NLP Chunking Studio: A Unified Framework for Semantic Chunking and Knowledge-Infused Long-Document Processing

View at publisher Summarise with VIRAT
Authors: Parul Sood, Agraj Chandra, Sanskriti Shrey, Himani Chauhan
Cited by: 0
References: 14

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Processing long documents that exceed the context window of modern transformers remains a major challenge in Natural Language Processing (NLP). Conventional fixed-stride chunking severs semantic links and introduces irrelevant content, leading to suboptimal performance. This paper presents NLP Chunking Studio, a unified framework for long-document processing that integrates: semantic chunking and reconstruction, knowledge graph infusion, a Retrieve-Enhance-Verify (REV) engine, and logical entity weighting (LEW) for triple refinement. We provide detailed design, algorithmic description, and evaluation on machine reading comprehension (CoQA, QuAC, TriviaQA) and industrial corpora (technical reports, contracts). Results show consistent gains in F1, EM and processing efficiency compared to representative baselines. The framework emphasizes interpretability and modularity, making it suitable for high-stakes and domain-specific applications.

Fields of Study & Subjects

Topic Modeling Natural Language Processing Techniques Information Retrieval and Search Behavior

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
867–874
Publication date
21 January 2026
Language
EN