“Moving money forward“ Money Forward

Contact us

Press Release

  • HOME
  • Press Release
  • A paper by Money Forward Lab on an information extraction method from tables in question-answering systems (Table Retrieval) has been accepted at LREC-COLING 2024, an international conference in the field of natural language processing

2024-05-17

A paper by Money Forward Lab on an information extraction method from tables in question-answering systems (Table Retrieval) has been accepted at LREC-COLING 2024, an international conference in the field of natural language processing

Money Forward, Inc. is pleased to announce that a paper by a member of our research and development organization, Money Forward Lab, has been accepted by LREC-COLING 2024, an international conference in the field of natural language processing. LREC-COLING is a joint conference of LREC, the leading international conference on language resources, and COLING, one of the most prominent international conferences in the field of natural language processing.
The accepted paper presents a new method of information extraction from data tables (Table Retrieval) in question-answering systems and will be presented at LREC-COLING 2024, to be held in Turin, Italy, in May 2024.

■About the Paper
Title: Sub-Table Rescorer for Table Question Answering
Author: Atsushi Kojima
URL: https://lrec-coling-2024.org/list-of-accepted-papers/ Submission ID: 185

Overview
Question-answering systems are one of the systems that provide appropriate answers to questions expressed in natural language. However, the conventional table question-answering system has the problem where if the table used to answer a question is too long, part of it is truncated and not referenced, reducing the accuracy of the answer. Therefore, we have devised a new method called the Sub Table Rescorer. In this method, a long table is first divided into smaller “sub-tables.” The relevance of each sub-table to the question is then calculated, and the more relevant sub-tables are prioritized to arrive at more appropriate answers. In fact, when tested using the WikiTableQuestions dataset*, it increased the probability of returning the correct answer by up to 6.3% over the traditional method.
*One of the datasets used in the field of natural language processing, used to evaluate the performance of question-and-answer systems.

■About Money Forward Lab
Money Forward Lab is engaged in research and development that makes full use of technology and data with the mission of bringing smiles and amazement to people’s lives by unraveling the mechanisms of money. Currently, we are focusing on research and development to realize an “Autonomous BackOffice”, a worldview where future issues that a company’s back office may face are predicted and actions to solve those issues are proposed and executed.

Main Research
・Domain optimization of large-scale language models
・Credit risk assessment
・Financial behavior analysis

■About Money Forward
Name: Money Forward, Inc.
Location: 21F Tamachi Station Tower S, 3-1-21 Shibaura, Minato-ku, Tokyo 108-0023
Representative: Yosuke Tsuji, Representative Director, President and CEO
Establishment: May 2012
Overview: Leading Fintech/SaaS company in Japan / Listed on Prime Section of Tokyo Stock Exchange
URL: https://corp.moneyforward.com/en/
Main Services:
“Money Forward ME”, a service to visualize personal finance https://moneyforward.com/me
“Money Forward Cloud”, SaaS platform for back-office operation https://biz.moneyforward.com/

*All company names and product/service names (including logos) are trademarks or registered trademarks of their respective owners.

For further information, please contact
Public Relations: pr@moneyforward.co.jp

page top