Using Syntactic and Shallow Semantic Kernels to Improve Multi-Modality Manifold-Ranking for Topic-Focused Multi-Document Summarization

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 1098–1106, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Using Syntactic and Shallow Semantic Kernels to Improve Multi-Modality Manifold-Ranking for Topic-Focused Multi-Document Summarization Yllias Chali University of Lethbridge Lethbridge, AB, Canada chali@cs.uleth.ca Sadid A. Hasan University of Lethbridge Lethbridge, AB, Canada hasan@cs.uleth.ca Kaisar Imam University of Lethbridge Lethbridge, AB, Canada imam@uleth.ca Abstract Multi-modality manifold-ranking is re- cently used successfully in topic-focused multi-document summarization. This ap- proach is based on Bag-Of-Words (BOW) assumption where the pair-wise similar- ity values between sentences are computed using the standard cosine similarity mea- sure (TF*IDF). However, the major lim- itation of the TF*IDF approach is that it only retains the frequency of the words and disregards the syntactic and semantic information. In this paper, we propose the use of syntactic and shallow semantic ker- nels for computing the relevance between the sentences. We argue that the addi- tion of syntactic and semantic information can improve the performance of the multi- modality manifold-ranking algorithm. Ex- tensive experiments on the DUC bench- mark datasets prove the effectiveness of our approach. 1 Introduction Text summarization is a good way to compress a huge amount of information into a concise form by selecting the most important information and discarding redundant information. According to Mani (2001), automatic text summarization takes a partially-structured source text from multiple texts written about the same topic, extracts in- formation content from it, and presents the most important content to the user in a manner sensi- tive to the user’s needs. In contrast to summariz- ing one document that is termed as single docu- ment summarization, multi-document summariza- tion deals with multiple documents as sources that are related to one main topic under consideration. As compared to generic summarization that must contain the core information central to the source documents, the main goal of topic-focused multi- document summarization (i.e. query-based multi- document summarization) is to create from the documents a summary that can answer the need for information expressed in the topic or explain the topic (Wan et al., 2007). In this paper, we con- sider the problem of producing extraction-based1 topic-focused multi-document summaries given a collection of documents. In recent years, a variety of manifold-ranking based methods are applied successfully to topic- focused multi-document summarization. The ba- sic manifold-ranking method is a typical graph- based summarization method that makes uni- form use of the sentence-to-sentence relation- ships and the sentence-to-topic relationships in a manifold-ranking process (Wan et al., 2007). In the multi-modality manifold-ranking algorithm, sentence relationships are classified into within- document relationships and cross-document re- lationships, and each kind of relationships are considered as a separate modality (graph) (Wan and Xiao, 2009). These methods are based on Bag-Of-Words (BOW) assumption where the pair-wise similarity values between the sentences are computed using the standard cosine measure (TF*IDF). The major limitation of the TF*IDF ap- proach is that it only retains the frequency of the words and does not take into account …

Content truncated to 500 words