社会媒介倾向性检索旨在诸如博客、微博等Web 2.0 媒介上检索大众对热点话题的观点看法。社会媒介文本相对于传统文本具有文本短、表达不规范等特点，更重要的是，社会媒介以用户为单位组织文档，包含了大量的用户个性化信息和反映社会影响力的用户交互关系。目前倾向性检索研究尚未能结合社会媒介文本的上述诸多特点，使得检索性能大打折扣。因此，本项目拟研究融合用户信息的一体化社会媒介倾向性检索模型。具体内容包括：①抽取、量化用户显式和隐式交互关系，基于张量表示的方法对用户交互关系建模，采用张量分解方法挖掘用户潜在联系，用以度量社会影响力；②统计分析倾向用语风格、内容自相似度、用户活跃度等个性化特征，基于因子分析模型挖掘这些特征之间的内在联系，以整体度量用户个性化特征；③基于文本内容、社会影响力和个性化特征，设计融合用户信息的一体化倾向性检索新算法，减少缺乏考虑用户因素所带来的性能偏差，提高倾向性检索效果。

There is a growing research in opinion retrieval as more and more social media (such as Blog, Microblog, etc) are becoming an important platform to share opinions or comments. The goal of social media opinion retrieval is to find relevant and opinionate documents in social media according to a user's query. Furthermore, documents in social media is organized according to user such that there is a lot of user's personal characteristics information and user social relations reflecting user social influence. However, existing research fails to incorporate the user level information, which discount the effectiveness of opinion retrieval. In this project, our proposal presents a novel unified model for social media opinion retrieval which takes advantage of user social influence and user's personal characteristics.Specifically, the proposal constitutes by the following three parts: Firstly, user's explicit and implicit relations are exploited to analyze user social influence based on tensor-based approach; Tensor decomposition is used to discover latent factors to measure user social influence. Secondly, features are extracted for user's personal characteristics such as content self-similarity, user's opinion stylistics, user's activity; factor analysis model is employed to mining common factors reflecting the corrections of user's personal characteristics; Score functions are constructed to measure these user's personalities via these common factors. Finally, an unified opinion retrieval framework is introduced to leverage content, user social influence and user's personal characteristics to rank documents. The presented method would reduce retrieval bias caused by neglecting user information and thus improve the effect of opinion retrieval in social media.