Study Of Sentiment Orientation Analysis Based On Network Comments

With the rapid development of the Internet, communication platform such as forum, blog and others are emerging. There is a growing accustomed subjective remarks published on the Internet. They involve many aspects such as the policy, the hot events and products, which contains a huge amount of information, but in order to get people’s perspectives and attitudes from many characters, events, products in a short period of time is often very difficult. Therefore, it leads to an important research directions in the field of natural language processing-orientation analysis.The purpose of orientation analysis is to determine the attitude (or sentiment orientation) of the entire text by text mining and analysising the viewpoints, opinions, emotions, likes and dislikes, etc., of the subjective information in the text. A number of areas are related such as artificial intelligence, machine learning, data mining and natural language processing, it has attracted many researchers to study and research.In this paper, we focus on the network review text sentiment orientation analysis, the corpus source is build up with standard corpus and self-gathering network comments. Emphasis the access network reviews and manual annotation process. The experiment using three model approachs:hidden Markov model, sentiment lexicon and support vector machine to analysis its orientation:positive (support), negative (opposite) orientation.Firstly, we introduce the method of text classification, principle of hidden Markov models and support vector machine model in system. Later, we introduce the preprocessing of the corpus, including word segmentation and part of speech labeling, and went without the wording in details. Then we choose the appropriate corpus for training and learning, get the final model and calculate the sentiment orientation of the text in accordance with the three models. Based on the sentiment orientation, we have also focused on the establishment of a method based on the the HowNet basis sentiment lexicon thesaurus, and the establishment of effective rules to calculate the final sentence orientation, the accuracy rate is more than90%. Based on support vector machine model, we using TF-IDF algorithm to calculate text feature weights, and then select a certain characteristic value learning and training, at last cross-test the classification model, the accuracy rate has reached90%or more. Finally, we compare and analysis the results of three experiments deeply.Finally, this paper focuses on the field of hotel reviews of the qunar website and the analysis the sentiment orientation by classification of comments. We classify the comments into six different classes in evaluation of object categories and then do the orientation analysis for each class. The experiments show that the position and views of this classification orientation analysis is more accurate and detailed to response the attitude of customers. It can help managers to grasp customers’ favorite or aversion for all aspects of the hotel quickly, it has a practical significance.