crossread_assist_ai_agents_workflow

to help reader connect stuffs between 2 sources
Requirement ?

• identify keyword’s types between 2 sources
  • technical terms, nouns, proper nouns(specific name)
• identify semantic relationship between each keywords?
• identify atomic core (summary) from each source’s content (more abstract compared to the previous)
  • e.g. core(summary) of the 33rd paragraph: Steve Jobs did a right thing when designing Macintosh

Questions

• are those two sources related by keywords
  • keywords linking
• are those two sources emitted similar reasoning
  • relationship between keywords comparison
• are those two sources emitted similar conclusion
  • core idea comparison: consensus, different, opposite ?

1. get bag of keywords from each source
   1. (separate between each sources, each source uses different AI)for each chapter, each paragraph, list out important keywords (technical terms, common nouns, proper nouns)
2. get semantic relationship between keywords
   1. (separate between each sources) from the previous step, find relationship between each keywords (meaningful triples, not all keywords)
   2. concat, compare between 2 sources (compare between two sources, if some keywords are founded in both sources, are they have the same relationship?)
3. get summary
   1. (separate between each sources, each source uses different AI) for each chapter, each paragraph, get summary
   2. the center AI find connection between two sources

https://gemini.google.com/share/20843eaf9c0c

• handle synonym
  • Entity Linking
  • coreference resolution
    • https://arxiv.org/html/2504.05767v1
    • thaicoref https://arxiv.org/abs/2406.06000
    • https://neurosys.com/blog/intro-to-coreference-resolution-in-nlp
• Comparing “relationships” is hard. Is “Steve founded Apple” the same as “Apple was created by Steve”? A standard string comparison will fail here. You need Vector Embeddings or Natural Language Inference (NLI) to detect if these statements entail each other.
  • vector embedding
  • NLI
• What it is: Determining if Sentence A implies, contradicts, or is neutral towards Sentence B.
  • NLI
  • Textual Entailment

created on: Wed Dec 10 2025