Search

CN-117131180-B - E-commerce customer service question and answer method and device, equipment and medium thereof

CN117131180BCN 117131180 BCN117131180 BCN 117131180BCN-117131180-B

Abstract

The application relates to an e-commerce customer service question answering method, a device, equipment and a medium thereof in the technical field of e-commerce information processing, wherein the method comprises the steps of responding to a customer service starting event and acquiring a question text; the method comprises the steps of carrying out semantic matching on the questioning text and text contents of a plurality of different structures of each preset explanation document, determining a matching score corresponding to each text content, determining a correlation score between each preset explanation document and the questioning document according to the matching score, screening at least one preset explanation document with the correlation score meeting preset conditions as a candidate document, and determining a reply text according to the at least one candidate document by adopting a large language model. The application can accurately reply the question.

Inventors

  • JIANG ZHISHENG

Assignees

  • 广州商研网络科技有限公司

Dates

Publication Date
20260508
Application Date
20230913

Claims (10)

  1. 1. The E-commerce customer service questioning and answering method is characterized by comprising the following steps of: Responding to a customer service enabling event, and acquiring a question text; carrying out semantic matching on the questioning text and text contents of a plurality of different structures of each preset description document, and determining a matching score corresponding to each text content; Determining a correlation score between the correspondence of each preset description document and the question text according to the matching score; screening at least one preset description document with the relevant scores meeting preset conditions as a candidate document, and determining a reply text according to the at least one candidate document by adopting a large language model; The method comprises the steps of obtaining a document title of a preset description document to serve as text content of a first structure, generating a document summary to serve as text content of a second structure according to the original text of the preset description document by adopting a preset abstract generation model, and segmenting the original text of the preset description document to obtain the text content of a third structure of each document fragment.
  2. 2. The e-commerce customer service question-answering method according to claim 1, wherein the semantic matching is performed on the question text and text contents of a plurality of different structures of each preset description document, and a matching score corresponding to each text content is determined, comprising the following steps: determining a questioning semantic vector according to the semantics of the questioning text by adopting a preset text vectorization model, and determining text semantic vectors corresponding to each text content according to the semantics of a plurality of text contents of different structures of each preset explanatory document; And determining the similarity between the question semantic vector and each text semantic vector as a matching score corresponding to each text content.
  3. 3. The e-commerce customer service question-answering method according to claim 1, wherein determining a correlation score between each of the preset description document correspondences and the question text according to the matching score comprises: and carrying out weighted summation on the matching scores corresponding to the text contents of each preset description document to obtain the correlation score corresponding to each preset description document.
  4. 4. The e-commerce customer service question-answering method according to claim 1, wherein the steps of segmenting an original text of the preset description document to obtain each document segment therein, include the steps of: Sliding the original text of the preset description document by adopting a sliding window with a preset window size to obtain a word by a preset step length, and obtaining a document fragment to be verified corresponding to each word; Determining the relativity between the document fragments to be checked according to the semantics of the document fragments to be checked; When the correlation degree meets the preset condition, confirming that the document fragment to be checked passes the check and taking the document fragment as the document fragment; and responding to a verification failure event, and iterating the process after adjusting the window size and/or the step length of the current sliding window until the sliding window passes the verification on the document fragment to be verified, which is obtained by sliding the original text by the step length and correspondingly extracting words.
  5. 5. The e-commerce customer service question-answering method according to claim 1, wherein the steps of segmenting an original text of the preset description document to obtain each document segment therein, include the steps of: Performing sentence granularity segmentation on the original text of the preset description document to obtain a corresponding sentence sequence; And determining whether an upper sentence relationship and a lower sentence relationship exist between every two adjacent clauses by adopting a text semantic model on the basis of semantics corresponding to every two adjacent clauses in the sentence sequence in sequence, cutting the every two adjacent clauses without the upper sentence relationship into segments, and determining corresponding document fragments.
  6. 6. The e-commerce customer service questioning and answering method according to claim 1, wherein after determining the reply text from the at least one candidate document using a large language model, comprising the steps of: checking the compliance and the authenticity of the reply text by adopting a preset reply check model, and determining the check passing rate of the reply text; And pushing the reply text to a user when the verification passing rate reaches a preset standard.
  7. 7. An e-commerce customer service answering apparatus, comprising: the event response module is used for responding to the customer service enabling event and acquiring a question text; The score determining module is used for carrying out semantic matching on the questioning text and text contents of a plurality of different structures of each preset description document, and determining a matching score corresponding to each text content; the score determining module is used for determining the correlation score between the corresponding preset description document and the question text according to the matching score; The reply determining module is used for screening at least one preset description document with the correlation score meeting a preset condition as a candidate document, and determining a reply text according to the at least one candidate document by adopting a large language model; The method comprises the steps of obtaining a document title of a preset description document to serve as text content of a first structure, generating a document summary to serve as text content of a second structure according to the original text of the preset description document by adopting a preset abstract generation model, and segmenting the original text of the preset description document to obtain the text content of a third structure of each document fragment.
  8. 8. The e-commerce customer service answering device according to claim 7, wherein the score determining module comprises a text vectorization sub-module for determining a questioning semantic vector according to the semantics of the questioning text by using a preset text vectorization model and determining a text semantic vector corresponding to each text content according to the semantics of a plurality of text contents of different structures of each preset explanatory document, and a similarity determining sub-module for determining the similarity between the questioning semantic vector and each text semantic vector as a matching score corresponding to each text content.
  9. 9. A computer device comprising a central processor and a memory, characterized in that the central processor is arranged to invoke a computer program stored in the memory for performing the steps of the method according to any of claims 1 to 6.
  10. 10. A computer-readable storage medium, characterized in that it stores in the form of computer-readable instructions a computer program implemented according to the method of any one of claims 1 to 6, which, when invoked by a computer, performs the steps comprised by the corresponding method.

Description

E-commerce customer service question and answer method and device, equipment and medium thereof Technical Field The present application relates to the field of electronic commerce information processing technologies, and in particular, to an electronic commerce customer service inquiry and answering method, and a corresponding apparatus, computer device, and computer readable storage medium thereof. Background Along with the rapid development of electronic commerce, merchant users and buyer users in an electronic commerce platform are rapidly increased, and accordingly questions of the users are also rapidly increased, so that intelligent customer service is developed for solving the questions of a plurality of users. The intelligent customer service can identify the questions of the user and give reasonable replies. In the conventional technology, in order to implement intelligent customer service, a search engine is generally adopted to search out a description document containing keywords from a plurality of description documents for replying to a user by taking all or part of the text in the question text of the user as keywords, however, in such an implementation manner, on one hand, a party needing to describe that the document contains the keywords can be searched out, so that the description document related to the question semantics but without keywords cannot be searched out, which results in inaccurate or incomplete reply, and on the other hand, the reply is easily affected by spelling errors, synonyms, ambiguities and grammar changes, and the reply is also inaccurate or incomplete. In view of the shortcomings of the conventional technology, the inventor conducts research in the related field for a long time, and develops a new way for solving the problem in the field of electronic commerce. Disclosure of Invention It is therefore a primary object of the present application to solve at least one of the above problems and provide an e-commerce customer service answering method, and corresponding apparatus, computer device, computer-readable storage medium. In order to meet the purposes of the application, the application adopts the following technical scheme: The application provides an e-commerce customer service inquiry and answering method which is suitable for one of the purposes of the application, and comprises the following steps: Responding to a customer service enabling event, and acquiring a question text; carrying out semantic matching on the questioning text and text contents of a plurality of different structures of each preset description document, and determining a matching score corresponding to each text content; Determining a correlation score between the correspondence of each preset description document and the question file according to the matching score; screening at least one preset description document with the correlation score meeting preset conditions as a candidate document, and determining a reply text according to the at least one candidate document by adopting a large language model. In a further embodiment, the semantic matching is performed on the question text and text contents of a plurality of different structures of each preset description document, and a matching score corresponding to each text content is determined, which includes the following steps: determining a questioning semantic vector according to the semantics of the questioning text by adopting a preset text vectorization model, and determining text semantic vectors corresponding to each text content according to the semantics of a plurality of text contents of different structures of each preset explanatory document; And determining the similarity between the question semantic vector and each text semantic vector as a matching score corresponding to each text content. In a further embodiment, determining a relevance score between each preset description document and the question according to the matching score includes: and carrying out weighted summation on the matching scores corresponding to the text contents of each preset description document to obtain the correlation score corresponding to each preset description document. In a further embodiment, before responding to the customer service enablement event, the method comprises the steps of: acquiring a document title of a preset description document as text content of a first structure; generating a document summary as text content of a second structure according to the original text of the preset description document by adopting a preset abstract generation model; Segmenting the original text of the preset description document to obtain each document fragment serving as text content of a third structure. In a further embodiment, segmenting an original text of the preset description document to obtain each document segment therein, including the following steps: Sliding the original text of the preset description document by adopting a slid