面向散文阅读理解的文本情感分析关键技术研究

作为体现认知智能的典型任务，机器阅读理解要求机器根据给定的文本内容，对相关的问题做出正确回答。这就需要机器具有语义分析能力，甚至是情感语义分析能力。本项目针对散文阅读理解中的情感分析任务，聚焦情感表达的隐式性、情感承载意象的多样性、情感变化原因的复杂性等语言特点，综合运用表示学习、图神经网络、认知计算和知识图谱等技术，采用信息融合、嵌入表示、联合训练、注意力等策略和机制，重点开展修辞格识别与情绪分类、隐喻识别与情绪分析、意象识别与思想情感分析、情感变化及原因发现等方面的研究，并开发一个面向散文赏析类问题的解答系统。本项目将拓宽文本情感分析研究的关注领域，丰富和深化文本情感分析的方法和技术，对提升智能阅读机器人的认知能力，助力智慧教育具有重要应用价值。

As one of the typical tasks that embody cognitive intelligence, machine reading comprehension ask the machine to provide the correct answers of the relevant questions according to the given text contents. This requires the machine to have the ability of semantic analysis, even the ability of emotional semantic analysis. . For the emotion analysis task for prose reading comprehension, focusing on the language characteristics of proses such as the implicitness of emotional expression, the diversity of emotional imagery and the complexity of the causes of emotional change, this project emphatically carries out the researches on the identification and emotion classification of figures of speech, the identification and emotion analysis of metaphor, the identification of imagery and thought emotion analysis, and the discovery of emotion change and its cause by synthetically utilizing various kinds of technologies such as representation learning, graph neural network, cognitive computing and knowledge graph, with the help of the strategies such as information fusion, embedding presentation, joint training, and attention mechanism. And a solution system for the questions of prose appreciation is finally developed. . This project will broaden the fields focused on by the researches of text emotion analysis, enrich and deepen the methods and technologies of text emotion analysis. And it has the significant application value for promoting the cognitive ability of intelligent reading robots, and for facilitating intelligent education.