Research On Intelligent Question Answering For Virtual Learning Environment

The rapid development of science and technology are leading the society into the era of big data.At the same time,people are also facing practical problems such as information overload,data explosion and information iteration update.The huge amount web page information of traditional search engine single retrieval feedback can no longer meet the needs of people to quickly obtain accurate information.As a hot spot of research in the field of artificial intelligence and natural language processing,intelligent question answering(QA)can provide users with efficient and accurate answers and intelligent personalized services.It is gradually applied to smart e-commerce,smart home,intelligent teaching,multimedia human-computer interaction and other commercial areas.Nevertheless,due to the diversity of semantics of natural sentences and the lack of restricted-domain QA datasets with research value,the intelligent QA research still faces quite a few bottlenecks.In this thesis,the open-domain English and restricted-domain Chinese intelligent QA algorithm is studied,and the model is tested and verified.Furthermore,based on artificial intelligence markup language(AIML)rules and Unity 3D platform,this paper designs and implements a teaching system of intelligent QA for virtual learning environment,which verifies and expands the practical application value of the intelligent QA model.The main content of research is as follows:1.Researching on the intelligence English QA based on the open-domain.Aiming at the shortcomings of the traditional model that can't capture the dependency of the question and answer statements,a semantic understanding model based on the stacked BiLSTM neural network and the collaborative attention(co-attention)mechanism is proposed.The stacked BiLSTM neural network is resort to acquire the dependency correlation between the words of the statement.The attention mechanism and the co-attention mechanism generate the affinity matrix to capture the interaction of the question and answer sentences,which is able to obtain the further feature vector representation of the statement.Select the intelligent optimization algorithm to perform parameter optimization.The algorithm comparison and the result analysis are carried out on TREC8-13,Wiki-QA and SemEval2015 public QA datasets.The experiment proves that the model has high accuracy and robustness in sentence semantic expression.2.Studying on the intelligent Chinese QA based on restricted-domains.A restricted-domain QA dataset(IPS-VLQA)based on the inverted pendulum system experiment is constructed for the current situation of the lack of Chinese dataset in the restricted-domain.The jieba word segmentation toolkit is utilized to segment Chinese words and performed part-of-speech tagging to generate standard datasets.Based on the work of content 1,a layer of CNN is added before the feature extraction layer of the model,and the impacts of each word on the entire statement is calculated by the convolution layer.This model was verified and analyzed in the CCKS2018 and IPS-VLQA datasets.3.Designing an intelligent QA system used for the virtual learning environments.Based on the Unity3 D platform to build a virtual learning environment,design models of virtual role such as virtual teachers and virtual students.And then,sort out IPS-VLQA data sets based on AIML.The model of content 2 and the AIML are simultaneously utilized as the intelligent question answering engine to drive the role in the virtual learning environment,the intelligent question answering system for the virtual learning environment is constructed to realize the question and the answer demonstration.