Retrieval Augmented Imputation using Data Lake Tables

Chenyu Yang, Yuyu Luo, Chuanxuan Cui, Ju Fan, Chengliang Chai, Nan Tang

Keywords: data imputation, dense retrieval, contrastive learning

Abstract:

Data imputation is an essential problem in many data science applications. Existing methods often struggle to impute missing values in scenarios where there is a lack of sufficient data redundancy. In this paper, leveraging large language models (LLMs) and data lakes, we propose a novel approach for retrieval-augmented imputation called RAI, utilizing fine-grained tuple-level retrieval instead of traditional coarse-grained table-based retrieval. RAI addresses the challenges of retrieving relevant tuples for missing value imputation from a data lake, where tuples have heterogeneous attributes, diverse values, and missing values. Rather than simply searching for similar tables, RAI employs a tuple encoder to learn meaningful representations for capturing tuple similarities and differences, enabling effective identification of candidate tuples. The retrieved results are further refined by a tuple reranker. We also introduce a new benchmark, mvBench, to advance further research. Extensive experiments demonstrate that RAI significantly outperforms existing methods.
We conduct extensive experiments, demonstrating that RAI significantly outperforms state-of-the-art table-based retrieval-augmented imputation methods by 10.7%.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 6504