Fine grained sentiment analysis based on Bert

Abstract

The fine-grained sentiment mining based on online review data is helpful to analyze user pain points and improve user experience and business marketing. A fine-grained sentiment analysis model based on Bert is proposed. Commodity feature words and emotion words are extracted by dependency parsing; the polarity of emotion was analyzed by the combination of Bert and emotional words; through the corresponding emotional polarity of feature words, we can get the emotional tendency of commodity features. The experimental results in movie review data set and film review data set show that, the accuracy and F1 value of Bert with feature words and emotional words were 94.67% and 94.55%, which were better than other models.