With the increase of digital text and the rise of related metadata, there is a growing interest in finding ways to reduce information overload while still maintaining the most important and useful content. Focused multi-document summarization (MDS) is a process that seeks to condense collections of documents that are related by a query, question, topic, or category down to a passage of only several sentences.

Harabagiu and Lacatusu present research based on topic themes, a new method of topic representation. Topic themes not only improve all aspects of the MDS process, but they also improve one’s understanding of the performance of the various focus-based techniques and their various combinations, with different extraction, compression, and selection methods. These topic themes are basically simple predicate-argument structures. In short, the research compares two of their own novel representations with five state-of-the-art topic representations that use eight theme selection methods (in all, 40 MDS system combinations).

Because of the many explanations of the various MDS topic representation techniques, the fundamental MDS and evaluation measures, and the authors’ methodology, the paper is a bit verbose and information dense. That being said, the paper’s clear writing style makes it accessible to new computational linguistics and natural language processing students, who should read this paper in its entirety. However, experts--readers who are already very familiar with information retrieval and MDS--should use this source as a reference. This elucidation of the MDS field is a great example of thorough experimentation.