Data Extract: Mining Context from the Web for Dataset Extraction

Abstract

—In this paper we address the problem of dataset extraction from research articles. With the growing digital data repositories and the demand of data centric research in data mining community, finding appropriate dataset for a research problem has become an essential step in scientific research. But given the wide variety of data usage in scientific research it is very difficult to figure out which datasets are most useful for a particular research topic. To alleviate this problem, an automated dataset search engine is a powerful tool. In this work we propose a novel approach to extract dataset names from research articles. We propose a novel way of using " web intelligence " from academic search engines and online dictionaries to mine dataset names from research articles. We also show a comparison between different sources of " web knowledge " by comparing different academic search engines such as Google scholar, Microsoft academic search. The performance of this approach is evaluated using standard information retrieval metric such as precision, recall and F-measure. We get an F-measure of 80%. This accuracy is significant for an unsupervised approach.