Scientific Development Research

Dialogue intent recognition is one of the research hotspots in question answering systems, and intent recognition is the premise for question answering systems to give answers. This paper firstly builds the Fasttext intent recognition model, optimizes the model, and conducts multi-classification to study the effect of the algorithm. In order to reduce short text, the word repetition rate is low, the word co-occurrence is not suitable, and the keyword extraction problem of the document set with low language norm is used in the input stage of the Fasttext model to use the TextRank algorithm and the TF-IDF algorithm to extract the key. The clause is input to the training model, and TF-IDF is used to extract the keywords of the text as a feature supplement to improve the effect of text classification. Get the user question, input it into the Fasttext model for intent classification, and return the final answer to the user.

TextRank; Fasttext; Convolutional Neural Network Method