Research And Implementation Of Character A Recognization And Emotion Recognization In Chinese Novels

With the development of the mobile Internet,audio novels have increasingly become an important entertainment for people’s daily lives.Novel suppliers need to invest a lot of manpower and material resources in artificially synthesizing an audio novel.The automatic synthesis of audio novels has become a research hotspot in recent years.The key to the automatic synthesis of automatic audio novels is to recognize the characters in the dialogue of the characters in the novel and the emotions of the dialogues,and then use the gender,age,personality and other conditions of the characters to automatically synthesize the relevant voices in line with the character settings,and at the same time according to the character dialogue,the result of emotion recognition is used to set relevant dialogue emotions.At present,the research on novels mainly focuses on foreign languages,and there is very little research on Chinese novels.In view of the differences between Chinese and foreign languages,this article focuses on Chinese novel dialogue recognition(or named quote attribution)and novel dialogue emotion recognition research.The main contents and innovations of the research are as follows:1)Aiming at the problem of character recognition in novel dialogues,in order to fully express the difference in dialogue types and the semantic features before and after the text,a method for character recognition in Chinese novel dialogues based on Rule-BertAtten is proposed.First,the dialogue is divided into four main categories.According to different dialogue types,the Rule judgment and BertAtten methods are used respectively.Experiments show that this method has higher accuracy.2)Aiming at the lack of emotional recognition data set in Chinese novels dialogues,this article dynamically manually annotates the World of Plainness emotional recognition data set.The reference content includes the sentence types obtained based on regular and syntactic analysis,and the emotional tendency score of SnowNLP analysis,as well as the results of the NRC emotional dictionary scoring.After many considerations,the emotions of Chinese conversations are divided into none,like,disgust,anger,happiness,fear,sadness and surprise eight emotions.3)On the issue of emotion recognition in novel dialogues,in view of the imbalance of Chinese novel emotion data sets(there are more none emotions),a strategy of first two-classifications and then seven-classification is proposed.The data set is first divided into two parts according to the none emotion and other emotions,and then using SnowNLP’s emotional tendency and NRC’s emotional dictionary rules screens out some of the dialogues that are obviously not neutral,and then innovatively propose a dialogue emotional addition strategy for the other seven emotion dialogues,and finally seven-classification,summary of results.The experimental results show that the first two classification,then the rule screening,and then the emotional word addition,the final seven classification strategy has a higher accuracy rate than the direct eight classification.