arXiv cs.CY· Fengming Liu, Mingyu Shu·· 23 小时前AI 评分39
研究揭示嵌入余弦相似度主要反映文本长度而非同质化
Embedding Cosine Measures Text Length Before It Measures Homogenization
AI 导读
对 606 万条 Reddit 评论的重新测量发现,未中心化的 DeBERTa-v3 余弦相似度主要追踪文本长度:将评论从 10 词截断到 150 词,平均余弦相似度提升幅度是原报告断点的 8 倍。
正文 · 原文
Abstract:Mean pairwise cosine similarity between transformer embeddings is a standard measure of whether online discourse is converging under generative AI. When the embeddings are mean-pooled outputs of an encoder never trained for sentence similarity, this measure mostly tracks text length. In 6.06 million Reddit comments, an uncentered DeBERTa-v3 measure had suggested that only conservative discourse broke toward homogeneity after late 2022. Truncating the same comments from 10 to 150 words raises their mean cosine by eight times the reported break. A placebo that swaps every comment for a random one of equal length reproduces 81% of the asymmetry, and a fixed-content simulation reproduces 78%. The leading principal direction carries 99.9% of raw DeBERTa homogeneity, and once length is handled the asymmetry disappears under every encoder. On 575,000 Stack Exchange answers the measure reports a post-ChatGPT homogenization of technical sites that the placebo fully reproduces, although answers grew longer. A census of 31 studies locates the risk where anisotropic encoders meet time-series or length-unequal designs. A reverse control on a well-known experiment shows that centering with renormalization can reverse a genuine effect. We propose a low-cost diagnostic protocol. Corrected measures show that the two political camps converged between 2019 and 2022 and stopped converging before ChatGPT. Cross-camp replies engage their targets more closely than same-camp replies.
| Comments: | 11 pages, 10 figures, 2 tables. Re-measurement of arXiv:2608.13674. v2: added co-author Mingyu Shu |
| Subjects: | Computation and Language (cs.CL); Computers and Society (cs.CY); Social and Information Networks (cs.SI) |
| Cite as: | arXiv:2611.21519 [cs.CL] |
| (or arXiv:2611.21519v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.21519 arXiv-issued DOI via DataCite |
Submission history
From: Fengming Liu [view email]
[v1]
Mon, 16 Nov 2026 18:12:15 UTC (206 KB)
[v2]
Wed, 18 Nov 2026 08:46:51 UTC (201 KB)
来源:arXiv cs.CY · arxiv.org