Research And Implementation Of Chinese Opinion Interpretation Classification Method Based On Attention Model

Opinion explanation classification is a basic task in the field of explanatory opinion mining,and has an extremely important impact on the subsequent research work.In this paper,we construct a large-scale,high-quality Chinese opinion explanation classification corpus of phone domain and hotel domain in Chinese product reviews.By analyzing the structural laws of opinion explanations in the corpus and the semantic logic relationship between opinions and opinion explanations,we explore different methods for explanations classifications.We will conduct research from the following three aspects:(1)Build corpus for explanation classification: according to the intrinsic characteristics of opinion explanation,we divide opinion explanations into three categories: reasons,suggestions and conditions.Based on this classification system,we construct a large-scale,high-quality Chinese opinion explanation classification corpus.Through the analysis of the corpus,we find that different opinion explanations have different expressions,and there is a potential connection between opinions and opinion explanations.(2)Chinese opinion explanation classification based on attention mechanism: according to the semantic relationship between opinions and opinion explanations,we attempt to apply Self Attention,Vanilla Attention and Contextualized Attention to the opinion interpretation classification task,and explore different attention methods to enhance the capability of classification.We use Focal Loss function to adjust the training direction of the model and alleviate the negative impact caused by class imbalance.To increase the capability of classification,we also try to utilize the Cw2 vec method to extract Chinese character stroke features for training and improve the quality of Chinese word representations.(3)Integrated system of opinion explanation classification: In this paper,we explore various attention mechanism to raise the capability of classification,and then trained a model which is better fit the opinion explanation classification.We applied this best model to build up an integrated system of opinion explanation classification,and this system can carry on semantic classification to the opinion explanation of phone domain and hotel domain.