Research On Sentiment Analysis Of Online Product Reviews Based On Implicit Features

In the era of big data,product reviews have become one of the most valuable data resources.Data mining in massive product reviews is one of most significant application in big data analysis.Product reviews contain the attitude towards the objects that the customers have evaluated,so mining the product reviews can develop commercial value and social value.In order to get more scientific and refined decision-making basis,the thesis combines existing research and relevant technologies to explores the research on sentiment analysis of online product reviews based on implicit features from the perspective of feature granularity.The mainly work is as follows:1.Aim at extraction of implicit features in the field of Chinese product reviews,a novel scheme based on hybrid rules is proposed.This scheme combines statistical rule,dependency parsing and conditional probability in order to taking the effect of adjective words,verbs and nouns into account.The proposed scheme is tested on a public cellphone reviews corpus,and results show that the hybrid rules are helpful to find implicit features in reviews effectively.2.Identifing the subjective review sentences about product feaures.A new emotional dictionary is constructed by integrating existing emotional dictionary with adjective words in corpus.Aim at the the subjective review sentences about explicit product feaures extraction,using the new emotional dictionary and a syntactic analysis tool to judge whether explicit features and emotional words occur together and exist dependence relationship.According to emotional words appearing in implicit sentences to extract subjective review sentences about implicit product feaures.3.Analying the sentiment orientation of subjective review sentences,and results are shown according to product feature clusters.The thesis proposes a new method based on the emotional dictionary and multiple classifiers combining.Firstly,using the method based on the emotional dictionary to calculate orientation value of each subjective review sentences and choose some sentences as the training data according to the value.This method can avoid to choose training data by manual work.Then,combining the self-training with ensemble learning and obtaining two different classifiers.The two classification results are used to judge the orientation of the rest sentences.The new method proposed in this thesis can effectively obtain the sentiment orientation of product features in product reviews and provide more detailed decision basis.Potential consumers can make more scientific and accurate purchasing decisions based on mining results,and manufacturers can also improve their products to meet consumer demands.