Research Of Sentiment Tendency Analysis For Goods Evaluation Based On Text Classification

In recent years, the booming e-commerce makes more and more people prefer on-line shopping. On the one hand, in order to improve customer satisfaction, online mer-chants usually allow customers to evaluate the purchased goods, resulting in the rapidgrowth of the number of goods evaluation. On the other hand, due to the limitations ofonline shopping itself, likely to cause to distinguish the quality of goods difficultly, thereality does not match the goods description information and other defects. Therefore, inorder to decide whether or not to buy, customers have to look at past reviews in a largenumber of goods evaluations and know about the others’ opinions of goods and services.At the same time the merchants can improve product quality and enhance competitive-ness through customer feedback. Therefore, data analysis techniques which in order toeffectively obtain evaluation information——sentiment analysis has been paid moreand more attention of scholars.The main contents of sentiment analysis include text subjective and objective con-tent identification, emotional intensity calculation, emotional tendency classificationand so on. It is mainly based on text mining and data mining, while integrate into thetechnology of text understanding. Among them, the emotional tendency classification isthe focus of this study, its main goal is to classify the positive or negative text emotionand it can be seen as a special text classification problem.This paper first introduces the relevant background and research status of senti-ment analysis, then describes in detail several classical feature selection methods andtext classification algorithms. By summing up the existing methods and improve theclassification accuracy and speed, this research proposed the feature selection algorithmfor sentiment tendency——matrix projection (MP, for short) and the classification al-gorithm for emotional tendency——normalized vector(NLV, for short) to realize sen-timent analysis of goods evaluation.The feature selection method based on the matrix projection considers the termdocument frequency and the average frequency occurs in all texts. Based on a variety ofclassification algorithms, the contrast of several other typical feature selection algo-rithms such as document frequency(DF),information gain(IG),chi-square parity(CHI)shows the MP feature selection method is effective.On the one hand, the classification algorithm based on normalized vector com-pressed the text vector space into normalized feature vectors. On the other hand, it re- duced the feature weights between high frequency words and low frequency words bynormalized function, and strengthen the ability of classification of low frequency words.This paper compared with three classification algorithms(KNN, NaiveBayes, SVM) onthe real goods evaluation data set, the results show that the classification algorithmbased on normalized vector has high classification accuracy and speed. The algorithm interms of classification accuracy compared with the KNN method has obviousadvantages, although slightly lower than SVM, but can predict emotional tendencies ofgoods evaluations faster.