An interactive visualization method for scientific literature exploration based on textual and image content

Abstract

Exploring digital libraries of scientific articles is an essential task to the research community, in both academia and industry. Traditional paradigm presents briefly textual information by article, in a list-based layout without almost visual resources. This design does not allow exploratory operations – as clustering, filtering, or aggregating – on the results delivered by the digital library. Consider the case where the analyst has a referenced article and wants to look for related works. Except for the articles which share citations or common keywords, the retrieved results will be limited to those which fulfill a syntactic match. If instead of having an article as a reference, the user has an image, the process of finding and explore articles with similar content becomes almost non-viable. In this project we propose a visual analytic tool for exploring and analyze scientific document collections by combining image and textual information extracted from articles. Our method relies on the interaction between a ContentBased Image Retrieval (Content-based image retrieval (CBIR)) widget and an interactive multidimensional projection method for mapping similarity relations in 2D space. We exploit other visual resources for representing author and topic information articles according to the selections and preferences of the user. Moreover, a user exploration log is stored for comparing two or more timestamps during the information extraction process, allowing to discover patterns and relations hidden. The effectiveness of our methodology is shown through two case studies and user evaluation, which attest to the usefulness of the proposed framework in exploring scientific document collections.