The anchor-document data derived from web graphs offers a wealth of paired information for training dense retrieval models in an unsupervised manner. However, the presence of inherent noise invariably compromises the robustness of training dense retrieval models, consequently hurting the performance. In this paper, we introduce WebDRO, an efficient approach for clustering the web graph data and optimizing group weights to enhance the robustness of the pretraining process of dense retrieval models on web graphs. Initially, we build an embedding model for clustering anchor-document pairs. Specifically, we contrastively train the embedding model for link prediction, which guides the embedding model in capturing the inherent document features behind the web graph links. Subsequently, we employ the group distributional robust optimization to recalibrate the weights across different clusters of anchor-document pairs during training dense retrieval models, directing the model to assign higher weights to clusters with higher loss and focus more on worst-case scenarios. Our experiments conducted on MS MARCO and BEIR demonstrate that our method can effectively improve retrieval performance in unsupervised training settings. Further analysis confirms the stability and validity of group weights learned by WebDRO. All codes will be released via GitHub.
翻译:暂无翻译