Working Notes on Late Interaction Dynamics: Analyzing Targeted Behaviors of Late Interaction Models
This work addresses performance bottlenecks in retrieval models for researchers, but it is incremental as it validates existing theories without introducing new methods.
The paper tackled the understudied dynamics of Late Interaction models by analyzing length bias and similarity distribution beyond top scores on the NanoBEIR benchmark, finding that theoretical length bias holds in practice and the MaxSim operator efficiently exploits token-level similarities.
While Late Interaction models exhibit strong retrieval performance, many of their underlying dynamics remain understudied, potentially hiding performance bottlenecks. In this work, we focus on two topics in Late Interaction retrieval: a length bias that arises when using multi-vector scoring, and the similarity distribution beyond the best scores pooled by the MaxSim operator. We analyze these behaviors for state-of-the-art models on the NanoBEIR benchmark. Results show that while the theoretical length bias of causal Late Interaction models holds in practice, bi-directional models can also suffer from it in extreme cases. We also note that no significant similarity trend lies beyond the top-1 document token, validating that the MaxSim operator efficiently exploits the token-level similarity scores.