Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{193683,
        author = {Naresh Kumar Yalla and V. Kamakshi Prasad},
        title = {A Comprehensive Survey on Interpretable Sentence-Level Relevance and Importance Modeling: From Statistical Heuristics to Transformer-Based and Correlation-Guided Frameworks},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {10},
        pages = {861-873},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=193683},
        abstract = {Sentence-level relevance scoring underpins modern information retrieval, extractive summarization, question answering, and decision-support systems, where ranking quality directly influences interpretability, reliability, and downstream reasoning. Research in this area has evolved from statistical heuristics and graph-based centrality to neural scoring–selection architectures, transformer encoders, and emerging large language model (LLM)-assisted estimation. Despite substantial gains in semantic expressiveness, persistent structural challenges remain, including fragmented operational definitions of relevance, redundancy amplification arising from correlated signals, limited transparency of factor interactions, evaluation protocols overly dependent on lexical overlap, and instability under domain shift or prompt variation.
This survey provides a structured and governance-oriented synthesis of sentence-level relevance modeling. We formalize relevance as a context-conditioned, multi-factor scoring function and analyze how lexical, semantic, structural, discourse, and informativeness signals are integrated across paradigms. Beyond chronological review, we organize existing methods through the lenses of multi-factor interaction modeling, correlation-aware signal governance, and reliability-sensitive evaluation. Particular emphasis is placed on redundancy control, feature decorrelation, representation stability, and faithfulness-aligned validation mechanisms essential for preventing proxy dominance and unstable attribution.
By comparatively analyzing modeling assumptions and evaluation practices across tasks, this survey identifies systemic gaps and articulates research directions toward explainable, correlation-governed, and stability-calibrated sentence-level relevance frameworks. We argue that advancing the field requires repositioning relevance modeling from performance-centric optimization toward governance-aware and reliability-sensitive design suitable for real-world deployment.},
        keywords = {Sentence-level relevance, extractive summarization, information retrieval, explainable AI, correlation-aware modeling, faithfulness, transformer models, large language models},
        month = {March},
        }