We're sorry but this app doesn't work properly without JavaScript enabled. Please enable it to continue.

This lesson's interactive features are locked, please to keep using them

Conflict Resolution in Summaries

Documents often disagree – especially if the types of documents you're searching through are user-generated.

Think about movie reviews for example. Not all reviews of the same movie are going to agree on all points, and you want your LLM summary to give a balanced view of the resulting documents.

Types of Conflicts

Factual conflicts:

  • Doc 1: "Released in 2014"
  • Doc 2: "Released in 2015"

Opinion conflicts:

  • Doc 1: "Masterpiece of cinema"
  • Doc 2: "Overrated and boring"

Emphasis conflicts:

  • Doc 1: "Focuses on action scenes"
  • Doc 2: "Focuses on emotional journey"

Conflict Detection

One way to get better performance surrounding conflicts is to build an entire LLM-step into your pipeline that simply identifies any conflicts in the results before attempting to summarize them.

We're not going to build conflict resolution into Webflyx, because we have a curated dataset of movies themselves, not user-generated reviews; but it's still an important concept to understand.