句子主题相似度判定算法开发需求及应用场景问询

I’m working on building an algorithm to assess whether a follow-up sentence fits into an existing narrative sequence. The core task is: given an initial starting sentence, determine if subsequent sentences are appropriate to add as complementary content that maintains the ongoing story or topic thread.

Judgment Examples #

• Valid starting sentence:

  My dog loves to drink water.
  This is a valid opening sentence that meets our baseline requirements.

• Fitting follow-up (shared subject):

  The dog hates cats.
  This sentence is acceptable—both the starting sentence and this one revolve around the core subject of "the dog," maintaining thematic consistency.

• Fitting follow-up (clear pronoun reference):

  It enjoys walks on the beach.
  This sentence works because the pronoun "it" clearly refers back to the dog established earlier, aligning with logical narrative flow.

• Non-fitting follow-up (off-topic):

  Pizza is great with pineapple on top.
  This sentence is not acceptable, as it completely breaks away from the dog-focused narrative thread built by the previous three sentences.

Context & Use Case #

I’ve already built a library that generates Markov text chains from input text. After applying grammatical corrections, the library produces syntactically valid sentences—but I need a way to string these sentences together into coherent, thematically consistent paragraphs. This coherence-checking algorithm will be the key technical component to make that possible.

Content of this question originates from Stack Exchange, asked by Bryant Makes Programs