A Simple Approach to Financial Relation Classification with Pre-trained Language Models

Le Qiu; Bo Peng; Yu Yin Hsu; Emmanuele Chersoni

A Simple Approach to Financial Relation Classification with Pre-trained Language Models

Le Qiu, Bo Peng, Yu Yin Hsu, Emmanuele Chersoni

Research output: Chapter in book / Conference proceeding › Conference article published in proceeding or book › Academic research › peer-review

Abstract

The paper serves as an experimental report submitted to the KDF.SIGIR 2023 shared task on relation extraction, focusing on the REFinD dataset. Motivated by recent advancements on Pre-trained Language Models (PLMs), we propose a simple, yet effective approach that leverages popular PLMs such as BERT, and RoBERTa to address this challenge. The approach capitalizes on the inherent capabilities of PLMs to encode sequences and enrich the semantics of the representations at the entity level.We go beyond the lexical and semantic levels by incorporating supplementary information to tackle the challenges in this task of financial relation classification. In the paper, we detail and justify the approach and report the results of our ablation studies.

Original language	English
Title of host publication	Proceedings of the 4th Workshop on Knowledge Discovery from Unstructured Data in Financial Services
Publisher	Association for Computing Machinery
Publication status	Published - Jul 2023
Event	The 4th Workshop on Knowledge Discovery from Unstructured Data in Financial Services - Taipei International Convention Center, Taipei, Taiwan Duration: 27 Jul 2023 → 27 Jul 2023 https://kdf-workshop.github.io/kdf23/

Workshop

Workshop	The 4th Workshop on Knowledge Discovery from Unstructured Data in Financial Services
Abbreviated title	SIGIR 23 KDF
Country/Territory	Taiwan
City	Taipei
Period	27/07/23 → 27/07/23
Internet address	https://kdf-workshop.github.io/kdf23/

Keywords

financial relation extraction
relation classification
shortest dependency path (SDP)

More information

https://kdf-workshop.github.io/kdf23/assets/images/kdf_s3.pdf

Cite this

@inproceedings{27dabcc1026047ebabf181b6937286a3,

title = "A Simple Approach to Financial Relation Classification with Pre-trained Language Models",

abstract = "The paper serves as an experimental report submitted to the KDF.SIGIR 2023 shared task on relation extraction, focusing on the REFinD dataset. Motivated by recent advancements on Pre-trained Language Models (PLMs), we propose a simple, yet effective approach that leverages popular PLMs such as BERT, and RoBERTa to address this challenge. The approach capitalizes on the inherent capabilities of PLMs to encode sequences and enrich the semantics of the representations at the entity level.We go beyond the lexical and semantic levels by incorporating supplementary information to tackle the challenges in this task of financial relation classification. In the paper, we detail and justify the approach and report the results of our ablation studies.",

keywords = "financial relation extraction, relation classification, shortest dependency path (SDP)",

author = "Le Qiu and Bo Peng and Hsu, {Yu Yin} and Emmanuele Chersoni",

year = "2023",

month = jul,

language = "English",

booktitle = "Proceedings of the 4th Workshop on Knowledge Discovery from Unstructured Data in Financial Services",

publisher = "Association for Computing Machinery",

note = "The 4th Workshop on Knowledge Discovery from Unstructured Data in Financial Services, SIGIR 23 KDF ; Conference date: 27-07-2023 Through 27-07-2023",

url = "https://kdf-workshop.github.io/kdf23/",

}

Qiu, L, Peng, B, Hsu, YY & Chersoni, E 2023, A Simple Approach to Financial Relation Classification with Pre-trained Language Models. in Proceedings of the 4th Workshop on Knowledge Discovery from Unstructured Data in Financial Services. Association for Computing Machinery, The 4th Workshop on Knowledge Discovery from Unstructured Data in Financial Services, Taipei, Taiwan, 27/07/23. <https://kdf-workshop.github.io/kdf23/assets/images/kdf_s3.pdf>

A Simple Approach to Financial Relation Classification with Pre-trained Language Models. / Qiu, Le; Peng, Bo; Hsu, Yu Yin et al.
Proceedings of the 4th Workshop on Knowledge Discovery from Unstructured Data in Financial Services. Association for Computing Machinery, 2023.

Research output: Chapter in book / Conference proceeding › Conference article published in proceeding or book › Academic research › peer-review

TY - GEN

T1 - A Simple Approach to Financial Relation Classification with Pre-trained Language Models

AU - Qiu, Le

AU - Peng, Bo

AU - Hsu, Yu Yin

AU - Chersoni, Emmanuele

PY - 2023/7

Y1 - 2023/7

N2 - The paper serves as an experimental report submitted to the KDF.SIGIR 2023 shared task on relation extraction, focusing on the REFinD dataset. Motivated by recent advancements on Pre-trained Language Models (PLMs), we propose a simple, yet effective approach that leverages popular PLMs such as BERT, and RoBERTa to address this challenge. The approach capitalizes on the inherent capabilities of PLMs to encode sequences and enrich the semantics of the representations at the entity level.We go beyond the lexical and semantic levels by incorporating supplementary information to tackle the challenges in this task of financial relation classification. In the paper, we detail and justify the approach and report the results of our ablation studies.

AB - The paper serves as an experimental report submitted to the KDF.SIGIR 2023 shared task on relation extraction, focusing on the REFinD dataset. Motivated by recent advancements on Pre-trained Language Models (PLMs), we propose a simple, yet effective approach that leverages popular PLMs such as BERT, and RoBERTa to address this challenge. The approach capitalizes on the inherent capabilities of PLMs to encode sequences and enrich the semantics of the representations at the entity level.We go beyond the lexical and semantic levels by incorporating supplementary information to tackle the challenges in this task of financial relation classification. In the paper, we detail and justify the approach and report the results of our ablation studies.

KW - financial relation extraction

KW - relation classification

KW - shortest dependency path (SDP)

M3 - Conference article published in proceeding or book

BT - Proceedings of the 4th Workshop on Knowledge Discovery from Unstructured Data in Financial Services

PB - Association for Computing Machinery

T2 - The 4th Workshop on Knowledge Discovery from Unstructured Data in Financial Services

Y2 - 27 July 2023 through 27 July 2023

ER -

A Simple Approach to Financial Relation Classification with Pre-trained Language Models

Abstract

Workshop

Keywords

More information

Fingerprint

Cite this