A Novel Enterprise Reputation Mining Approach Based on Sentiment Analysis Technology

With the rapid growth of user-supplied content in the Internet, sentiment analysis for online users' reviews has attracted more and more attentions. In this paper, we aim to mining enterprise reputation by the sentiment analysis technology on the microblogging platform. Framework of the enterprise reputation mining system is presented in advance. In this framework, the original documents are pre-processed by four steps: (1) Stop words removing, (2) Stemming, (3) Converting the document to vector space, and (4) Term weighting. Then, using the sentiment words base and psychological model base, we propose an improved LDA topic model to mine sentiments from users' reviews. The LDA topic model is a structure with multi-layers, including: (1) document collection layer, (2) document layer, and (3) word layer. To solve the sentiment analysis problem, three types of latent variables are considered in the improved LDA model, which are (1) distribution of the joint sentiment and topic-document, (2) distribution of the joint sentiment and topic-word, and (3) distribution of the sentiment and document. With the given modified LDA topic model, the enterprise reputation mining task is transformed to a three points classification problem, and "Positive," "Negative," and "Neutral" are used as the sentiment analyzing results. In the experiment, we select six enterprises which produce electronic products to construct dataset, and the experimental data are collected from Twitter. Compared with other methods, the proposed algorithm can effectively mine the enterprise reputation with high accuracy.