汉语解释性意见挖掘关键技术研究

Recent years have seen many advances made in the area of opinion mining. However, most current opinion mining systems are only concerned with mining positive or negative opinions from opinionated texts, and ignore the detailed explanation of the underlying facts, reasons, conditions and/or suggestions for the opinions. Actually, such explanatory information is extremely important for decision-making and information service based on opinion mining. To overcome this challenge, this proposal will focus on developing innovative technologies for explanatory opinion mining from Chinese online reviews. The major content of this research includes: (1) First, we will investigate the internal structures of Chinese opinion explanatory expressions in different domains and the correlation mechanism between opinions and their explanations, and thus construct classification systems for opinion explanations and explanatory opinion relationships, respectively. Furthermore, we will build an annotated corpus of Chinese explanatory opinions after drafting a scheme for explanatory opinion annotation. (2) Based on the theoretical analysis and the annotated corpus, we will continue to explore the key issues in Chinese explanatory opinion mining, such as explanatory opinion relation recognition, opinion explanatory paraphrase recognition and opinion explanatory sentence summarization, and thus develop methods and technologies for Chinese explanatory opinion extraction, aggregation and fusion. (3) Finally, we will exploit the above innovative methods and technologies to construct an aspect-based multi-mode abstraction framework and system for Chinese explanatory opinion summarization, and further assess the effectiveness of the proposed methods. The implementation of this project will not only expand the research area of opinion mining, but also facilitate more accurate decision-making with respect to a set of given opinions, and thus have prospects for applications in opinion question and answering, intelligent customer services, recommendation dialog systems and business intelligence.

意见挖掘研究近年来取得很大进展，但现有意见挖掘系统大多只关注褒贬意见而忽视其背后的事实、原因、条件或建议等意见解释信息,而这些信息对基于意见挖掘的决策和信息服务极为重要。为此，本项目拟对汉语解释性意见挖掘展开研究，即研究产生褒贬意见的原因、条件等意见解释信息，主要研究内容包括：(1)研究意见解释表达的的基本理论及其语言学表示特点,建立意见解释语义与关系分类体系，制定标注规范，构建相应语料；(2)以上述理论分析及所构建的语料为基础，研究解释性意见关系识别、意见解释复述识别和意见解释句子摘要生成等关键问题及相应的意见解释抽取、聚集与融合方法；(3)基于上述方法构建解释性意见摘要框架，实现一个多文档汉语解释性意见摘要系统，并验证所提出方法的有效性。本项目的实施不仅可以拓展意见挖掘的研究领域，而且可以为用户的精准决策提供意见解释信息支持，在意见问答、智能客服和推荐对话等领域具有广阔的应用前景。

意见挖掘研究近年来取得很大进展，但现有意见挖掘系统大多只关注褒贬意见而忽视其背后的事实、原因、条件或建议等意见解释信息,而这些信息对基于意见挖掘的决策和信息服务极为重要。本项目以研究汉语解释性意见挖掘关键技术为目标，重点探索了意见解释检测与分类、解释性意见要素识别、解释性意见抽取、意见解释聚合与融合等关键问题，并取得了以下研究进展：(1) 本项目构建了一个涵盖酒店服务、手机产品和金融等领域六万个评论的包括解释性意见要素及其关系、意见解释类别和多粒度情感极性等标注信息的多层次解释意见标注语料库；(2) 针对汉语意见解释特点，提出面向意见文本的词法和依存句法分析、多粒度的意见解释识别、基于序列标注的解释性意见要素识别、多特征融合的意见解释分类、基于关系识别的解释性意见信息抽取以及意见驱动的意见解释聚集与融合等一系列解释性意见挖掘关键技术；(3) 形成一个完整的基于意见文本解释性意见信息结构意见解释摘要的解释性意见挖掘技术框架，构建一个面向产品评论的汉语解释性意见挖掘平台，并实验验证了方法的有效性。此外，项目组共发表学术论文21篇，培养硕士研究生10人。本项目研究成果为用户的精准决策提供意见解释信息支持，可广泛应用于智能客服、产品推荐和精准产品设计等相关领域。