Design And Implementation Of Retrieval Question Answering System Based On Intelligent Recommendation

With the rapid development of the Internet,there are more and more Internet information.As the main way for people to obtain Internet information,search engine plays an increasingly important role in daily life.However,in the process of retrieval,there are some problems:it is often difficult for users to give concise and accurate queries directly according to their search intention,which leads to the search engine can not provide the most suitable results for users;In addition,search engines provide users with web pages containing a large amount of spam information,allowing users to obtain results after a long time of reading,which makes the retrieval process inefficient and the experience worse.This paper proposes a retrieval question answering system based on user behavior intelligent suggestion method,which improves the user's retrieval efficiency and optimizes the user's retrieval experience.This paper mainly completes the following three parts:1.This paper proposes a query recommendation model based on click model and network embedding.We propose a feedback construction method based on query timing and user's query perception attention to model the behavior of multiple queries in a session,so as to capture the most interested documents;we use word embedding and graph embedding to model the query reformulation relationship between adjacent queries,enhance the relevance of user's query,improve the interpretability of query expression,and improve the accuracy of query suggestion model.2.We use large-scale pre-training model to fine tune in the downstream reading comprehension task.By improving the pre-training task,the model parameters can be reduced without damaging the effect of the model;combined with the technology of named entity recognition and part of speech tagging,according to the semantic,part of speech,category and other information of the query words,a question expression conforming to the habit of question answering is constructed,so that users can get accurate and concise answers directly from the short query.3.Based on the query suggestion model and reading comprehension model proposed in this paper,we use Elastic Search open-source search engine to build a retrieval question answering system to simulate the whole process of user retrieval.The model proposed in this paper has achieved significant performance improvement on the public data in their respective fields.Through the design of model comparison and ablation experiments,the effectiveness and feasibility of the model proposed in this paper are verified.