Optimizing resume information extraction through TSHD segmentation and advanced deep learning techniques

Abstract

This research focuses on a significant factor in the natural language processing area, which is extracting information from unstructured textual data through efficient methods in order to pull useful insights and structured representations from this data. This research attempts to boost the effectiveness of information retrieval systems through computational analysis. This paradigm is explored in this work using question answering models in an extractive style, a modern information extraction approach, creating a new methodology combining the topic segmentation based on headings detection (TSHD) segmentation algorithm and deep learning methods. The TSHD algorithm breaks documents into sections in which certain topics are addressed. Refined extraction models are then used to process these disjoint segments leading to more accurate and contextjudicious extraction compared to naive whole-document extraction approaches. We empirically validate this approach using the stanford question answering dataset (SQuAD) 1.1 dataset, with a specific adaptation to resumes. Experimental results show that the performance metrics increase by 7.4% in exact match (EM) and by 7.8% in F1-score. This can be concluded from these results illustrating the feasibility of the proposed approach in the automated information extraction frameworks such as resume processing.