Object And Sentiment Analysis Of Texts Based On Probabilistic Graphical Model

With the development of Web2.0 and e-commerce, more and more users express their views on a variety of products and services in blogs and forums. Therefore, a large number of comments on products and services by users have arisen on the Internet. In the face of these reviews, users need to find suitable information to guide their consumption behavior. Besides, product and service providers also need to analysis the views and comments of users to find consumers’demands, thereby improving the products and services, to seek greater commercial profits. However, discovering information through manual methods of a large volume of textual data is extremely difficult. Hence, there is an urgent need to study advanced technology for detecting and processing these comments rapidly.Online review usually relates to the specific object. This paper focuses on studying object sentiment classification issues based on automobile comments. The main contents are as follows:(1) Establishing database and constructing modelThis paper establishes the comment text database by obtaining, analysis and sorting relevant comment text that contains the object name. Based on the theory of probabilistic graphical models, the Object and Topic Sentiment Unification model (OTSU model) was constructed corresponding to the text generation process "objectâ†'topic and sentiment". Besides, the Unsupervised Object and Sentiment Unification model (UOSU model) was constructed corresponding to the text generation process "objectâ†'sentimentâ†'topic".(2) Object and sentiment classification of texts based on the OTSU modelOTSU model detects object and topic-sentiment from texts simultaneously. Using the OTSU model can achieve words generated by the specific object and topic-sentiment. Besides, the object distribution and topic-sentiment distribution for a text can be obtained at the same time. And the sentiment of text’s object can be obtained through the topic-sentiment distribution. This paper based on the OTSU model, object sentiment classification was conducted for texts with single object and multiple objects respectively. The OTSU model has been evaluated on the automobile review dataset. The experimental results show that the words for specific topic-sentiment discovered by OTSU model are indeed coherent and informative. In addition, for object sentiment classification of texts with single object and multiple objects, the accuracy, recall rate, F-measure of OTSU model reached 70.63%,70.78%,70.57% and 70.94%,71.09%, 70.89% respectively. OTSU model can extract the objects and the corresponding sentiment polarity from texts simultaneously, which benefit people analysis and application of the relevant data.(3) Object and sentiment classification of texts based on the UOSU modelUOSU model detects object, sentiment and topic from texts simultaneously. Using the UOSU model can achieve words generated by the specific object, sentiment and topic. Besides, the object and sentiment distribution for a text can be obtained at the same time. Unlike to the OTSU model, the sentiment of text’s object obtained through the topic-sentiment distribution by UOSU model. Similarly, object sentiment classification was conducted for texts with single object and multiple objects respectively for the UOSU model. The experimental results show that the words for specific topic discovered by UOSU model are indeed coherent and informative. In addition, for object sentiment classification of texts with single object and multiple objects, the accuracy, recall rate, F-measure of UOSU model reached 74.19%,73.97%,74.06% and 73.53%,73.50%,72.97% respectively.