Understanding Matching Mechanisms in Cross-Encoders
This work addresses the lack of detailed understanding of matching processes in cross-encoders for researchers in information retrieval, but it is incremental as it builds on existing interpretability efforts.
The paper tackled the problem of understanding the internal matching mechanisms of cross-encoders in neural IR, finding that straightforward methods can reveal insights such as the crucial roles of specific attention heads and the underlying mechanism for matching detection.
Neural IR architectures, particularly cross-encoders, are highly effective models whose internal mechanisms are mostly unknown. Most works trying to explain their behavior focused on high-level processes (e.g., what in the input influences the prediction, does the model adhere to known IR axioms) but fall short of describing the matching process. Instead of Mechanistic Interpretability approaches which specifically aim at explaining the hidden mechanisms of neural models, we demonstrate that more straightforward methods can already provide valuable insights. In this paper, we first focus on the attention process and extract causal insights highlighting the crucial roles of some attention heads in this process. Second, we provide an interpretation of the mechanism underlying matching detection.