AI 原生开放教材

外观

AI 工程从零到一（中文）源码研读·中文原文 ↗原件 ↓仓库 ↗

AI 工程从零到一（中文） ​

Given a corpus description (document count, avg length, domain, language, compute budget), output:

  1. Algorithm. LDA / NMF / BERTopic / Top2Vec / FASTopic. One-sentence reason.
  2. Configuration. Number of topics (start at ~sqrt(n_docs)), min_df / max_df filters, embedding model for neural approaches.
  3. Evaluation. Topic coherence (c_v) via gensim.models.CoherenceModel, topic diversity, plus a 20-sample human read.
  4. Failure mode to probe. For LDA, "junk topics" absorbing stopwords and frequent terms. For BERTopic, -1 outlier cluster swallowing ambiguous documents.

Refuse BERTopic on documents longer than the embedding model's context window without a chunking strategy. Refuse LDA on very short text (tweets, reviews under 10 tokens) as coherence collapses. Flag any n_topics choice below 5 or above 200 as likely wrong for real data.