为了使个性化虚拟人更加形象生动,能根据用户输入的文本做出表情动作,运用自然语言处理技术对中文和英文文本进行语义和分类处理,分析出动作和情感信息。采用潜在语义方法从文本中提取出动作语义信息,利用hownet计算词汇相似度,使用K最近邻方法将文本情感信息分为6类:愤怒、厌恶、恐惧、喜悦、悲伤和惊讶。实验结果为:语料文本分类准确率为87.5%,系统能从用户输入的文本中提取出情感、动作信息,使虚拟人做出相应表情变化。%To make individual avatars more vivid and the animations of avatars can be changed according to the input text,our researches focus on natural language processing technology and achieve the Chinese and English text semantic analysis and sentiment classification.The system uses latent semantic analysis to extract action semantic information,uses hownet to calculate word similarity,and uses K-nearest neighborhood to divide the text into six categories: anger,disgust,fear,joy,sadness and surprise.Experimental results show that the accuracy rate of text corpus classification is 87.5%.Finally,the system can extract valid emotion and action information from the input text,and the individual avatars changed by the relevant emotion results.
展开▼