Entity-Conditioned Question Generation for Robust Attention Distribution in Neural Information Retrieval
This addresses a robustness issue in neural information retrieval for users relying on accurate entity attention, though it is incremental as it builds on existing supervised models.
The paper tackled the problem of sparse attention patterns in supervised neural information retrieval models, which cause key phrases like named entities to receive low attention and lead to under-performance. The result showed that using a novel synthetic data generation method improved attention patterns and retrieval performance on two public benchmarks, including in zero-shot settings.
We show that supervised neural information retrieval (IR) models are prone to learning sparse attention patterns over passage tokens, which can result in key phrases including named entities receiving low attention weights, eventually leading to model under-performance. Using a novel targeted synthetic data generation method that identifies poorly attended entities and conditions the generation episodes on those, we teach neural IR to attend more uniformly and robustly to all entities in a given passage. On two public IR benchmarks, we empirically show that the proposed method helps improve both the model's attention patterns and retrieval performance, including in zero-shot settings.