FinIR 2025
The 2nd Workshop on Financial Information Retrieval
in the Era of Generative AI
July 17, 2025 (Padua, Italy)
Co-located with The 48th International ACM SIGIR Conference on Research and Development in Information Retrieval
ABOUT
Recent advancements in Generative AI, such as Large Language Models (LLMs), have demonstrated remarkable success across various general tasks. Extensive studies have explored leveraging generative models in finance, but significant challenges persist. This FinIR workshop aims to explore potential approaches and research directions to address these challenges by equipping generative models with advanced Information Retrieval (IR) models. Specifically, this FinIR workshop seeks to provide a platform for discussing innovative ideas that facilitate the advancement of IR technology to enrich generative models in finance from four key perspectives: (i) financial IR techniques, (ii) financial IR benchmarking and evaluation, (iii) financial systems and agents/assistants, (iv) and trustworthiness, privacy and security when applying financial IR and generative models. This workshop aims to deepen understanding, accelerate progress, and support the advancement of IR technology to enhance generative models to address financial challenges.
PROGRAM
The FinIR workshop will take place in meeting room PALLADIO B on July 17th, 2025.
| Session | Time | Agenda |
|---|---|---|
| Opening | 08:30–08:40 | Opening Remarks |
| Keynote 1 | 08:40–09:20 | Iadh Ounis: Beyond Profit: Evaluating Financial Recommenders with Real-World Transactions |
|
Paper Presentation
|
09:20–09:30 | Financial Risk Identification through Dual-view Adaptation |
| 09:30–09:40 | Testing the Effectiveness of Knowledge Editing in Financial Polarity Editing | |
| 09:40–09:50 | Are LLMs Rational Investors? A Study on the Financial Bias in LLMs | |
| 09:50–10:00 | FinLight: Bridging the Annotation Gap for Highlight Extraction in Financial Reports | |
| 10:00–10:10 | FinS-Pilot: A Benchmark for Online Financial System | |
| 10:10–10:20 | Towards Temporal-Aware Multi-Modal Retrieval Augmented Generation in Finance | |
| 10:20–10:30 | Agentic Retrieval of Topics and Insights from Earning Calls | |
| 10:30–10:40 | Structuring the Unstructured: A Multi-Agent System for Extracting and Querying Financial KPIs and Guidance | |
| Break | 10:40–11:00 | Coffee Break |
| Keynote 2 | 11:00–11:40 | Yongjae Lee: LLMs in Portfolio Selection: Embedding, Bias, and Optimization |
| Keynote 3 | 11:40–12:20 | Rex Ying: Retrieval-Augmented Generation for Time Series |
| Closing | 12:20–12:30 | Closing Remarks |
| Poster Session | 13:00–14:30 | Poster Presentations |
KEYNOTE SPEAKERS
Iadh Ounis
University of Glasgow
Beyond Profit: Evaluating Financial Recommenders with Real-World Transactions
Abstract: Financial asset recommendation is becoming an increasingly important and impactful research topic, as recommender systems become increasingly embedded into trading platforms, robo-advisors and financial chat-bots. However, unlike more traditional recommendation system domains such as product or movie recommendation, there has been comparatively little research into how to evaluate financial recommenders. In this talk, we will introduce the task of financial asset recommendation, and discuss the types of data typically used by financial recommender systems. We will also present a recently released dataset designed for the transaction-based evaluation of recommendation quality. We will explore the key differences between profitability -- commonly used for evaluation in investment contexts -- and transaction-based evaluation metrics, highlighting why these two approaches often yield uncorrelated results. The talk concludes with a forward-looking discussion on how large language model (LLM) agents can be leveraged to enhance both the recommendation and evaluation of financial assets.
Yongjae Lee
Ulsan National Institute of Science and Technology
LLMs in Portfolio Selection: Embedding, Bias, and Optimization
Abstract: Large Language Models (LLMs) are reshaping the landscape of investment decision-making with their capacity for large-scale textual analysis and reasoning. However, concerns about hallucination and reliability continue to limit their direct application in high-stakes financial settings. This talk presents a structured overview of how LLMs can be effectively integrated into the portfolio selection process under uncertainty. We explore three core aspects: (1) fine-tuning LLM embedding models to align investment themes with financial assets, (2) identifying and analyzing biases in LLMs within investment contexts such as sector or strategy preference, and (3) incorporating LLM-generated views into portfolio optimization using the Black-Litterman model. Together, these components offer a practical roadmap for utilizing LLMs in investment workflows, balancing innovation with risk awareness.
Rex (Zhitao) Ying
Yale University
Retrieval-Augmented Generation for Time Series
Abstract: The ubiquity of dynamic data in domains such as weather, healthcare, and energy underscores a growing need for effective interpretation and retrieval of time-series data. These data are inherently tied to domain-specific contexts, such as clinical notes or weather narratives, making cross-modal retrieval essential not only for downstream tasks but also for developing robust time-series foundation models by retrieval-augmented generation (RAG). Despite the increasing demand, time-series retrieval remains largely underexplored. Existing methods often lack semantic grounding, struggle to align heterogeneous modalities, and have limited capacity for handling multi-channel signals. Here I will present our new work, a generic multimodal retriever that grounds time-series embeddings in aligned textual context. The model supports flexible cross-modal retrieval modes, including Text-to-Timeseries and Timeseries-to-Text, effectively linking linguistic descriptions with complex temporal patterns. Our experiments demonstrate its dual utility, as both an effective encoder for downstream applications and a general-purpose retriever to enhance time-series models.
CALL FOR PAPERS
We welcome contributions of research papers from a wide range of topics, including but not limited to the following topics of interest:
-
Financial IR Techniques
- Financial retrieval models including table retrieval, multi-modal retrieval, real-time retrieval, and generative retrieval, etc.
- Techniques for financial query analysis, query rephrase, query expansion, etc.
- Techniques for financial data processing, de-duplication, representation, indexing, filtering, and ranking, etc.
-
Financial IR Evaluation
- Financial IR tasks and benchmarks for evaluating IR models in finance.
- Evaluation metrics and methodologies for measuring financial IR models.
- Evaluation with simulators, such as market simulator.
-
Financial Systems and Agents
- Utilizing IR and generative models to develop systems and applications for financial data analysis, stock price prediction, stock recommendation, portfolio optimization, financial content generation, event forecasting, etc.
- Financial conversational agents and personal financial assistants powered by IR and generative models.
- Multi-agent systems in finance for algorithmic trading, policy design, behavior modeling, risk assessment, etc.
-
Trustworthiness, Privacy and Security
- Exploring and resolving the challenges related to trustworthiness and reliability in applying IR and generative models in finance, including distinguishing AI-generated financial content and ensuring factual consistency, etc.
- Privacy-preserving access and processing financial data, such as sensitive information anonymization and privacy regulations compliance (e.g., PDPA and GDPR).
- Financial data security when using IR technology and generative models, such as financial misinformation detection and data leakage prevention.
PAPER SUBMISSION GUIDLINES
Submissions may range from 4 to 8 pages, with no limit on the number of pages for references. Authors can choose the appropriate length for their paper, as no distinction is made between long and short submissions. All submissions are double-blind and will be peer reviewed by the program committee and judged by their relevance to the workshop, especially to the topics, and their potential to generate discussion.
All submissions should be prepared according to the standard double-column ACM SIG proceedings format. Additional information about formatting and style files is available on the ACM website. Papers must be submitted to easychair at Submission site.
Note that the accepted paper will not be included in the proceedings, so you are free to submit it to other venues.
For inquires about the workshop and submissions, please email to finir2025@easychair.org
ACCEPTED PAPERS
Are LLMs Rational Investors? A Study on the Financial Bias in LLMs
Financial Risk Identification through Dual-view Adaptation
FinLight: Bridging the Annotation Gap for Highlight Extraction in Financial Reports
Testing the Effectiveness of Knowledge Editing in Financial Polarity Editing
Agentic Retrieval of Topics and Insights from Earning Calls
FinS-Pilot: A Benchmark for Online Financial System
Towards Temporal-Aware Multi-Modal Retrieval Augemented Generation in Finance
Structuring the Unstructured: A Multi-Agent System for Extracting and Querying Financial KPIs and Guidance
IMPORTANT DATES
Paper submission due: Apr 30, 2025
Paper acceptance notification: May 21, 2025
Workshop day: July 17, 2025
All times are 23:59, AoE (Anywhere on Earth)
ORGANIZING COMMITTEE
Fengbin Zhu National University of Singapore
Yunshan Ma Singapore Management University
Fuli Feng University of Science and Technology of China
Chao Wang
6Estates Pte Ltd
Singapore
Huanbo Luan
6Estates Pte Ltd
Singapore
Guangnan Ye
Fudan University
China
Shuo Zhang
Bloomberg
United Kingdom
Dhagash Mehta
BlackRock
USA
Pingping Chen
Goldman Sachs
Hong Kong
Bing Xiang
Goldman Sachs
USA
