WeaverBird: Empowering Financial Decision-Making with
Large Language Model, Knowledge Base, and Search Engine
Siqiao Xue
♦,∗
, Fan Zhou
♦,∗
, Yi Xu
♦
, Ming Jin
♣
, Qingsong Wen
♥
, Hongyan Hao
♦
, Qingyang Dai
♦
and
Caigao Jiang
♦
, Hongyu Zhao
♦
, Shuo Xie
♦
, Jianshan He
♦
, James Zhang
♦
, Hongyuan Mei
♠
♦Ant Group, ♣Monash University, ♥Alibaba DAMO Academy,
♠
TTIC
{siqiao.xsq,hanlian.zf,haolin.xy}@alibaba-inc.com,ming.jin@monash.edu,qingsongedu@gmail.com
hongyuan@ttic.edu
ABSTRACT
We present WeaverBird, an intelligent dialogue system tailored for
the nance sector. Our system harnesses a large language model
of GPT architecture that has been tuned using extensive corpora
of nancial texts. As a result, our system possesses the capability
to understand complex nancial queries, such as “How should I
manage my investments during ination?”, and provide informed
responses. Furthermore, our system integrates a local knowledge
base and search engine to retrieve relevant information. The nal
system responses are conditioned on the web search and include
proper references to the sources, thus enjoying enhanced credibility.
Comparative evaluations on a wide range of nancial question-
answering highlight our system’s superior performance over other
models. Experience WeaverBird for yourself through our live demo
at https://weaverbird.ttic.edu and view our concise 2-minute video
at https://www.youtube.com/watch?v=yofgeqnlrMc. Our code has also
been made publicly available at https://github.com/ant-research/n_
domain_llm.
CCS CONCEPTS
• Social and professional topics
→
Computing and business;
• Computing methodologies
→
Natural language processing.
KEYWORDS
large language models, nance, dialogue system
ACM Reference Format:
Siqiao Xue
♦,∗
, Fan Zhou
♦,∗
, Yi Xu
♦
, Ming Jin
♣
, Qingsong Wen
♥
, Hongyan
Hao
♦
, Qingyang Dai
♦
and Caigao Jiang
♦
, Hongyu Zhao
♦
, Shuo Xie
♦
, Jian-
shan He
♦
, James Zhang
♦
, Hongyuan Mei
♠
. 2023. WeaverBird: Empowering
Financial Decision-Making with Large Language Model, Knowledge Base,
and Search Engine. In Proceedings of Make sure to enter the correct conference
title from your rights conrmation emai (Conference acronym ’XX). ACM,
New York, NY, USA, 10 pages. https://doi.org/XXXXXXX.XXXXXXX
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for prot or commercial advantage and that copies bear this notice and the full citation
on the rst page. Copyrights for components of this work owned by others than ACM
must be honored. Abstracting with credit is permitted. To copy otherwise, or republish,
to post on servers or to redistribute to lists, requires prior specic permission and/or a
fee. Request permissions from permissions@acm.org.
Conference acronym ’XX, June 03–05, 2023, Woodstock, NY
© 2023 Association for Computing Machinery.
ACM ISBN 978-1-4503-XXXX-X/18/06. . .$15.00
https://doi.org/XXXXXXX.XXXXXXX
1 INTRODUCTION
Everyone faces nancial decisions from time to time. From ques-
tions like “What does the earnings report indicate for stock X?” to
“How should I navigate my investments during ination?”, many
individuals prefer consulting nancial professionals. Yet, not every-
one has the means to access such expert services.
In this paper, we introduce WeaverBird
1
, an intelligent dialogue
system tailored for the nancial sector. Our primary objective is to
bridge the disparity in accessing nancial advisory services and to
democratize nancial expertise. Acting as an intelligent nancial
assistant, WeaverBird is adept at understanding complex economic
and nancial queries, oering nancial perspectives, and enabling
users to make informed nancial decisions. Through the launch of
our system, we aspire to ensure that people across varied nancial
backgrounds gain access to trustworthy nancial advice.
Another key motivation for developing a nance dialogue sys-
tem is to support future research in this area. A comparison of
our WeaverBird system to other competitive approaches is sum-
marized in Table 1. Throughout the development process of our
system, we have amassed extensive collections of nance-centric
text documents, which have been leveraged in ne-tuning language
foundation models. This resource will be available to the research
community for future exploration and analysis. Notably, our collec-
tion features a substantial corpus of Chinese nancial documents,
addressing the gap in the predominantly English-based nance cor-
pora and oering a valuable addition. This equips our WeaverBird
system with a distinctive edge in processing Chinese queries and
delving into discussions concerning Chinese rms and equities. Fur-
thermore, we have constructed a collection of challenging nancial
question-answer pairs, which can serve as a benchmark for evalu-
ating the performance of future models and systems. Each question
stems from a selected document, such as an industry research re-
port. Consequently, answering these questions correctly requires
the discovery of the specic document or equivalent information
from the local knowledge base or across the internet. Our Weaver-
Bird system harnesses the latest research advancements in large
language models, enhanced by our novel training and inference
methods, which we will elaborate on in section 2. This system sets
a robust baseline for evaluating future models and systems in the
nance domain.
*
These authors contributed equally to this work.
1
This name embodies our vision for WeaverBird, symbolizing our commitment to
delivering comprehensive and intelligent support in the nance domain. It represents
the intricate and skillful weaving of nancial knowledge and information, empowering
users with a robust and reliable resource.
arXiv:2308.05361v3 [cs.CL] 2 Dec 2023
资源评论
pk_xz123456
- 粉丝: 2284
- 资源: 2353
下载权益
C知道特权
VIP文章
课程特权
VIP享7折，此内容立减5.97元 开通VIP
最新资源
资源上传下载、课程学习等过程中有任何疑问或建议，欢迎提出宝贵意见哦~我们会及时处理！
点击此处反馈
