Developing text mining and visualization system for numerical information pairs

Abstract

We constructed a system for this study to automatically extract numerical pairs from documents related to a certain topic and display them in graphs. It first extracts two units and one item expression from documents, it then extracts numerical pairs from sentences including the two units and the item expression, and it finally arranges the pairs and displays them in graphs. When we judged the extraction of a correct graph as the top output in the experiments to be correct, our best system accuracy was 0.2222 in Evaluation A and 0.4444 in Evaluation B. When we judged the extraction of a correct graph in the top five outputs to be correct, the best accuracy rose to 0.3889 in Evaluation A and 0.5000 in Evaluation B. (In Evaluation A, a graph where 75\% or more of the points was correct was judged to be correct and in Evaluation B, a graph where 50\% or more of the points was correct was judged to be correct.) Our system is convenient and effective because it can output a graph that includes numerical pairs at these levels of accuracy when only given a set of documents as input. We also showed some graphs that were automatically constructed by our methods and taught us many interesting things in this paper.