Abstract:
Finding pertinent material fast is becoming more and more challenging for researchers
and students as a result of the research publications' rapid development into digital
libraries. The problem is exacerbated by subscription-based platforms, which frequently
offer restricted access to entire publications, forcing consumers to make their decisions
only on abstracts. To address these challenges, this study explores multi-document
summarization techniques aimed at improving document retrieval and relevance
assessment. We evaluate Maximal Marginal Relevance (MMR), Centroid, and
PageRank algorithms, assessing their performance on datasets such as Multi-XScience
and a manually curated set using ROUGE metrics. The findings demonstrate that MMR
has the best recall-to-precision ratio, making it an excellent choice for summarizing a
variety of scientific document sets. Centroid is appropriate in situations when speed is
of the essence since, although being marginally less accurate, it drastically cuts down
on processing time. Even if it is less useful in this situation, PageRank nonetheless
offers insightful information for ranking-based strategies. To facilitate more effective
navigation of large datasets and enable users to make well-informed decisions, this
study highlights the significance of incorporating these summarizing techniques into
digital library systems. The findings optimize the trade-off between computing speed
and summary quality, which advances current attempts to improve scientific knowledge
retrieval.
