CN-121996787-A - Tag processing method, tag processing system, electronic device and computer program product

Abstract

The application discloses a tag processing method, a tag processing system, electronic equipment and a computer program product, and relates to the fields of large model technology and natural language processing. The method comprises the steps of obtaining historical dialogue data, wherein the historical dialogue data are used for representing dialogue data generated by at least one round of interaction reply between a historical service object and a historical serviced object in a historical period, analyzing the historical dialogue data by utilizing a data processing model to obtain a plurality of initial dialogue labels of the serviced object, and clustering the initial dialogue labels according to a clustering strategy to obtain clustered initial dialogue labels, wherein the clustering strategy is used for representing a rule for clustering the initial dialogue labels, and the clustered initial dialogue labels are used for enabling at least one round of interaction reply between the service object and the serviced object. The application solves the technical problem of low label processing efficiency.

Inventors

• ZHOU KANG

Assignees

• 阿里云计算有限公司

Dates

Publication Date

20260508

Application Date

20241107

Claims (20)

1. 1. A method of processing a label, comprising: acquiring historical dialogue data, wherein the historical dialogue data is used for representing dialogue data generated by at least one round of interactive reply between a historical service object and a historical serviced object in a historical period; analyzing the historical dialogue data by using a data processing model to obtain a plurality of initial dialogue labels of the served object, wherein the data processing model is obtained by training a generating model by using dialogue data samples, and the initial dialogue labels are used for representing dialogue intentions of the historical served object; And clustering the initial dialogue labels according to a clustering strategy to obtain clustered initial dialogue labels, wherein the clustering strategy is used for representing a rule for clustering the initial dialogue labels, and the clustered initial dialogue labels are used for enabling at least one round of interaction reply between a service object and a served object.
2. 2. The method of claim 1, wherein analyzing the historical dialog data using a data processing model to obtain a plurality of initial dialog tags for the served object comprises: And analyzing the historical dialogue data by using the data processing model to obtain the initial dialogue tag and an interpretation text of the initial dialogue tag, wherein the interpretation text is used for representing the meaning of the initial dialogue tag.
3. 3. The method of claim 2, wherein the historical dialog data includes sub-historical dialog data for different rounds, analyzing the historical dialog data using the data processing model to obtain the initial dialog tag, and the interpretation text of the initial dialog tag, comprising: Sequencing the sub-history dialogue data according to the turn of the sub-history dialogue data to obtain sequenced sub-history dialogue data; And analyzing the sequenced sub-history dialogue data by using the data processing model to obtain the initial dialogue labels and the interpretation text of the sequenced sub-history dialogue data.
4. 4. A method according to claim 3, wherein analyzing the ordered sub-historical dialog data using the data processing model to obtain the initial dialog tag of the ordered sub-historical dialog data, and the interpretation text, comprises: Marking identification information on the sequenced sub-history dialogue data to obtain marked sub-history dialogue data, wherein the identification information is used for indicating that the sub-history dialogue data is from the history service object or the history served object; And analyzing the marked sub-history dialogue data according to the turns of the sub-history dialogue data by using the data processing model to obtain the initial dialogue tag of the marked sub-history dialogue data and the interpretation text.
5. 5. The method of claim 2, wherein clustering the initial session tags according to a clustering strategy to obtain clustered initial session tags comprises: performing de-duplication processing on the initial dialogue tag and the interpretation text respectively; vector encoding is carried out on the initial dialogue tag after the duplication removal to obtain the encoded initial dialogue tag, and vector encoding is carried out on the interpretation text after the duplication removal to obtain the encoded interpretation text; And clustering the coded initial dialogue labels according to the clustering strategy to obtain a clustering result, and generating the clustered initial dialogue labels based on the clustering result and the coded interpretation text.
6. 6. The method of claim 5, wherein clustering the encoded initial session tags according to the clustering strategy to obtain a clustering result comprises: And clustering the coded initial dialogue labels by using the clustering strategy in a clustering model to obtain the clustering result, wherein the clustering model is obtained by using machine learning training.
7. 7. The method of claim 5, wherein generating the clustered initial dialog tag based on the clustered results and the encoded interpretation text comprises: And analyzing the clustering result and the encoded interpretation text by using the data processing model to obtain the clustered initial dialogue labels.
8. 8. The method according to claim 1, wherein the method further comprises: And adjusting the clustered initial dialogue labels according to an adjustment strategy corresponding to the clustering strategy to obtain at least one target dialogue label.
9. 9. The method of claim 8, wherein the clustering policy includes a first clustering policy and a second clustering policy, the similarity between different initial session tags in a cluster corresponding to the first clustering policy is smaller than the similarity between different initial session tags in a cluster corresponding to the second clustering policy, and the adjusting the clustered initial session tags according to an adjustment policy corresponding to the clustering policy, to obtain at least one target session tag includes: Responding to the clustering strategy as the first clustering strategy, and splitting the clustered initial conversation label according to a first adjustment strategy corresponding to the first clustering strategy to obtain the target conversation label, wherein the first adjustment strategy is used for representing a rule for splitting the clustered initial conversation label; And responding to the clustering strategy as the second clustering strategy, and merging the initial dialogue labels after different clustering according to a second adjustment strategy corresponding to the second clustering strategy to obtain the target dialogue labels, wherein the first adjustment strategy is used for representing rules for merging different initial dialogue labels after clustering.
10. 10. The method of claim 9, wherein responding to the clustering policy being the first clustering policy, splitting the clustered initial session tags according to a first adjustment policy corresponding to the first clustering policy to obtain the target session tags, comprising: Responding to the clustering strategy as the first adjustment strategy, and splitting the clustered initial dialogue labels by utilizing the data processing model according to the first adjustment strategy to obtain a plurality of sub dialogue labels and a plurality of label categories; and determining the sub-dialogue labels classified into the corresponding label categories as the target dialogue labels.
11. 11. The method according to claim 10, wherein the method further comprises: And in response to at least one first target sub-dialog tag in the sub-dialog tags not being classified into the tag class, or the first target sub-dialog tag being classified into a different tag class, reclassifying the first target sub-dialog tag into a tag class to which a second target sub-dialog tag in the sub-dialog tags belongs, wherein the similarity between the second target sub-dialog tag and the first target sub-dialog tag is greater than a first similarity threshold.
12. 12. The method of claim 9, wherein, in response to the clustering policy being the second clustering policy, merging the initial session tags after different clustering according to a second adjustment policy corresponding to the second clustering policy to obtain the target session tag, including: And responding to the clustering strategy as the second clustering strategy, and merging the initial dialogue labels after different clusters with the similarity larger than a second similarity threshold by using the data processing model to obtain the target dialogue labels.
13. 13. The method according to any one of claims 8 to 12, further comprising: determining the total number of dialogue rounds of interactive reply between the service object and the object to be serviced by utilizing the target dialogue label and the number of dialogue rounds of interactive reply failure by utilizing the target dialogue label; Determining a quality index of the target dialogue tag based on the number of dialogue rounds and the total number of dialogue rounds, wherein the quality index is used for representing the frequency of failure in interactive reply by using the target dialogue tag; and adjusting the clustered initial dialogue labels by using the quality index.
14. 14. A reply information generation method, characterized by comprising: displaying inquiry information of the served object on an operation interface; Responding to an information reply operation acted on the operation interface, and displaying target dialogue labels matched with the inquiry information in a dialogue label tree on the operation interface, wherein the dialogue label tree is obtained by establishing different target dialogue labels according to a tree data structure, the target dialogue labels are used for enabling a service object and the object to be served to carry out at least one round of interaction reply, the target dialogue labels are obtained by adjusting clustered initial dialogue labels according to an adjustment strategy corresponding to a clustering strategy, the clustering strategy is used for representing rules for clustering the initial dialogue labels, the clustered initial dialogue labels are obtained by carrying out clustering processing on the initial dialogue labels according to the clustering strategy, the initial dialogue labels are used for representing dialogue intention of historical objects to be served, and are obtained by analyzing historical dialogue data by utilizing a data processing model, the data processing model is obtained by training a dialogue data sample pair, and the historical dialogue data is used for representing dialogue data generated by enabling the historical service object and the historical objects to carry out at least one round of interaction reply in a historical period; And displaying dialogue flow information matched with the query information on the operation interface, wherein the dialogue flow information is generated according to the target dialogue label matched with the query information, and the dialogue flow information is used for representing the process of interactive reply with the served object.
15. 15. A reply information generation method, characterized by comprising: displaying inquiry information of the served object on an operation interface; Responding to an information reply operation acted on the operation interface, displaying target dialogue labels matched with the inquiry information in a dialogue label library on the operation interface, wherein the dialogue label library comprises different target dialogue labels, the target dialogue labels are used for enabling a service object and a served object to conduct at least one round of interaction reply, the target dialogue labels are obtained by adjusting clustered initial dialogue labels according to an adjustment strategy corresponding to a clustering strategy, the clustering strategy is used for representing rules for clustering the initial dialogue labels, the clustered initial dialogue labels are obtained by clustering the initial dialogue labels according to the clustering strategy, the initial dialogue labels are used for representing dialogue intentions of historical served objects, and are obtained by analyzing historical dialogue data by using a data processing model, the data processing model is obtained by training a generated model by using dialogue data samples, and the historical dialogue data are used for representing dialogue data generated by enabling a historical service object and the historical served object to conduct at least one round of interaction reply in a historical time period; And displaying reply information matched with the inquiry information on the operation interface.
16. 16. A method of processing a label, comprising: Acquiring historical dialogue data by calling a first interface, wherein the first interface comprises a first parameter, the parameter value of the first parameter is the historical dialogue data, and the historical dialogue data is used for representing dialogue data generated by at least one round of interaction reply between a historical service object and a historical served object in a historical period; analyzing the historical dialogue data by using a data processing model to obtain a plurality of initial dialogue labels of the served object, wherein the data processing model is obtained by training a generating model by using dialogue data samples, and the initial dialogue labels are used for representing dialogue intentions of the historical served object; clustering the initial dialogue labels according to a clustering strategy to obtain clustered initial dialogue labels, wherein the clustering strategy is used for representing a rule for clustering the initial dialogue labels, and the clustered initial dialogue labels are used for enabling at least one round of interaction reply between a service object and a served object; And outputting the clustered initial dialogue labels by calling a second interface, wherein the second interface comprises a second parameter, and the parameter value of the second parameter is the clustered initial dialogue labels.
17. 17. A label processing system, comprising: The client is used for uploading historical dialogue data, wherein the historical dialogue data is used for representing dialogue data generated by at least one round of interactive reply between a historical service object and a historical served object in a historical period; The server is used for analyzing the historical dialogue data by utilizing a data processing model to obtain a plurality of initial dialogue labels of the served object, wherein the data processing model is obtained by training a generated model by utilizing dialogue data samples, the initial dialogue labels are used for representing dialogue intentions of the historical served object, clustering is carried out on the initial dialogue labels according to a clustering strategy to obtain the clustered initial dialogue labels, the clustering strategy is used for representing a rule for clustering the initial dialogue labels, the clustered initial dialogue labels are used for enabling at least one round of interaction reply between the served object and the served object, and the clustered initial dialogue labels are returned to the client.
18. 18. An electronic device, comprising: A memory storing an executable program; A processor for executing the program, wherein the program when run performs the method of any one of claims 1 to 16.
19. 19. A computer readable storage medium, characterized in that the computer readable storage medium comprises a stored executable program, wherein the executable program when run controls a device in which the storage medium is located to perform the method of any one of claims 1 to 16.
20. 20. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 16.

Description

Tag processing method, tag processing system, electronic device and computer program product Technical Field The present application relates to the field of large model technology and natural language processing, and in particular, to a tag processing method, system, electronic device, and computer program product. Background Currently, with the continuous development and application of artificial intelligence technology, dialogue systems are widely used. The intention recognition capability is a core link for understanding the demands of users and providing accurate services, and has become a key for realizing intelligent interaction of a dialogue system. In the related art, the process of intent recognition relies on a manually defined tagging system, i.e., by a professional presetting a series of possible user intentions as tags, using the tag training model described above to enable recognition and classification of the user's intent in a conversation. However, with the continuous expansion of application scenes and the increasing diversity of user expressions, the limitations of manually defined tag systems are increasingly highlighted. For the manpower cost of manually defining the label system, the method needs a large number of professionals to analyze, set and adjust for constructing and maintaining a comprehensive label system, which is time-consuming and labor-consuming and has high cost. For maintenance and update of a manually defined tag system, continuous update and optimization of the tag system is required as user expressions change and new scenes appear. The manual update speed is slow, and the scene task development and the change of the user demands are difficult to keep pace with. For flexibility, when an unknown or non-standardized user intention is met by the manually defined label system, new dialogue labels are often required to be manually added or existing labels are required to be adjusted, so that complete automatic processing cannot be realized, and the response speed and the adaptability of the system are reduced. Aiming at the processing efficiency of a manual definition tag system, the manual processing cannot be efficiently completed in the face of massive dialogue data. Therefore, there is still a technical problem that the label processing cannot be performed effectively. In view of the above problems, no effective solution has been proposed at present. Disclosure of Invention The embodiment of the application provides a tag processing method, a tag processing system, electronic equipment and a computer program product, which are used for at least solving the technical problem of low tag processing efficiency. According to one aspect of the embodiment of the application, a tag processing method is provided. The method comprises the steps of obtaining historical dialogue data, wherein the historical dialogue data are used for representing dialogue data generated by at least one round of interaction reply between a historical service object and a historical serviced object in a historical period, analyzing the historical dialogue data by utilizing a data processing model to obtain a plurality of initial dialogue labels of the serviced object, wherein the data processing model is obtained by training a generating model by utilizing dialogue data samples, the initial dialogue labels are used for representing dialogue intentions of the historical serviced object, clustering the initial dialogue labels according to a clustering strategy to obtain clustered initial dialogue labels, wherein the clustering strategy is used for representing rules for clustering the initial dialogue labels, and the clustered initial dialogue labels are used for enabling at least one round of interaction reply between the service object and the serviced object. According to another aspect of the embodiment of the present application, there is provided a method of generating reply information. The method comprises the steps of displaying inquiry information of a served object on an operation interface, responding to an information response operation acted on the operation interface, displaying target dialogue labels matched with the inquiry information in a dialogue label tree on the operation interface, wherein the dialogue label tree is obtained by building different target dialogue labels according to a tree data structure, the target dialogue labels are used for enabling the served object and the served object to conduct at least one round of interactive response, the target dialogue labels are obtained by adjusting clustered initial dialogue labels according to an adjustment strategy corresponding to a clustering strategy, the clustering strategy is used for representing rules for clustering the initial dialogue labels, the clustered initial dialogue labels are obtained by clustering the initial dialogue labels according to the clustering strategy, the initial dialogue labels are used for represe