Research And Realization Of Common Refractive Method For Product Reviews

Coreference is a widespread phenomenon in Natural language. Coreference resolution is a natural language processing task that combinedifferent expressions ina discourse that refer to the same entity using context and expressions’ characters. With the rapid development of network media recent years, product reviews, as the important data which users express their opinion on products has been studied by many researchers. While, in many product reviews, people always express one product aspect with different words and noun phrases.This phenomenon common makes product attributes description is too many, trivial, and is not conducive to computer analyzing. Therefore, product reviews coreference resolution is one of the important issues in Opinion Mining.According to the characteristics of Chinese language product reviews, this paper discuss about Chinese product attributes Coreference resolution problem and build our System using two methods. Specifically, our research concerns the following three aspects:(1). Aiming at Chinese product reviews Coreference resolution problem, this paper analysis the characteristic of Chinese opinion paraphrase and discussed the associated of cor-refered product aspect expressions. And we also determined the definition of similarity calculation methods from literal, semantic and context of product aspects. This lay a good foundation for the Coreference resolution system.(2). This paper built a binary classifier for product aspects Coreference resolution based on supervised machine learning method. And discuss the specific details of Mention detection, classification decision and linking process. The experimental results show that the binary classification framework in with literal, semantic and context features are quite effective for our task.(3). This paper also solve the problem using Clustering method. Hierarchical clustering and K-Means algorithm are used in our system. And we combined with the density and maximum minimum principle to solve the initial clustering center selection problem. The experimental results show that this principle can effectively improve the Coreference resolution performance on Clustering method.