Intelligent response method and system based on context dialogue semantic recognition

[0025] Example:

[0026] see figure 1 , the intelligent answering system of the present invention comprises mobile equipment, telephone routing, voice-to-text module, answering module and customer service seat display, wherein mobile equipment, telephone routing and voice-to-text module are connected successively, and answering module is connected with mobile equipment, voice to text module respectively , Seat monitor connection.

[0027] There are two main ways for customers to access the intelligent answering system. One is to dial the customer service phone through the mobile device, and convert the text into the answering module through the phone routing and voice-to-text module. The other is to use the APP of the mobile device, online answering The text is directly input to the answering module by means of pages, etc. see figure 2 , the present invention adds context-based dialogue semantics recognition function to the processing method of the traditional intelligent answering system. To context-based dialogue semantic recognition. Specifically, semantic recognition models the dialogue structure of the context generated by the current question, analyzes the context information when the user asks the current question through machine learning methods, and identifies the context of the current question by analyzing the sentence composition structure of the current question. The missing sentence components are combined with the context for topic analysis and semantic analysis to expand and understand the semantic information of the current question, and then retrieve the answer that matches the semantics of the current question from the system. After the customer service agent checks the corresponding answer through the monitor, he replies to the customer through the phone.

[0028] The present invention is based on the intelligent response method of context dialog semantic recognition, and the process is as follows:

[0029] 1. Problem analysis, formatting and natural language processing of the questions entered by customers.

[0030] The purpose of question analysis is to analyze the semantic type of the question and determine the intention of the user to ask the question. The key step is to extract the keywords. For example, if the user wants to ask "I want to ask about the tariff introduction of the 4G Feixiang package", the content must be formatted first, such as converting English characters into uppercase 4g into 4G, replacing typos, etc. After formatting , start text segmentation, and then extract keywords after the word segmentation is completed. For example, in the current example, what the customer wants to ask is tariff introduction, and the keywords are "4G Feixiang package", "rate" and "introduction". Words such as "I think" and "ask me" are used as secondary words or directly filtered. At the same time, part-of-speech analysis is performed on the extracted keywords, and then the processed results are transmitted to the next process for processing.

[0031] 2. Information retrieval, based on the enterprise knowledge base, using the recommendation algorithm to retrieve knowledge points.

[0032] Information retrieval is to use the corresponding recommendation matching algorithm from the enterprise knowledge base according to the results of problem analysis to match the answers, and then transfer the corresponding hit knowledge point records to the fourth step according to the matching degree. If there is no knowledge with a high matching degree Click to skip to the third step.

[0033] The recommendation matching algorithm will analyze the grammar and syntax of the input results of the problem analysis, and then sort the knowledge points in the knowledge base according to the matching degree according to the keywords and syntactic structure of the query statement, and finally return the highest matching degree according to the results record of knowledge points.

[0034] 3. Semantic recognition, which is based on contextual semantic recognition of conversations.

[0035] When there is no knowledge point record with high matching degree only based on the content of the currently input consulting question, start to analyze the structure of the current consulting question and identify the semantics of the question in combination with the context content, so as to accurately identify the answer to the question.

[0036] 1) Structural analysis

[0037] First, analyze the structure of the current consulting question, obtain the composition of the question through the sentence composition analysis model, and then judge what the default composition of the question is, and obtain the composition of the question and the record of the default composition.

[0038] 2) Semantic recognition

[0039] Through the analysis of people's spoken language habits, it is found that the default question components usually exist in the context of the dialogue, then through the analysis of the context components, combined with the sentence components of the query questions, the user's semantic recognition is jointly carried out, and finally the sentence's The basic ingredients complement and complete.

[0040] 3) Problem reconstruction

[0041] By inputting the components of the consulting question, the question is recombined and generated into a sentence expression that the system can explain.

[0042] 4) Information retrieval

[0043] For the regrouped questions, re-retrieve the information, and output the search results after completion for the customer service personnel to choose.

[0044] 4. Answer extraction, extracting the content of knowledge points and returning them in a colloquial way.