A Linear-Chain CRF-Based Learning Approach for Web Opinion Mining

The task of opinion mining from product reviews is to extract the product entities, opinions on the entities and determine whether the opinions are positive, negative or neutral. Reasonable performance on this task has been achieved by employing rule-based, statistical approaches or generative learning models such as hidden Markov model (HMMs). In this paper, we proposed a discriminative model using linear-chain Conditional random field (CRFs) for opinion mining and extraction. CRFs can naturally incorporate arbitrary, non-independent features of the input without making conditional independence assumptions among the features. This can be particularly important for opinion mining on product reviews. We evaluate our approach base on three criteria: recall, precision and F-score of extracted entities, opinions and their polarity. Compared to other methods, our approach is more effective for opinion mining tasks.