Abstract: The validity of evidence value assessment directly affects judicial fairness. The traditional assessment model that relies on expert experience suffers from problems such as subjective bias. The likelihood ratio (LR) based on the Bayesian framework provides an objective measurement of probative force by calculating the probability ratio of evidence under the prosecution and defense hypotheses, and has become a key tool for connecting the technical and legal dimensions. This paper systematically reviews the research status of the LR method in the three major fields: physical evidence, forensic medical examination, and audio-visual material identification. From a technical perspective, the LR method covers feature-based and score-based feature acquisition methods, as well as model construction techniques using classical statistics and artificial intelligence. System performance calibration and evaluation techniques continue to improve, providing dual guarantees for the reliability and accuracy of LR results. Current research shows new trends from experience-driven to data-driven, from single-modal to multi-modal fusion, and from laboratory analysis to real-time intelligent on-site detection. The LR method not only promotes technological updates but also facilitates the transformation of judicial proof concepts to an “experience-data” hybrid model. However, China still faces challenges in database construction, standardized validation, interdisciplinary talent cultivation, and judicial acceptance. This paper aims to provide a theoretical reference for building an interpretable, verifiable, and scalable evidence assessment framework, and to assist in the standardization of forensic science evidence assessment and the practice of judicial fairness in China.

Key words: Bayesian theory, scientific evidence, judicial proof, statistics

摘要：  证据价值评估的有效性直接影响司法公正，传统依赖专家经验的评估模式存在主观偏差等问题。基于贝叶斯框架的似然比（likelihood ratio，LR）通过计算证据在控辩双方假设下的概率比，为证明力提供客观度量，已成为连接技术维度与法律维度的关键工具。本文系统综述了LR方法在物证类、法医类和声像资料鉴定三大领域的研究现状。从技术路径看，LR方法涵盖特征基与分数基的特征获取方式，以及经典统计学与人工智能的模型构建技术，系统性能校准与评价技术持续完善，为LR结果的可靠性与准确性提供了双重保障。当前研究呈现经验驱动向数据驱动、单模态向多模态融合、实验室分析向实时智能化现场检测的新趋势。LR方法不仅推动技术更新，更促进司法证明理念向“经验-数据”混合模式转变。但我国在数据库建设、标准化验证、跨学科人才培养和司法接受度等方面仍面临挑战。本文旨在为构建可解释、可验证、可推广的证据评估框架提供理论参照，助力我国法庭科学证据评估标准化与司法公正实践。

关键词: 贝叶斯理论, 科学证据, 司法证明, 统计学