Research On Calculation Method Of Text Similarity Based On Deep Learning In Intelligent Question Answering System

With the rapid development of information technology,smart community,by making full use of the integrated application of technologies such as the Internet and the Internet of Things,has become a new mode of social management innovation in the new situation.Among them,property consulting is an indispensable part of property management.And based on the property consulting,the intelligent question answering system can solve many problems,such as unskilled staff,limited working hours and energy,achieve real-time consultation and reduce time and labor costs.In this paper,based on the question matching problem in the question answering system,we have done a deep research on the method of sentence similarity calculation by using natural language processing technology.This research mainly focuses on the similarity calculation method of short sentences and long sentences,which is based on the deep learning.The specific contributions are as follows:(1)Because the traditional method of sentence similarity calculation,which is based on statistical model,can only consider the surface features,such as word frequency and sentence length,and ignore the semantic and syntactic structure of sentences.A sentence similarity computing method based on Word2vec and dependency syntax is proposed.Based on the community consultation problem database,the method uses word vector model to map words into vector space of fixed dimension,fully expresses lexical semantic information,and the processing of the sentence is simplified as a vector in a vector space.And then this method acquires syntactic structure information from dependency parser,and fully considers the syntactic relations between words.The experiment shows that the method can deeply understand the semantic information of the question and improve the accuracy of the question matching.(2)The sentence features extracted by the original siamese network model contain limited information,and the spatial distance used to calculate the similarity of two sentences,ignores the semantic information of the two sentences.Aiming at this problem,a Siamese network model based on improved features is proposed for similarity calculation.This model combines attention mechanism,feature aggregation,and word-char vector to operate the features extracted by the neural network,and excavate deeper sentence features.The algorithm is tested on the open corpus,and it can get better results than the original siamese network.(3)In order to achieve the complementary advantages of multiple features,a siamese network model based on multi-feature fusion is proposed for similarity calculation.This paper proposes an improved siamese network model based on cascade feature fusion and interactive feature fusion to effectively fuse the improved features.The algorithm test is carried out on the open corpus,which shows that the siamese network model based on multi-feature fusion is better than the single-feature model.(4)Based on the research of the method of statement similarity calculation about deep learning,this paper designs a simple question answering system,includeing client development,communication protocol development,server-side development,database design and other content.