Deep feature engineering using full-text publications

Abstract

We have observed a rapid proliferation in scientific literature and advancements in web technologies has shifted information dissemination to digital libraries [1]. In general, the research conducted by scientific community is articulated through scholarly publications pertaining high quality algorithms along other algorithmic specific metadata such as achieved results, deployed datasets and runtime complexity. According to estimation, approximately 900 algorithms are published in top core conferences during the years 2005-2009 [2]. With this significant increase in algorithms reported in these conferences, more efficient search systems with advance searching capabilities must be designed to search for an algorithm and its supported metadata such as evaluation results like precision, recall etc., particular dataset on which an algorithm executed or the time complexity achieved by that algorithm from full body text of an article. Such advanced search systems could support researchers and software engineers looking for cutting edge algorithmic solutions. Recently, state designed to search for an algorithm from full text articles [3-5]. In this work, we designed an advanced search engine for full text publications that leverages the deep learning techniques to classify algorithmic specific metadata and further to improve searching capabilities for a search system.