Reproduce “A Multi-Granularity-Aware Aspect Learning Model for Multi-Aspect Dense Retrieval” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xiaojie Sun, Keping Bi, Jiafeng Guo, Sihui Yang, Qishen Zhang, Zhongyi Liu, Guannan Zhang, Xueqi Cheng
The MUlti-granulaRity-aware Aspect Learning model (MURAL) enhances dense retrieval by effectively integrating explicit multi-granularity aspect information from structured data into item representations. It achieves substantial performance improvements over state-of-the-art multi-aspect dense retrieval models on large-scale e-commerce and mini-program search datasets.
Key insights
- Capturing aspect information at multiple granularities (phrase, word, token) is crucial for a nuanced understanding of semantic relationships between aspect values, leading to improved retrieval accuracy.
- Dedicated aspect tokens combined with the MLM objective enable robust aspect learning even when explicit aspect-value annotations are sparse or missing.
- Allowing aspect embeddings to specialize via 'unshared' embedding tables, separate from general PLM token embeddings, optimizes their ability to represent specific aspect semantics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.