Answer-biased summarization

Yulianti, E 2018, Answer-biased summarization, Doctor of Philosophy (PhD), Science, RMIT University.


Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Yulianti.pdf Yulianti.pdf Click to show the corresponding preview/stream application/pdf; Bytes
Title Answer-biased summarization
Author(s) Yulianti, E
Year 2018
Abstract Displaying answers directly on the search results page has been shown to have a positive effect on a user's search experience. It leads to good abandonment, a scenario in which users have already found what they require on the results page and therefore do not need to read the documents. A substantial amount of a user's time can then consequently be saved. However, research on finding answers to non-factoid queries has not been extensively explored, even though these types of queries represent the most frequently asked questions on the Web. This situation then raises a need to improve the answers to non-factoid queries in search results. In this thesis, we try to solve this issue by improving search result summaries. We performed an answer-biased summarization of documents in the search results. These answer-biased summaries are expected to contain answers to the user's query. The main challenge of this task is the lexical gap between the query and the sentences containing answers in the document. The answer-bearing sentences may share different vocabularies with the queries. We propose some techniques to improve the extraction of answer-biased summaries from documents. First, we utilize the related content from community question answering (CQA) site to guide the selection of sentences that bear answers. Three optimization-based and one learning-to-rank based methods are proposed.

The effect of the quality of the related CQA content on the accuracy of the generated summaries is also analyzed. Next, we use the semantic and context information for generating answer-biased summaries from documents. Finally, we extend our work on answer-biased summarization for ad-hoc retrieval. Here, the quality features extracted from the summaries are incorporated into ranking functions. Our results show that the related CQA content can be used to improve the creation of answer-biased summaries from documents. There is a significant effect of the CQA content quality on the accuracy of optimization-based summaries. However, such significance is not found on learning-to-rank-based summaries. Further, the semantic and context features are also shown to boost the accuracy of answer-biased summaries. Then, in our extension work on ad-hoc retrieval, the result shows that our ranking methods using answer-biased summaries can give significant improvement over state-of-the-art ranking models.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Science
Subjects Information Retrieval and Web Search
Keyword(s) document summarization
answers
search results
answer-biased summaries
community question answering (CQA)
optimization
learning-to-rank
semantic
context
document ranking
Versions
Version Filter Type
Access Statistics: 69 Abstract Views, 67 File Downloads  -  Detailed Statistics
Created: Mon, 09 Jul 2018, 10:12:29 EST by Denise Paciocco
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us