Decision-Based Retrieval
A retrieval paradigm that treats retrieval as a decision problem rather than a similarity-matching problem. By incorporating goal-awareness and information-gain criteria, it actively evaluates which content should genuinely be placed into the context window.
Timeline (last 90 days)
Information gain—the new, non-redundant facts or reasoning basis that a piece of content contributes relative to the existing context—is the core judgment dimension of decision-based retrieval.
In multi-hop reasoning scenarios, each hop's retrieval should incrementally supplement based on the conclusions of the previous hop, and the decision model can explicitly track the known information set to avoid redundant recall.
All Facts (2)
Information gain—the new, non-redundant facts or reasoning basis that a piece of content contributes relative to the existing context—is the core judgment dimension of decision-based retrieval.
50%UnverifiedIn multi-hop reasoning scenarios, each hop's retrieval should incrementally supplement based on the conclusions of the previous hop, and the decision model can explicitly track the known information set to avoid redundant recall.
50%