Research & Publications

My research focuses on natural language processing and applied machine learning, with an emphasis on transformer models, rigorous evaluation and interpretable results.

  • Natural Language Processing
  • Text Summarisation
  • Transformer Models
  • Interpretable Machine Learning
  • Applied Data Science

Published Research

  1. 2026
    Journal Article Peer-reviewed Open Access

    A Comparative Study of Hybrid Text Summarization Techniques Using Traditional, Static, and Contextual Word Embeddings

    Lilakant Pokhrel, Sangita Pokhrel, Swathi Ganesan, Nalinda Somasiri

    Artificial Intelligence and Applications, Bon View Publishing, 2026 (Online First)

    Summary

    This paper builds a hybrid extractive and abstractive summarisation system and compares word embeddings across three families: traditional (Count Vector, TF-IDF), static (Word2Vec, GloVe, FastText) and contextual (ELMo, GPT, BERT). Eight summarisation techniques were integrated into a Django web application and evaluated with ROUGE on the CNN/DailyMail dataset. GPT gave the strongest results, followed by TextRank and ELMo-based systems.

    Cite

    Pokhrel, L., Pokhrel, S., Ganesan, S., & Somasiri, N. (2026). A comparative study of hybrid text summarization techniques using traditional, static, and contextual word embeddings. Artificial Intelligence and Applications. https://doi.org/10.47852/bonviewAIA62026330

    • Text summarization
    • Hybrid summarization
    • Contextual embeddings
    • Word embeddings
    • NLP
    Article PDF DOI BibTeX
  2. 2024
    MSc Dissertation

    Text Summarisation with Deep Learning Integration

    Lilakant Pokhrel

    MSc Data Science, York St John University, 2024 (Distinction)

    Abstract

    Currently, there is an exponential growth in textual data due to the rapid expansion of internet technology and various social and entertainment platforms. This is causing widespread negative effects on document and text management, text classification, and information retrieval, among other areas. To address the challenge of translating large volumes of text into natural-sounding language that retains citations, quotes, and references, summarization research is gaining significant international attention. The focus is on developing an effective hybrid (extractive and abstractive) summarization system by implementing different word embedding and summarization techniques. The study examines various embedding methods, from traditional approaches (Count Vector, TF-IDF) to static word embeddings (Word2Vec, GloVe, FastText), and contextualized embeddings (ELMo, GPT, BERT). It employs efficient extractive methods (TF-IDF, Textrank, LSA, Word2Vec, ELMo with K-means) and abstractive methods (Bi-LSTM, GPT) to create an effective summarization system. By integrating eight summarization techniques into Django-based (Python) web applications, the research finds that GPT performs particularly well, followed by TextRank and Elmo-based systems, with their ROUGE scores compared. The study also discusses the challenges faced in developing such a system and outlines future directions for ongoing research.

    • NLP
    • BERT
    • T5
    • Abstractive summarisation
    • ROUGE
    PDF DOI Code

Under Review & In Progress

Papers currently under review will be listed here.

Interested in collaborating?

I welcome research collaborations and conversations in NLP, machine learning and data science.

Get in touch View projects