属性情感分类是社交媒体文本情感分析与挖掘领域的重要研究问题。现有方法主要利用文本内部特征构建分类模型，由于文本包含的语义信息有限，导致分类性能受限。考虑到社交媒体中的用户交互信息以及开放知识库的属性知识可以提供更多语义知识来源，本项目拟开展基于多源语义表示学习的社交媒体文本属性情感分类研究。具体包括：利用文本结构与情感信息进行注意力机制下的表示学习，研究基于上下文语境的文本情感语义表示方法；通过异质网络节点嵌入模拟面向用户、内容和属性，研究融合用户交互信息的文本情感语义表示方法；通过多源约束下的知识表示对属性进行情感推理与消歧，研究融合开放知识库的文本情感语义表示方法；应用端到端的学习机制，实现文本、用户与属性知识的统一语义表示，研究基于多源语义表示的属性情感分类方法。在此基础上，构建面向证券领域的情感分类示范应用。本研究将推动社交媒体文本情感分析的基础研究，也为社交媒体相关应用提供借鉴。

Aspect-based sentiment classification is an importance topic in sentiment analysis, opinion mining and social media analysis. Recently, researchers focus on text features and text classification models. However, the confined semantic information from the text limits the performance of classifier. In practice, the interactive information between users in social media and the knowledge of aspect from knowledge base will provide more semantic information. In this project, we aim to use multi-source representation learning to improve aspect-based sentiment analysis in social media. Firstly, we will utilize syntactic structure and sentiment information by attention mechanism to map the text into low dimensional vector space in order to represent the text based on the context. Then, we will introduce user interactive information into representation learning by heterogeneous network embedding. In the knowledge representation, we plan to use multi-source restricted knowledge representation to learn the representation of aspect and introduce the aspect level knowledge representation into sentiment analysis by attention mechanism. Finally, we will design an End-to-End architecture to map the text, user and aspect knowledge into an unified representation for sentiment analysis in social media. The result of our work will improve the research of sentiment analysis in social media, and it will be also implemented in practical applications such as financial prediction by sentiment analysis.