Research On Choice Question Solving Method Based On Supervised Learning

The choice question is composed of a question sentence and several candidates,being similar to the multiple-choice question in the examination paper.This type of choice question belongs to the field of artificial intelligence,focuses on how to predict the correct answer from several candidates,and involves many techniques in different fields,such as natural language processing,information retrieval,and machine learning.In recent years,intelligent question answering technology has extensive application prospect in industry,and receives great attention in academia.This thesis regards the solution to the choice question as a ranking problem for candidate sentences.At first,our model extracts three kinds of features between the question and the candidate,i.e.statistical feature,relevance feature based on information retrieval model and semantic similarity feature.And then two kinds of supervised learning methods,i.e.Ranking SVM and logistic regression,are used to sort the candidates.Finally,the right answer is selected according to the sorting result of the candidates.In addition,our model analyzes the composition of the question sentence and the candidates,and puts forward the correction rules for the three types of questions,i.e.interrogative as subject,interrogative as object and interrogative as attribute,and then the results of Ranking SVM and logistic regression method have been improved according to the correction rule.The experimental results have been evaluated by MRR,MAP and accuracy,which are common metrics in the field of information retrieval.The experimental results show that the proposed method to choice question based on supervised learning is effective.