Research On Product Evaluation Classification And Recommendation Algorithm Based On Chinese Micro-blogging

In recent years, micro-blogging is an emerging internet media platform, and its features are short content, fast spread, a large number of users and so on. Sentiment analysis on micro-blogging is an important and meaningful part of this field. Users may want to get information from micro-blogging, when they do shopping or other behaviors online. Aiming at text chara-cteristics like irregular format, network language, omitted component, and classification prob-lems like scarce labeled data, difficult manual annotation existed in product evaluation data mining on Chinese micro-blogging, this paper mainly carries out the following research work.According to text characteristics, proposes a method for constructing emotion evaluation unit sets. This method respectively constructs three dictionaries of evaluation words, adverbs, and evaluation object words, and formulates some rules for component complement and unit construction. It not only ensures a better accuracy and comprehensiveness of information ext-raction, but makes an attempt to streamline word sets and improve efficiency. The experiment results show that it is better than the method based on syntactic path.According to classification problems, proposes an classification algorithm based on semi-supervised learning on graph, called LP-SVM, which combines the label propagation process and support vector machine. The algorithm not only realizes classification of a small amount of labeled samples, but avoids producing no classifier and retraining for a new data, which is caused by semi-supervised learning on graph. Using this algorithm, the paper makes feature extraction and classification for emotion evaluation units. Experiment results show that this algorithm is better than traditional and transductive support vector machine.Combining with practical application, this paper also proposes a product recommendation algorithm based on evaluation classification. The algorithm combines the results of evaluation classification and text features of Chinese micro-blogging, and formulates the product recom-mendation indexes and its calculation method. The product recommendation scheme obtained by experiment is basically same as a professional website, which fully verifies the accuracy of this algorithm.