arXiv cs.AI· Luc Hazenoot, Zhaochun Ren, Amirhossein Zohrehvand·· 3 天前AI 评分22
用 LLM 激活值测量文本中的概念内容：基于概念向量与线性探针的 ESG 证据
Measuring Concept Content in Text from LLM Activations: ESG Evidence from Concept Vectors and Linear Probes
AI 导读
研究比较了 Recursive Feature Machine（RFM）与线性探针从 LLM 激活值中提取概念的效果，在人工标注的 ESG 金融文本数据集上，最佳线性探针准确率比微调领域分类器低 0.6 至 3.3 个百分点，但在所有对比中均优于模型自身回答，且准确率全面超过 RFM 概念向量。在合成混合测试中，线性探针的均值池化分数比概念向量分数更能跟踪段落中概念句的数量。
正文 · 原文
Abstract:Recent work shows that Large Language Models (LLMs) encode more information in their activations than they show in their responses. We ask whether the information contained in activations could be used as a source of measurement for the presence of concepts. We compare two algorithms for extraction of concepts, i.e., Recursive Feature Machine (RFM) algorithm and linear probing against an embedding baseline, surface baselines, and the model's own response using a human-annotated Environmental, Social and Governance (ESG) dataset of financial texts. The best linear probe ranges from 0.6 to 3.3 percentage points below fine-tuned domain classifiers. The probes also outscore the model's own response in all comparisons. Furthermore, the probes outperform the RFM concept vectors in all accuracy comparisons. In a synthetic mixture test, their mean-pooled score tracks the number of concept sentences in a paragraph better than the concept-vector score. Put together, a linear probe on a frozen LLM is a cost-effective and generalizable alternative to fine-tuning a domain classifier.
| Comments: | 26 pages, 3 figures, 16 tables |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); General Economics (econ.GN) |
| ACM classes: | I.2.7; I.2.6; J.4 |
| Cite as: | arXiv:2608.07208 [cs.CL] |
| (or arXiv:2608.07208v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2608.07208 arXiv-issued DOI via DataCite |
Submission history
From: Amirhossein Zohrehvand [view email]
[v1]
Fri, 7 Aug 2026 13:21:50 UTC (106 KB)
[v2]
Sun, 11 Oct 2026 20:23:01 UTC (86 KB)
来源：arXiv cs.AI · arxiv.org