{
  "context": "Evaluates how well the solution builds and uses a gensim-based topic modeling pipeline: preparing corpora, training topics, inferring distributions, summarizing topics, and persisting artifacts for reuse. Focuses on correct application of Gensim primitives rather than custom reimplementation.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "Dictionary prep",
      "description": "Raw texts are tokenized (e.g., via gensim.utils.simple_preprocess), optional stopwords are removed, and a corpora.Dictionary is built then applied with doc2bow to create the training corpus without reimplementing these steps.",
      "max_score": 20
    },
    {
      "name": "Model training",
      "description": "Topic model is trained with gensim.models.LdaModel or LdaMulticore using the bag-of-words corpus and num_topics/passes parameters instead of a custom algorithm.",
      "max_score": 30
    },
    {
      "name": "Inference",
      "description": "Inference for new text converts tokens with the same Dictionary and uses LdaModel.get_document_topics (or equivalent __getitem__) to produce a normalized topic distribution ordered by probability.",
      "max_score": 20
    },
    {
      "name": "Topic terms",
      "description": "Top terms per topic are obtained through model helpers such as show_topic/print_topics with the requested topn and preserve descending weight order instead of manual sorting.",
      "max_score": 15
    },
    {
      "name": "Persistence",
      "description": "Model and dictionary persistence relies on built-in save/load methods (e.g., LdaModel.save/load and Dictionary.save/load) so a reloaded pipeline reproduces prior inference results.",
      "max_score": 15
    }
  ]
}