`包裹，长文用`
`/`
`划分小节。 3. 列表内容用`
- `，引用用`
`，代码用`
`，重要句子可用``。
4. 保留原文所有信息，不得增删改内容。
5. 不要输出``/``/``等文档级标签。
正文内容：
High-granularity topic modelling extracts hundreds of niche topics from a corpus, yet this regime is rarely studied and is more commonly framed as a failure mode: "over-clustering," in which coherent themes fragment into redundant subtopics. We argue that this framing conflates genuine redundancy with topics that are semantically similar but meaningfully distinct, and that the central challenge in high-granularity modelling is separating the two. Recent post-hoc refinement methods address topic redundancy through either embedding-based similarity thresholding or LLM-guided merging. We show that neither suffices in high-granularity settings and propose embedding-gated iterative merge (EGIM), a hybrid approach that iteratively generates context-rich topic representations, uses embedding similarity to propose candidate merges across the topic space, then applies LLM judgement to accept only genuinely redundant pairs while preserving niche distinctions. Across two patent corpora and one legal corpus, LLM judgement rejects 83% of embedding-identified candidates as non-redundant. EGIM improves discriminability on merged topics by 25-55% over a prior iterative LLM baseline while better preserving the original topic distribution.
分析正文结构：
这是一段学术/科技新闻或摘要，可以分为几个自然段，或者加一些小标题以符合“长文用``/`
`划分小节”的要求。虽然文本不长，但可以稍微划分一下，或者就作为一个或两个自然段。
原文本其实是一整段。为了满足“每个自然段落用`
`包裹，长文用`
`，代码用``，重要句子可用``。 4. 保留原文所有信息，不得增删改内容。 5. 不要输出``/``/``等文档级标签。 正文内容： High-granularity topic modelling extracts hundreds of niche topics from a corpus, yet this regime is rarely studied and is more commonly framed as a failure mode: "over-clustering," in which coherent themes fragment into redundant subtopics. We argue that this framing conflates genuine redundancy with topics that are semantically similar but meaningfully distinct, and that the central challenge in high-granularity modelling is separating the two. Recent post-hoc refinement methods address topic redundancy through either embedding-based similarity thresholding or LLM-guided merging. We show that neither suffices in high-granularity settings and propose embedding-gated iterative merge (EGIM), a hybrid approach that iteratively generates context-rich topic representations, uses embedding similarity to propose candidate merges across the topic space, then applies LLM judgement to accept only genuinely redundant pairs while preserving niche distinctions. Across two patent corpora and one legal corpus, LLM judgement rejects 83% of embedding-identified candidates as non-redundant. EGIM improves discriminability on merged topics by 25-55% over a prior iterative LLM baseline while better preserving the original topic distribution. 分析正文结构： 这是一段学术/科技新闻或摘要，可以分为几个自然段，或者加一些小标题以符合“长文用``/`
`划分小节”的要求。虽然文本不长，但可以稍微划分一下，或者就作为一个或两个自然段。 原文本其实是一整段。为了满足“每个自然段落用`
`包裹，长文用`
`/`
`划分小节”，我可以将其拆分为几个语义小节。 小节划分构思：
背景与挑战
High-granularity topic modelling extracts hundreds of niche topics from a corpus, yet this regime is rarely studied and is more commonly framed as a failure mode: "over-clustering," in which coherent themes fragment into redundant subtopics. We argue that this framing conflates genuine redundancy with topics that are semantically similar but meaningfully distinct, and that the central challenge in high-granularity modelling is separating the two.
现有方法与新提出方案
Recent post-hoc refinement methods address topic redundancy through either embedding-based similarity thresholding or LLM-guided merging. We show that neither suffices in high-granularity settings and propose embedding-gated iterative merge (EGIM), a hybrid approach that iteratively generates context-rich topic representations, uses embedding similarity to propose candidate merges across the topic space, then applies LLM judgement to accept only genuinely redundant pairs while preserving niche distinctions.
实验结果
Across two patent corpora and one legal corpus, LLM judgement rejects 83% of embedding-identified candidates as non-redundant. EGIM improves discriminability on merged topics by 25-55% over a prior iterative LLM baseline while better preserving the original topic distribution.
检查是否增删改内容： 原文本： 1. High-granularity topic modelling extracts hundreds of niche topics from a corpus, yet this regime is rarely studied and is more commonly framed as a failure mode: "over-clustering," in which coherent themes fragment into redundant subtopics. 2. We argue that this framing conflates genuine redundancy with topics that are semantically similar but meaningfully distinct, and that the central challenge in high-granularity modelling is separating the two. 3. Recent post-hoc refinement methods address topic redundancy through either embedding-based similarity thresholding or LLM-guided merging. 4. We show that neither suffices in high-granularity settings and propose embedding-gated iterative merge (EGIM), a hybrid approach that iteratively generates context-rich topic representations, uses embedding similarity to propose candidate merges across the topic space, then applies LLM judgement to accept only genuinely redundant pairs while preserving niche distinctions. 5. Across two patent corpora and one legal corpus, LLM judgement rejects 83% of embedding-identified candidates as non-redundant. 6. EGIM improves discriminability on merged topics by 25-55% over a prior iterative LLM baseline while better preserving the original topic distribution. 我划分的段落： 段1包含了1和2。 段2包含了3和4。 段3包含了5和6。 小标题是我加的（"背景与挑战"，"现有方法与新提出方案"，"实验结果"）。规则说“不得增删改内容”，加小标题算不算增改内容？ 为了严格遵守“不得增删改内容”，最好不要自己写小标题，因为小标题属于新增的内容。 那如果不用小标题，只用``包裹自然段落，也可以。原文本实际上是一段，但为了排版好看，可以按句群拆分成几个`
`，只要不改变文字本身。 或者，把原文本视为一个自然段，直接用一个`
`包裹。 “每个自然段落用
包裹，长文用
/
划分小节” —— 这不是强制要求必须用小标题，而是说如果需要划分小节就用。 为了安全起见，避免添加任何原文没有的文字（包括小标题），我只使用`
`和``。 原文本： High-granularity topic modelling extracts hundreds of niche topics from a corpus, yet this regime is rarely studied and is more commonly framed as a failure mode: "over-clustering," in which coherent themes fragment into redundant subtopics. We argue that this framing conflates genuine redundancy with topics that are semantically similar but meaningfully distinct, and that the central challenge in high-granularity modelling is separating the two. Recent post-hoc refinement methods address topic redundancy through either embedding-based similarity thresholding or LLM-guided merging. We show that neither suffices in high-granularity settings and propose embedding-gated iterative merge (EGIM), a hybrid approach that iteratively generates context-rich topic representations, uses embedding similarity to propose candidate merges across the topic space, then applies LLM judgement to accept only genuinely redundant pairs while preserving niche distinctions. Across two patent corpora and one legal corpus, LLM judgement rejects 83% of embedding-identified candidates as non-redundant. EGIM improves discriminability on merged topics by 25-55% over a prior iterative LLM baseline while better preserving the original topic distribution. 我可以把它分成几个
，按原有的句子逻辑：
High-granularity topic modelling extracts hundreds of niche topics from a corpus, yet this regime is rarely studied and is more commonly framed as a failure mode: "over-clustering," in which coherent themes fragment into redundant subtopics. We argue that this framing conflates genuine redundancy with topics that are semantically similar but meaningfully distinct, and that the central challenge in high-granularity modelling is separating the two.
Recent post-hoc refinement methods address topic redundancy through either embedding-based similarity thresholding or LLM-guided merging. We show that neither suffices in high-granularity settings and propose embedding-gated iterative merge (EGIM), a hybrid approach that iteratively generates context-rich topic representations, uses embedding similarity to propose candidate merges across the topic space, then applies LLM judgement to accept only genuinely redundant pairs while preserving niche distinctions.
Across two patent corpora and one