基于语义分析的评价对象-情感词对抽取

发布时间：2018-10-12 15:41

【摘要】：评价对象-情感词对是情感词及其所修饰评价对象的组合,评价对象-情感词对的识别是细粒度情感分析的一个原子任务和关键任务.现有的研究大多集中在商品评论上,随着金融大数据的涌现,金融评论的情感挖掘意义凸显.与商品评论不同,中文金融评论中评价对象数目繁多且构成形式复杂,虚指评价对象和隐式评价对象也更常见;情感词的词性更丰富,其在句中的句法成分也更灵活、语义更丰富.针对金融评论的这些特点,该文提出了基于浅层语义与语法分析相结合的评价对象-情感词对抽取方法.考虑到金融评论多动词情感词,设计了语义角色标注与依存句法分析相结合的评价对象-情感词对抽取规则,解决了评价对象构成的复杂性问题;基于语义和领域知识对虚指评价对象进行了判别和替换,以明确其实际的指向和含义;基于特殊情感词搭配表、上下文搭配表及频繁搭配表提出了隐式评价对象识别的新思路,能有效地识别出缺省和隐含评价对象.在大规模的中文金融评论上进行了详细的实验测试,实验结果表明了该方法的有效性.
[Abstract]:Evaluation object-affective word pair is a combination of affective word and its modified object, and the recognition of evaluation-affective word pair is an atomic task and a key task in fine-grained emotional analysis. Most of the existing studies focus on commodity reviews. With the emergence of financial big data, the emotional mining significance of financial reviews is prominent. Different from commodity reviews, the number of evaluation objects in Chinese financial reviews is various and the forms of evaluation are complex, the virtual reference evaluation objects and implicit evaluation objects are more common, the part of speech of emotional words is more abundant, and the syntactic elements in sentences are more flexible. The semantics are richer. In view of these characteristics of financial reviews, this paper proposes an approach to extract affective word pairs, which is based on the combination of shallow semantic and grammatical analysis. Considering the multi-verb affective words in financial comment, an evaluation object-affective word pair extraction rule combining semantic role tagging and dependency syntactic analysis is designed, which solves the complexity problem of evaluation object constitution. Based on semantic and domain knowledge, virtual reference evaluation object is distinguished and replaced to clarify its actual direction and meaning. Based on special affective word collocation table, context collocation table and frequent collocation table, a new idea of implicit evaluation object recognition is proposed. Can effectively identify the default and implicit evaluation objects. The experimental results show the effectiveness of the method.
【作者单位】：江西财经大学信息管理学院;江西财经大学数据与知识工程江西省高校重点实验室;
【基金】：国家自然科学基金项目(61562032,61662027,61662032,61173146,61363039,61363010,61462037) 江西省自然科学基金重大项目(20152ACB20003) 江西省高等学校科技落地计划项目(KJLD12022,KJLD14035)资助~~
【分类号】：TP391.1

【相似文献】