Individual Linguistic Granular Computing: A Granulation–Degranulation-Based Approach
提出一种基于粒化-去粒化的个体语言粒度计算方法，用截断和镜像幂律分布处理语义歧义和语境变化，在航空发动机风险评估和在线评论语义分析中优于其他词计算模型和七种大语言模型。
This study proposes a granulation-degranulation-based approach for individual linguistic granular computing that addresses two types of uncertainty: ambiguity in semantic representation and contextual variability in semantic choices. Current studies typically use intervals as the formalism for granulation, with random sampling on these intervals to perform degranulation. In contrast, this study employs a more versatile probability-sampling-based degranulation method that uses truncated and mirrored power-law distributions. This method is based on three key assumptions derived from the central limit theorem and establishes a relationship between the power-law index and the uncertainty in interpreting linguistic terms. Moreover, to enhance alignment with human perception, this study designs an optimization framework that integrates interval-based granulation with the proposed probability-sampling-based degranulation method. The effectiveness and practicality of the proposed approach are validated through an experimental study on the risk assessment of aircraft engines. The results of the experiments on the semantic analysis of online reviews demonstrate that the proposed approach achieves superior performance, as evidenced by comparisons with other computing-with-words models and seven large language models.