Unverified50% confidenceFactExact time
一篇发表于arXiv的研究（arXiv:2609.10893v1）提出假设：在文本中显式注入句法和修辞结构信息能否帮助模型更好地识别文本的不连贯性
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
Unverified实验表明文本连贯性可以作为检测误导性内容的有效代理指标70% similarUnverified团队提出'可言语化'（verbalizability）概念：当图像内容可准确描述成文字时模型表现良好，反之大幅退化69% similarUnverifiedEarly embedding models like Word2Vec and GloVe could only handle word-level representations, while modern models can encode entire sentences or paragraphs67% similarUnverified作者通过用两个不同模型读取同一句子来抵消流畅度与文本质量的干扰66% similarUnverified近义词归一化可通过Sentence-BERT等句嵌入模型将词语映射为稠密向量，再通过余弦相似度计算语义接近程度65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/907931API
curl https://kongchang.com/api/v1/knowledge/claims/907931MCP
get_claim(id=907931)