Social Summarization via Automatically Discovered Social Context

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 483–490, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Social Summarization via Automatically Discovered Social Context Po Hu1 2, Cheng Sun1, Longfei Wu1, Donghong Ji1 and Chong Teng1 1Computer School, Wuhan University, China 2Department of Computer Science, Huazhong Normal University, China phu@mail.ccnu.edu.cn, gensun.cc@gmail.com, lonfee88@gmail.com, donghong_ji2000@yahoo.com.cn, tchong616@126.com Abstract Heavy research has been done in recent years on tasks of traditional summarization. How- ever, social context, which is critical in build- ing high-quality social summarizer for web documents, is usually neglected. To address this issue, we propose a novel summarization approach based on social context. In this ap- proach, social summarization is implemented by first employing the tripartite clustering al- gorithm to simultaneously discover document context and user context for a specified docu- ment. Then sentence relationships intra and in- ter documents plus intended user communities are taken into account to evaluate the signifi- cance of each sentence in different context views. Finally, a few sentences with highest overall scores are selected to form the sum- mary. Experimental results demonstrate the ef- fectiveness of the proposed approach and show the superior performance over several baselines. 1 Introduction Now, an increasing number of Web 2.0 applica- tions (e.g. Del.icio.us, CiteULike, LinkedIn) are allowing users to play more active roles such as annotating online documents with free-form tags, submitting opinions on the items they are inter- ested in, or participating in social networks, etc. Despite their focus on different resources, all of them have the common purpose of helping users organize, retrieve and share knowledge. The rich information offered by these systems provides additional clues for collaboratively summarizing online documents in a social con- text. However, most existing methods generate a summary based only on the information within each document or its neighboring documents, while the social context is usually ignored. This is an important issue that has not been exten- sively investigated in the summarization’s litera- ture. In this study, different user's social tagging history and their tagged documents are employed and a novel social summarization approach is proposed by considering both document context and user context in the sentence evaluation proc- ess. The intuition is that if an appropriate social context is available, then integration of social context knowledge into the existing sentence evaluation process will improve the performance of traditional summarization methods. The proposed approach takes into account both the mutual influences between documents and the impact from different user communities, which consists of the following phases. Firstly, the tripartite clustering algorithm is adopted to discover document context and user context by clustering documents, users and tags simultaneously, which is based on the inherent structure of the three kinds of objects. In the clustering result, each document cluster is re- garded as a context for each document in the cluster and all the user clusters are regarded as user communities with diverse information pref- erences. Secondly, the context-sensitive ranking algo- rithm is applied for evaluating the significance …

Content truncated to 500 words