@inproceedings{hennig-labor-09_topic-based-summarization-PLSA,
title={Topic-based multi-document summarization with probabilistic latent semantic analysis},
author={Hennig, Leonhard and Labor, DAI},
booktitle={Recent Advances in Natural Language Processing (RANLP)},
year={2009}
annote = {
\textbf{Overview}: This paper considers the problem of query-focused multidocument summarization, where a summary containing the information most relevant to a user's information need is produced from a set of topic-related documents. They propose a new method based on probabilistic latent semantic analysis (PLSA), which allows them to represent sentences and queries as probability distributions over latent topics. Their approach combines query focused and thematic features computed in the latent topic space to estimate the summary relevance of sentences. In addition, they also evaluate several different similarity measures for computing sentence-level feature scores.
\\
\\
\textbf{Approach}: They follow a greedy approach for selecting sentences, and penalize candidate sentences based on their similarity to the partial summary. Firstly, they associate sentences and queries with a representation in the latent topic space of a PLSA model by estimating their mixing proportions. They then compute several sentence-level features based on the similarity of sentence and query distributions over latent topics. Finally, they combine individual feature scores linearly into an overall sentence score to create a ranking, which we use to select sentences for the summary.
\\
\\
\textbf{What's new}: They examine how a PLSA model can be used to represent documents, sentences and queries in the context of multi-document summarization, and investigate which measures are most useful for computing similarities in the latent topic space. They do not only pick sentences from topics with the highest likelihood in
the training data, but compute a sentence's score based on a linear function of query-focused and thematic features.
Finally, their work outperforms the best reported results on DUC 2006 data, and also compares well on DUC 2007 data.
\\
\\
-- PRASHANT JALAN, 11523, Y11
}
}