一句话：指出局部相似匹配忽略周围类上下文，提出面向 Android 第三方库检测的全局匹配思路。
第三方库检测对维护 Android 供应链安全至关重要，但代码混淆使准确识别十分困难。已有工作采用模糊库签名以容忍混淆影响，却常产生多个高度相似的候选。现有方法多依赖局部相似度极大值，忽略周围类所包含的上下文知识，导致匹配结果次优。本文据此提出面向 Android 第三方库检测的全局匹配思路，旨在利用类间上下文缓解局部匹配偏差，从而提升混淆场景下的库检测质量。摘要未报告具体实验结果，但指出了从局部相似到全局上下文匹配的改进方向。
The detection of third-party libraries (TPLs) is pivotal in upholding the security of Android supply chains. A significant hurdle in this domain pertains to ensuring accurate detection, particularly in the face of prevalent code obfuscation techniques. Recently, multiple research endeavors have advocated for the adoption of fuzzy library signatures to accommodate specific code features that can be affected by code obfuscation. Despite the potential promise of this approach, the use of such abstractions often results in the identification of multiple candidates that bear a striking resemblance. Unfortunately, prevailing methodologies tend to rely on local maxima of similarity matches, disregarding the broader contextual knowledge encapsulated within surrounding classes. This oversight culminates in suboptimal matching outcomes.