Renxian Zhang

Twitter hyperlink recommendation with user-tweet-hyperlink three-way clustering

Twitter, the most famous micro-blogging service and online social network, collects millions of tweets every day. Due to the length limitation, users usually need to explore other ways to enrich the content of their tweets. Some studies have provided findings to suggest that users can benefit from added hyperlinks in tweets. In this paper, we focus on the hyperlinks in Twitter and propose a new application, called hyperlink recommendation in Twitter. We expect that the recommended hyperlinks can be used to enrich the information of user tweets. A three-way tensor is used to model the user-tweet-hyperlink collaborative relations. Two tensor-based clustering approaches, tensor decomposition-based clustering (TDC) and tensor approximation-based clustering (TAC) are developed to group the users, tweets and hyperlinks with similar interests, or similar contexts. Recommendation is then made based on the reconstructed tensor using cluster information. The evaluation results in terms of Mean Absolute Error (MAE) shows the advantages of both the TDC and TAC approaches over a baseline recommendation approach, i.e., memory-based collaborative filtering. Comparatively, the TAC approach achieves better performance than the TDC approach.

Enhancing diversity and coverage of document summaries through subspace clustering and clustering-based optimization

Abstract Sentence clustering has been successfully applied in document summarization to discover the topics conveyed in a collection of documents. However, existing clustering-based summarization approaches are seldom targeted for both diversity and coverage of summaries, which are believed to be the two key issues to determine the quality of summaries. The focus of this work is to explore a systematic approach that allows diversity and coverage to be tackled within an integrated clustering-based summarization framework. Given the fact that normally each topic can be described by a set of keywords and the choice of the keywords among the topics is topic-dependent, we take the advantage of the newly emerged subspace clustering to enable the flexibility of keyword selection and the improved quality of sentence clustering. On this basis, we develop two clustering-based optimization strategies, namely local optimization and global optimization to pursue our targets. Experimental results on the DUC datasets demonstrate effectiveness and robustness of the proposed approach.

Combining co-clustering with noise detection for theme-based summarization

To overcome the fact that the length of sentences is short and their content is limited, we regard words as independent text objects rather than features of sentences in sentence clustering and develop two co-clustering frameworks, namely integrated clustering and interactive clustering, to cluster sentences and words simultaneously. Since real-world datasets always contain noise, we incorporate noise detection and removal to enhance clustering of sentences and words. Meanwhile, a semisupervised approach is explored to incorporate the query information (and the sentence information in early document sets) in theme-based summarization. Thorough experimental studies are conducted. When evaluated on the DUC2005-2007 datasets and TAC 2008-2009 datasets, the performance of the two noise-detecting co-clustering approaches is comparable with that of the top three systems. The results also demonstrate that the interactive with noise detection algorithm is more effective than the noise-detecting integrated algorithm.

Learning features through feedback for blog distillation

The paper is focused on blogosphere research based on the TREC blog distillation task, and aims to explore unbiased and significant features automatically and efficiently. Feedback from faceted feeds is introduced to harvest relevant features and information gain is used to select discriminative features. The evaluation result shows that the selected feedback features can greatly improve the performance and adapt well to the terabyte data.

Information Ordering with an Event-Enriched Vector Space Model for Multi-Document News Summarization

Information ordering is a nontrivial task in multi‐document summarization (MDS), which typically relies on the traditional vector space model (VSM) notorious for semantic deficiency. In this article, we propose a novel event‐enriched VSM to alleviate the problem by building event semantics into sentence representations. The mediation of event information between sentence and term, especially in the news domain, has an intuitive appeal as well as technical advantage in common sentence‐level operations such as sentence similarity computation. Inspired by the block‐style writing by humans, we base the sentence ordering algorithm on sentence clustering. To accommodate the complexity introduced by event information, we adopt a soft‐to‐hard clustering strategy on the event and sentence levels, using expectation–maximization clustering and K‐means, respectively. For the purpose of cluster‐based sentence ordering, the event‐enriched VSM enables us to design an ordering algorithm to enhance event coherence computed between sentence and sentence–context pairs. Drawing on the findings of earlier research, we also incorporate topic continuity measures and time information into the scheme. We evaluate the performance of the model and its variants automatically and manually, with experimental results showing clear advantage of the event‐based model over baseline and non‐event‐based models in information ordering for multi‐document news summarization. We are confident that the event‐enriched VSM has even greater potential in summarization and beyond, which awaits further research.

Coherent narrative summarization with a cognitive model

Borrowing theories from cognitive psychology, we propose a computational model of human cognition.Using the cognitive model, we generate coherent narrative summaries.We propose a novel method of proposition-level extractive summarization.We verify the cognitive model and summarization method with narrative text data. For summary readers, coherence is no less important than informativeness and is ultimately measured in human terms. Taking a human cognitive perspective, this paper is aimed to generate coherent summaries of narrative text by developing a cognitive model. To model coherence with a cognitive background, we simulate the long-term human memory by building a semantic network from a large corpus like Wiki and design algorithms to account for the information flow among different compartments of human memory. Proposition is the basic processing unit for the model. After processing a whole narrative in a cyclic way, our model supplies information to be used for extractive summarization on the proposition level. Experimental results on two kinds of narrative text, newswire articles and fairy tales, show the superiority of our proposed model to several representative and popular methods.

Towards content-level coherence with aspect-guided summarization

The TAC 2010 summarization track initiated a new task—aspect-guided summarization—that centers on textual aspects embodied as particular kinds of information of a text. We observe that aspect-guided summaries not only address highly specific user need, but also facilitate content-level coherence by using aspect information. In this article, we present a full-fledged approach to aspect-guided summarization with a focus on summary coherence. Our summarization approach depends on two prerequisite subtasks: recognizing aspect-bearing sentences in order to do sentence extraction, and modeling aspect-based coherence with an HMM model in order to predict a coherent sentence ordering. Using the manually annotated TAC 2010 and 2010 datasets, we validated the effectiveness of our proposed methods for those subtasks. Drawing on the empirical results, we proceed to develop an aspect-guided summarizer based on a simple but robust base summarizer. With sentence selection guided by aspect information, our system is one of the best on TAC 2011. With sentence ordering predicted by the aspect-based HMM model, the summaries achieve good coherence.

LDA-Based Topic Formation and Topic-Sentence Reinforcement for Graph-Based Multi-document Summarization

In recent years graph-based ranking algorithms have attracted much attention in document summarization. This paper introduces our recent work on applying a topic model, namely LDA, in graph-based summarization. In the proposed approach, LDA is used to automatically identify a set of semantic topics from the documents to be summarized. The identified topics are then used to construct a bipartite graph to represent the documents. Topic-sentence reinforcement is implemented to calculate the salience scores of topics and sentences simultaneously. By incorporating the information embedded in the topics, the sentence ranking result can be improved. Experiments are conducted on the DUC 2004 data set to evaluate the effectiveness of the proposed approach.