A Question—answering System Based On Sentence Similarity Comparison

Recently,Internet applications are growing rapidly.Algorithms and applications designed with artificial intelligence has contributed a lot to the Internet era in all over the world.A computer with Internet access should understand users' intention,speculate users' purposes,and provide services to them.The Internet has well knowledge of what they have,what they desire,and what they are used to do.Thus,it becomes more efficient and convenient for people to explore the answers or the knowledge of their questions.From traditional searching-engines,to question-answering communities later on,the way of knowledge acquisition has turned into intelligent question-answering system via human-computer interaction.The way of traditional searching-engines are based on keywords searching,which cannot comprehend users' purposes at all.Meanwhile,the results are displayed in web link-like summaries.These results always contain lots of useless information and lead to inefficiency.Though the manner of question-answering communities eliminates some semantic comprehension tasks,it brings timeliness troubles and need review and correct the answers manually.When it comes to the intelligent question-answering system,users' intentions can be acquired by artificial intelligence algorithms and the questions will be answered in time and precisely.In this paper,the FAQ question-answering system project based on Zhejiang Huixin Tech Company are designed and put into practice.We implement data cleansing on QQ group chatting record,extracting Q A pairs,concluding to extended and standard QA databases.As the data gets more,we carry out different features and machine learning methods on the task.Finally we train a question-answering model and provide RESTful API for web service and Wechat.Getting through the result and concurrence load test,the FAQ question-answering system can perform a good accuracy at 83.75%,which promotes a lot compared with the original one and are competitive to other question-answering products.