Multimodal Banking Dataset: Understanding Client Needs through Event Sequences
Keywords: multimodal, multi-temporal, event sequence
Abstract: Financial organizations collect a huge amount of data about clients that typi-
cally has a temporal (sequential) structure and is collected from multiple sources
(modalities). However, despite the urgent practical need, developing deep learn-
ing techniques suitable to handle such data is limited by the absence of large open-
source multi-source real-world datasets of event sequences. To fill this gap mainly
caused by security reasons, we present the industrial-scale publicly available mul-
timodal banking dataset, MBD, that contains more than 2M corporate clients with
several data sources: 950M bank transactions, 1B geo position events, 5M em-
beddings of dialogues with technical support and monthly aggregated purchases
of four bank’s products. All entries are properly anonymized from real proprietary
bank data. Moreover, we introduce a novel multimodal benchmark incorporating
our MBD and two open-source financial datasets. We provide numerical results
demonstrating the superiority of fusion baselines over single-modal techniques
for each task. Moreover, our anonymization techniques still save all significant
information for introduced downstream tasks.
Primary Area: datasets and benchmarks
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.
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: 6922
Loading