Entity-Conditioned Question Generation for Robust Attention Distribution in Neural Information Retrieval

TL;DR

This paper introduces a method to improve neural information retrieval models by training them to distribute attention more evenly across passage entities, enhancing robustness and performance, especially in zero-shot scenarios.

Contribution

The paper proposes a novel synthetic data generation approach that conditions training on poorly attended entities to promote uniform attention in neural IR models.

Findings

Improved attention distribution over entities in IR models.

Enhanced retrieval performance on benchmark datasets.

Robustness gains observed in zero-shot settings.

Abstract

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.