Debasmita Das

Senior AI Specialist at Mastercard

Debasmita is presently working as a Project Lead in the AI Research Team of Mastercard. She has over 5 years of professional experience in the field of Machine Learning, NLP & Data Science, and have previously worked with organizations like J. P. Morgan & EXL. She graduated from IIM Lucknow in 2015. Her paper ‘Information Retrieval and Extraction on COVID-19 Clinical ArticlesUsing Graph Community Detection and Bio-BERT Embeddings’ got accepted (abstract only) in the ACL Workshop 2020 in the COVID Track.

WATCH LIVE: December 10 @ 3:35PM – 4:05PM ET

Deep Learning Algorithm to Rank-Order Resumes using Discriminative Embedding Space

In this paper, we present a method to rank-order resumes for specific roles after reducing the resumes and job descriptions (JDs) to a discriminative embedding space using a Bi-LSTM-CNN network. The algorithm uses a five-level skills hierarchy, developed in an unsupervised manner using a graph-based clustering method. This skills hierarchy is used to assess the semantic similarity between the skills present in a resume and a JD. The algorithm is trained on anonymized internal company data. This approach does not require any explicit feature extractions (e.g. professional experience, projects etc.) and mitigates against the gender / racial bias owing to historical hiring pattern. We tested the algorithm on over 60K anonymized resumes of recent lateral and campus hires based on the recruitment drives conducted by the organisation across various verticals for different regions globally and the results demonstrate that the algorithm can identify over 85% of selected candidates.