Scholarlink

<p>Computational sentiment analysis aims to automatically infer and quantify human opinions and emotions from text. Despite extensive work using machine learning and deep neural architectures, many existing methods depend heavily on labeled data, complex training, and rating supervision, which can limit robustness and interpretability in real-world settings. This study proposes an optimization-driven sentiment tagging framework, the Evaluation based on Distance from Average Solution and Game Theory-Based Sentiment Tagger (EDGT-ST), as an extension of earlier TOGT-ST/BGT-ST models. Rather than only expanding the output space, EDGT-ST modifies the underlying decision mechanism by replacing the TOPSIS-based, rating-dependent ranking with an EDAS-based appraisal that evaluates positive, negative, and neutral classes through their positive and negative deviations from an average multi-criteria solution. In this formulation, sentiment alternatives are scored using context, emotion, and word-count features only, and the resulting EDAS appraisement scores serve as payoffs in a non-cooperative game-theoretic layer, where the final tag is obtained from equilibrium behaviour among competing sentiment cues instead of proximity to explicit star ratings. The EDGT-ST model is evaluated on SemEval, Twitter, and domain-specific review datasets, where it performs rating-independent tertiary sentiment classification and maintains competitive or improved accuracy relative to existing mathematical and learning-based baselines. Its language‑agnostic behavior is demonstrated by applying the same training‑free framework directly to multiple English datasets and a Hindi review dataset, without language‑specific parameter tuning. Statistical analyses indicate that the EDAS-driven, rating-free decision mechanism yields stable sentiment tags under noisy and mixed reviews, while the integrated game-theoretic layer supports interpretable aggregation of multiple textual cues. These properties make EDGT-ST a lightweight and training-free alternative suitable for scenarios where ratings are unavailable or unreliable, and where consistent positive/negative/neutral tagging is required for downstream analytical and business applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Mathematical Optimization-Driven Approach for Enhanced Sentiment Categorization for Textual Data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月20日

查看全文
  • Neha Punetha,
  • Goonjan Jain

摘要

Computational sentiment analysis aims to automatically infer and quantify human opinions and emotions from text. Despite extensive work using machine learning and deep neural architectures, many existing methods depend heavily on labeled data, complex training, and rating supervision, which can limit robustness and interpretability in real-world settings. This study proposes an optimization-driven sentiment tagging framework, the Evaluation based on Distance from Average Solution and Game Theory-Based Sentiment Tagger (EDGT-ST), as an extension of earlier TOGT-ST/BGT-ST models. Rather than only expanding the output space, EDGT-ST modifies the underlying decision mechanism by replacing the TOPSIS-based, rating-dependent ranking with an EDAS-based appraisal that evaluates positive, negative, and neutral classes through their positive and negative deviations from an average multi-criteria solution. In this formulation, sentiment alternatives are scored using context, emotion, and word-count features only, and the resulting EDAS appraisement scores serve as payoffs in a non-cooperative game-theoretic layer, where the final tag is obtained from equilibrium behaviour among competing sentiment cues instead of proximity to explicit star ratings. The EDGT-ST model is evaluated on SemEval, Twitter, and domain-specific review datasets, where it performs rating-independent tertiary sentiment classification and maintains competitive or improved accuracy relative to existing mathematical and learning-based baselines. Its language‑agnostic behavior is demonstrated by applying the same training‑free framework directly to multiple English datasets and a Hindi review dataset, without language‑specific parameter tuning. Statistical analyses indicate that the EDAS-driven, rating-free decision mechanism yields stable sentiment tags under noisy and mixed reviews, while the integrated game-theoretic layer supports interpretable aggregation of multiple textual cues. These properties make EDGT-ST a lightweight and training-free alternative suitable for scenarios where ratings are unavailable or unreliable, and where consistent positive/negative/neutral tagging is required for downstream analytical and business applications.