Abstract: With the explosive growth of online social networks, Sina Weibo has become an important platform for people to express their opinions and emotion. Weibo not only reflects the opinions of the users, but also conveys   the opinions to other users, which can affect the opinions of other users. However, the typical features of Weibo such as short and flexiblity, bring some new challenges for opinion mining. When traditional approaches focusing on building a sophisticated feature space for text are used to analyze Weibo, the performance is usually poor. To solve this problem, this paper proposes an opinion mining method based on graph ranking for Sina Weibo. Firstly, utilize Boolean model to represent Weibo, and assign a pseudo label for each Weibo by logistic regression. Then, construct Weibo relationship graph based on context, and take advantage of restricted Boltzmann machine to extract high dimensional features. Finally, mine opinion for Weibo based on graph ranking. The experimental results show that the proposed method can mine opinion for Weibo effectively.

Key words: Weibo, opinion mining, graph ranking, restricted Boltzmann machine

摘要： 随着在线社交网络的爆炸式增长，微博已成为人们发表观点和表达情绪的重要平台。微博不仅可以反映用户的观点，还可以通过转发方式等传递观点，进而影响其他用户的观点。然而，微博以其简短、口语化等特点，给识别观点带来了新的挑战。仅仅基于文本进行观点分析的传统方法在分析微博观点倾向时，效果并不理想。为解决此问题，提出了一种基于图排序模型的微博观点信息识别算法。首先，利用布尔模型表示微博文本，并用逻辑回归进行观点分析获得伪标签；然后，利用上下文关系构建微博关系图，并利用受限玻尔兹曼机抽取高维特征；最后，基于图排序模型识别微博观点信息。实验结果表明，该算法能有效地对微博观点进行识别。

关键词: 微博, 观点识别, 图排序, 受限玻尔兹曼机