The Research And Design Of One Sentiment Analysis System Based On The Method Of Reflections Between Features And Polarities

With the growing availability of opinion-rich resources such as product reviews, one emerging research field related to opinion analysis and opinion mining is the study and summary of people’s important assessments on social issues or products from the rich sources of online opinion-based texts. This paper studies and designs a sentiment analysis system, it can identify opinions and polarities related to these opinions from raw textual reviews without explicit ratings, and can establish the reflections between opinions and polarities. The contents of this research have three major points as follows.Firstly, focus on the identification and extraction of features on the field of opinion analysis. Considering the attributes of features from the view of granule, features can be classified into two types, coarse features and fine features. In this section, the main attention is how to identify and extract the most representative coarse features of one object and select and filter the fine features according to these coarse features. In the research, the specific bootstrapping algorithms are utilized to study the terms and phrases related to every feature on one unlabelled textual dataset, these terms and phrases are used to predict the polarity of each feature. After analyzing and discussing the algorithm of ambiguity degree of terms related to features, which was proposed by the forerunners, and reckoning on existing problems, in this paper, a new algorithm of ambiguity degree is proposed. In the final part of this section, one formula of scores of the terms related to features is redefined.Secondly, since a single sentence or paragraph in real reviews often exhibits several aspects for user opinions, these sentences or paragraphs need the method of segmentations based on aspects. Otherwise, in the research of this paper, a circumstance is explored that between some different reviews the lengths have tremendous distances. Regarding these problems, a new segmentation model named LSMAS is proposed to address the challenge of identifying multiple single-aspect units in such sentences or paragraphs. It based on two types of segmentations and some formulas are modified or redefined.Thirdly, in this section, the research focuses on the sentiment analysis on reviews. The existing problem in general sentiment lexicon is inaccurate to identify the polarities of vocabularies or phrases which are depended on specific domains. In order to explain this difficulty, in this part one useful method is proposed that can establish a domain depended sentiment lexicon without the helps of other general sentiment lexicons. This method can solve the above problems reasonably. After that, utilizing the constructed sentiment lexicon, the features’polarities of "zones" can be analyzed and determined.(The "zones" are the units which are divided and produced by some methods of segmentations).Finally, in the section of evaluation, the experiments on the1000real reviews of enterprises’figures testify that the methods illustrated on above three sections in this paper can achieve a higher accuracy and have a better performance.