Research On The Opinion Question Oriented Answer Summarization Techniques In Cqa Services

With the development of CQA, the scale of question-answering resourceprovided by users grows rapidly, which does much favor to people who have similarquestions as those in CQA. A number of questions in the CQA corpus can becategorized as the opinion choosing questions, the askers of which aim to get theopinions about their own actions or ideas from the others. The problem for users toread answers of the questions is that there are too many answers provided by usersto have an overview of those efficiently. To solve this problem, we have research onthe following aspects:First, we treat the answers of the opinion choosing question as the texts whichcontain sentiment polarity, and use both supervised and unsupervised methods todecide the polarity of answers. By computing the sentiment orientation of wordswith WordNet, we calculate the polarity of answers with unsupervised learningmethod. As to supervised methods, we use the Na ve Bayes model, SVM model andthe maximum entropy model to classify answers into different polarities and use theensemble method to improve the performance of classification. Finally, theexperiments are conducted to test the effectiveness of the methods mentionedabove.Second, we have studied on summarizing answers in sentence level with amethod based on the MMR model. The feature sparse problem of short text issolved by expending word sets of answers with WordNet. After that, we present amethod which combines MMR model with sentiment polarity information whichcontains three improved MMR models by introducing sentiment information ofanswer into summarization. Then we give experiments to test the effectiveness ofthe new model with the ROUGE package. Experiments show the performance ofMMR model has obtained an improvement by combining sentiment informationwith it, and computing similarity between sentences more precisely with sentimentinformation makes the most significant improvement of model performance amongall methods.Finally, we build an opinion searching system based on the answer summarization resource of CQA. The system summarizes answers which is relatedto the user‘s query based on the resource of Yahoo! Answer, and presents them in afavorable way.