs in to the system, the role model question and answer module first determines the current system role and obtains the role assigned by the system. If the user has multiple roles, the user is switched to different roles for question and answer, and the corresponding results of the user are supported for scoring.

8. The device for controlling the results of model question-answering through role definition according to claim 1, characterized in that: The sharing methods in the model sharing module include role version sharing and automatic matching of multi-role model templates; the premise of automatic matching of multi-role model templates is to describe the model in a labeled manner and make it public.

9. The device for controlling the results of model question-answering through role definition according to claim 1, characterized in that: The main method of continuous learning in the model continuous learning module is to use the requirements in the collection template for the continuously generated role templates, and to propose "optimization suggestions" for role templates that do not have the same constraints to further optimize the role templates.