Search

CN-121979981-A - Man-machine interaction method and computing device

CN121979981ACN 121979981 ACN121979981 ACN 121979981ACN-121979981-A

Abstract

The embodiment of the application provides a man-machine interaction method and computing equipment, and relates to the technical field of artificial intelligence. The large language model is associated with a plurality of memory sets. The method comprises the steps of obtaining query information input by a user, determining a task target of the query information from a first memory set, wherein the first memory set is used for storing context information in a current session of the user with a large language model, determining supplementary information of the task target from a second memory set, supporting the task target and/or performing personalized configuration on the user, constructing a prompt word based on the task target and the supplementary information, and inputting the prompt word into the large language model to obtain response information of the query information. The method can improve the accuracy of the response of the man-machine interaction system and the adaptability to the habit of the user, and improve the use experience of the user.

Inventors

  • KONG WEIJIE
  • Tian Dengkui
  • JIN HANG

Assignees

  • 河南秦尉数字技术有限公司

Dates

Publication Date
20260505
Application Date
20260104

Claims (10)

  1. 1. A human-machine interaction method, characterized in that the method comprises: Acquiring inquiry information input by a user; Determining a task target of the query information from a first memory set according to the query information, wherein the first memory set is used for storing context information in a current session of a user and the large language model; The supplementary information is used for supporting the task target and/or is used for personalized configuration of the user; constructing a prompt word based on the task target and the supplementary information; And inputting the prompt word into the large language model to obtain response information of the query information.
  2. 2. The method of claim 1, wherein the second set of memory includes a first subset and a second subset, and wherein determining the supplemental information for the task objective from the second set of memory includes: determining a first search result from the first subset as the supplementary information according to the task target, wherein the first search result is used for supporting the task target; And determining a second search result from the second subset as the supplementary information according to the task target, wherein the second search result is used for personalized configuration of the user.
  3. 3. The method of claim 2, wherein the first subset is used to store topics for user sessions with the large language model and contextual information associated with the topics; Determining a first search result from the first subset as the supplemental information according to the task objective, including: in the case that the task objective relates to a topic stored in the first subset, determining the first search result from the first subset, wherein the first search result comprises the topic related to the task objective and associated context information.
  4. 4. The method of claim 2, wherein the second subset is used to store a user representation of the user; determining a second search result from the second subset as the supplemental information according to the task objective, including: In the case that the task objective relates to the personalized demand of the user, determining the second search result from the second subset, wherein the second search result comprises personalized configuration information for the user.
  5. 5. The method according to any one of claims 2-4, further comprising: And when the data volume in the first memory set is larger than a preset threshold value, migrating the context information with earliest time generated in the first memory set to the first sub-set.
  6. 6. The method according to any one of claims 2-5, further comprising: Acquiring target context information to be stored in the first subset; dividing the target context information into at least one segment, wherein one segment corresponds to one theme; for each of the at least one segment, storing the segment and a topic association corresponding to the segment into the first subset.
  7. 7. The method of claim 6, wherein the method further comprises: determining the theme popularity of each theme in the first subset according to the access frequency of the theme and the matching degree of the theme and the user requirement; and clearing the topics with the topic popularity less than the popularity threshold and the corresponding fragments from the first subset.
  8. 8. The method according to any one of claims 2-7, further comprising: the method comprises the steps of acquiring dynamic characteristics of a user from a historical session of the user, wherein the dynamic characteristics are used for indicating preference content of the user in the historical session process; acquiring static characteristics of the user, wherein the static characteristics are preconfigured preference contents of the user; And generating a user portrait of the user according to the dynamic characteristics and the static characteristics of the user, and storing the user portrait into a second subset.
  9. 9. The method according to any one of claims 1-8, wherein said constructing a hint word based on said task goal and said supplemental information comprises: the tracking identification is used for uniquely identifying the current session of the user and the large language model; and splicing the task target carrying the tracking identifier and the supplementary information to construct the prompt word.
  10. 10. A computing device, comprising a processor and a memory, the processor coupled to the memory; the memory is used for storing computer instructions; the computer instructions being loaded and executed by the processor to cause the computing device to implement the human-machine interaction method of any of claims 1-9.

Description

Man-machine interaction method and computing device Technical Field The embodiment of the application relates to the technical field of artificial intelligence, in particular to a man-machine interaction method and computing equipment. Background With the development of large language model (large language models, LLM) technology, man-machine interaction systems built on the basis of large language models have been widely used in multiple scenarios such as virtual assistants, intelligent customer service, educational coaching, programming assistance, document summarization, data analysis assistance, etc. However, the large language model in the current man-machine interaction system has limited understanding capability on the content input by the user, so that the responsive content has defects in the aspects of accuracy, scene suitability and the like, and the use experience of the user is affected. Disclosure of Invention The embodiment of the application provides a man-machine interaction method and computing equipment, which can improve the accuracy of a man-machine interaction system in responding to content and the adaptability to the habit of a user, and improve the use experience of the user. In a first aspect, an embodiment of the present application provides a human-computer interaction method, where the method includes obtaining query information input by a user, determining a task target of the query information from a first memory set according to the query information, where the first memory set is used to store context information in a current session between the user and a large language model, determining supplementary information of the task target from a second memory set, where the supplementary information is used to support the task target and/or is used to perform personalized configuration on the user, constructing a prompt word based on the task target and the supplementary information, and inputting the prompt word into the large language model to obtain response information of the query information. The embodiment of the application provides a man-machine interaction method, which comprises the steps of determining a task target from a first memory set and determining supplementary information of the task target from a second memory set according to query information after query information input by a user is acquired, and further constructing a prompt word based on the task target and the supplementary information. The task targets included in the prompt words can ensure that response information output by the large language model does not deviate from the current conversation scene, and continuity and accuracy of conversation are ensured. In addition, the supplementary information in the prompt word can be used for supporting a task target, so that the content of the response information can be more fit with the requirements of users, and the accuracy of the response information is further ensured. The supplementary information in the prompt word can also be used for personalized configuration information of the user, so that the suitability of the content of the response information and the habit of the user can be improved. Therefore, the response information of the query information is generated by constructing the prompt word input large language model, so that the session continuity in the human-computer interaction process can be ensured, the accuracy of the reply content can be ensured, the personalized requirements of the user can be met, and the use experience of the user can be effectively improved. In a possible implementation manner, the second memory set comprises a first sub-set and a second sub-set, and determining the supplementary information of the task target from the second memory set comprises determining a first search result from the first sub-set as the supplementary information according to the task target, wherein the first search result is used for supporting the task target, determining a second search result from the second sub-set as the supplementary information according to the task target, and the second search result is used for personalized configuration of a user. A specific scheme for determining the supplemental information is provided, and the feasibility of the scheme is ensured. In another possible implementation, the first subset is used for storing topics of conversations of the user with the large language model and context information associated with the topics, and determining the first search result from the first subset as supplemental information according to the task objective includes determining a second search result from a subset, where the task objective relates to the topics stored in the first subset, the second search result including the topics to which the task objective relates and associated context information. The method and the device provide a specific scheme for determining the supplementary information according