Term Weighting Scheme–Based Sentimental Analysis

In recent years,sentiment analysis has been the subject of growing interest on the part of the community of natural language processing researchers.Sentiment analysis can help people in companies and public administrations to learn more about the opinions of customers,and to assist them make some decisions.In this paper,we firstly introduce the background,the definition of the problem that which will allow a better understanding of the proposed objectives of the research and the contribution of the thesis.We also give some introductions of recent text classification methods,such as Probabilistic Algorithms(Naive Bayes),Algorithm of the Nearest Neighbor and Variants,Decision Trees or Classification and Vector Support Machines.Then we introduce the steps of constructing a sentimental analysis system,including preprocessing,feature extraction and performance evaluation.Finally,we pay more attention to a dataset which is made up of online hotel reviews,and apply a supervised machine learning approach Na?ve Bayes using unigram feature with two types of information(frequency and TF-IDF)to realize polarity classification of documents.As shown in our experimental results,in terms of accuracy,precision,recall and F_score,our model outperforms others.