Topic
LLMs & Text
Large language models, agents, sentiment and text as data in finance.
- Papers featured
- 600
- Last 12 months
- 34
- Cited 100+
- 109
- Top venue
- Machine learning
Featured per quarter
The last four quarters are highlighted.
Methods that come up
Click one to search for it.
Most cited
Featured papers in this topic with the most citations today.
- 7 Feb 202410,823cites
DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
Advancing Math Reasoning in Language Models: DeepSeekMath7B is a new language model that uses web data and Group Relative Policy Optimization for advanced mathematical reasoning, scoring high on the MATH benchmark.
Machine learning
- 16 Oct 20234,163cites
Mistral 7B
Superior Language Model: Mistral 7B v0.1 is a language model with 7 billion parameters that excels in reasoning, mathematics, and code generation, and has a version specifically designed to follow instructions.
Machine learning
- 7 Aug 20242,390cites
Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters
The research investigates enhancing Large Language Models' (LLMs) performance using more test-time computation, suggesting a compute-optimal scaling strategy based on prompt difficulty.
Machine learning
- 24 Jul 20242,107cites
AWQ: Activation-aware Weight Quantization for On-Device LLM Compression and Acceleration
The study suggests Activation-aware Weight Quantization (AWQ), a hardware-friendly method for quantizing large language models that reduces error and improves performance on various benchmarks.
Machine learningIn GetMobile: Mobile Computing and Communications
- 12 Dec 20241,838cites
Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling
The paper presents InternVL 2.5, a sophisticated multimodal large language model that performs well on various benchmarks, exceeding 70% on the MMMU benchmark.
Machine learning
- 16 Oct 20231,806cites
MemGPT: Towards LLMs as Operating Systems
Extended Context in LLMs: MemGPT is a system that manages different memory levels, providing extended context within large language models' limited context windows, enhancing document analysis and multi-session chat performance.
Machine learningFeatured 2×
- 5 Feb 20251,613cites
s1: Simple test-time scaling
The research presents a method called budget forcing, which uses a small dataset to achieve test-time scaling and improved reasoning performance in language modeling, particularly in competition math questions.
Machine learningIn Conference on Empirical Methods in Natural Language Processing
- 22 May 20241,602cites
DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
MoE Language Model: DeepSeek-V2, a language model with 236B parameters, offers enhanced performance and cost efficiency compared to its predecessor, ranking high among open-source models.
Machine learning
- 16 Oct 20231,410cites
SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models
Hallucination Detection for LLMs: The paper presents SelfCheckGPT, a new approach for fact-checking black-box model responses without an external database, proving its superior ability to detect and rank factual and non-factual sentences.
Machine learningIn Conference on Empirical Methods in Natural Language ProcessingFeatured 2×
- 8 May 20241,079cites
A Simple and Effective Pruning Approach for Large Language Models
Wanda, a new method, efficiently prunes weights in Large Language Models without retraining, offering a more efficient approach to inducing sparsity in pretrained models.
Machine learningIn International Conference on Learning Representations
- 12 Jun 20241,066cites
Simple and Effective Masked Diffusion Language Models
The performance of diffusion models in language modeling has been enhanced by using an effective training recipe and a simplified objective, setting a new standard among diffusion models.
Machine learningIn Neural Information Processing Systems
- 5 Feb 20251,065cites
TÜLU 3: Pushing Frontiers in Open Language Model Post-Training
Open Language Model Post-Training: The Tulu 3 model, a top-tier post-trained language model, is introduced, outperforming other models and providing a detailed guide for its use and adaptation.
Machine learning
Latest
- 6 Nov 20264fanfare
Thinking Fast, Trading Late: System One Decision Models vs the Market
A pre-registered blinded test finds that language-model readings of company news lose their predictive power in real time and that commercial models identify masked companies in two-thirds of cases.
arXiv
- 6 Nov 20264fanfare
LLMs Don't Know What They Know: Benchmarking Temporal Self-Knowledge for Financial Backtesting
A benchmark of 2,226 dated facts reveals that frontier language models underreport their knowledge and answer most facts they claim not to know, compromising backtest validity.
arXiv
- 6 Nov 20262fanfare
Multidimensional Views Behind Target-Price Revisions: An LLM-Based Representation of Analyst Outlooks
Question-based outlook scores extracted from analyst reports by an LLM improve prediction of target-price revisions beyond traditional sentiment measures.
arXiv
- 6 Nov 20264fanfare
The Financial Impact of Non-Financial Beliefs in Large Language Models
Following AI adoption, financial analysts revise high-ESG forecasts upward and become more bullish, but forecast accuracy deteriorates and errors rise for top-ESG firms.
SSRN
- 30 Oct 20263fanfare
FinLLM-KG: Knowledge Graph-Enhanced LLMs for Financial Sentiment Analysis in Algorithmic Trading
FinLLM-KG injects SEC filing facts into sentiment analysis at inference time, improving portfolio performance through explainable company-level context without retraining.
arXiv
- 30 Oct 20263fanfare
LiveSearch: Questions as Executable Verifier Code for a Live Benchmark of Professional-Source Financial Search Agents
A live benchmark stores financial questions as executable code so answers update daily against real data, preventing memorization and revealing that 70% of daily queries change their correct answer.
arXiv
- 30 Oct 20262fanfare
Grounding Long Financial Text with Semantically-Enriched Tables via Multimodal Learning
A transformer framework integrates semantically enriched table embeddings with unstructured text via cross-attention, showing structured data dominates but text adds incremental signal.
arXiv
- 30 Oct 20263fanfare
Generative AI and the Speed of Price Formation Around Earnings Announcements: International Evidence
The study shows that greater ChatGPT exposure in countries accelerates how quickly stock prices incorporate earnings-announcement information, with effects strongest where investors are sophisticated and data is transparent.
SSRN
- 30 Oct 20263fanfare
Opening the Black Box: Which News Explain LLM-Based Stock Ratings?
Research using FinBERT embeddings finds that firm-specific and fundamental news topics explain LLM stock ratings far better than sentiment alone, doubling the model's explanatory power.
SSRN
- 23 Oct 20260cites
From SEC Filings to Financial Signals: Evaluating LLMs on Stock Market Reactions and Analyst Forecasts
Analysis of 13 language models on 45,964 Form 10-Q filings shows finance-specialized models produce modest advantages in return ranking but lag analyst consensus in earnings forecasts.
arXiv
- 16 Oct 20260cites
How Well Do LLMs Perform in Peer Review? Evidence from Finance and Accounting
GPT-4o evaluates research correctness with predictive value for publication but cannot judge originality or significance, suggesting technical peer review can delegate to models while contribution assessment requires people.
arXiv
- 16 Oct 20264fanfare
From Algorithms to Agents: A Survey of LLM-Based Trading Systems
A survey of 47 LLM trading systems across six markets finds that reported margins shrink or vanish when rerun, with best evidence placing language models upstream of execution.
SSRN
- 16 Oct 20263fanfare
Fearful Analysts, Failing Deals: LLM-Based Analyst Sentiment in M&A Conference Calls
Language models analyzing analyst tone in M&A calls reveal that fear and anxiety about deals predict completion delays, failure, and long-run underperformance even when markets initially react favorably.
SSRN
- 16 Oct 20262fanfare
Reading the News Without Hindsight: NoLBERT Topics and the Cross-Section of Stock Returns
A news reader trained only on pre-1995 text produces stable topics but earns 1.12 monthly Sharpe via ensemble with other readers, significant before multiple-testing adjustment.
SSRN
- 16 Oct 20264fanfare
From Text to Evidence: LLMs in Economics and Finance
A review of LLM applications finds that contextual interpretation improves some tasks but measurement error persists, and trading signals do not always generate profits.
RePEc
- 16 Oct 20263fanfare
Measuring sentiment news with transformer-based language models
FinBERT-based sentiment measures agree more closely with human raters than dictionary methods and better distinguish positive, neutral, and negative financial news articles.
RePEc
- 9 Oct 20260cites
Thin Evidence, Thick Priors: How Language Models Substitute Identity for Missing Financial Facts
When financial facts are withdrawn from prompts, identity attributes explain 96% of variation in equity allocation advice from a language model, replacing missing evidence.
arXiv
- 9 Oct 20260cites
MintEval: Do LLMs Implement the Trading Strategy You Asked For? A Behavioural-Equivalence Benchmark for Natural-Language-to-Strategy Code
Introduces a benchmark that compares LLM-generated trading code to reference strategies bar by bar on identical data, revealing silent failures in implementation despite passing functional tests.
arXiv
- 9 Oct 20262fanfare
Large Language Models and Information Frictions in Corporate Bonds
Shows that large-language-model scores of default risk from earnings calls predict bond rating migration one year ahead where analysts disagreed, capturing information not yet priced in spreads.
SSRN
- 9 Oct 20263fanfare
Geoeconomic Fear and Sovereign Spreads
The study constructs a geoeconomic fear index from news articles and finds that higher fear predicts wider emerging-market sovereign spreads, especially for speculative-grade issuers.
SSRN
- 2 Oct 20260cites
Resolving the Missing Financial Data Crisis: A Generative AI Pipeline for SEC 10-K Extraction
Large language models outperform traditional methods at extracting missing financial data from SEC filings, with accuracy improving as model size matches document complexity.
arXiv
- 2 Oct 20263fanfare
Transformer-based CoVaR: Systemic Risk in Textual Information
Integrating financial news embeddings from large language models with market data, the study improves systemic risk forecasts using conditional value-at-risk without requiring large datasets.
RePEc
- 2 Oct 20262fanfare
One advisor for the whole world? Cross-country evidence on financial advice from large language models
Large language models provide nearly identical portfolio advice across twenty-one countries despite local differences, following retail finance conventions rather than academic prescriptions and ignoring household balance sheets.
RePEc
- 25 Sep 20260cites
Financial Language Models as Applied Artificial Intelligence Systems for News-Based Trading under Market Frictions
Introduces a market-friction-aware framework that converts timestamped financial news into auditable trading decisions while accounting for execution timing, transaction costs, and liquidity constraints.
arXiv
- 25 Sep 20260cites
From Tone to Trajectory: Continuous Sentiment and the Shape of Monetary Policy Communication
The study shows that how monetary policy sentiment unfolds across a press conference, not just its average tone, predicts rate changes and shapes forecaster expectations at the ECB and Fed.
arXiv
- 25 Sep 20260cites
FinInteract: Benchmarking Clarification and Intent Integration in Ambiguous Financial Question Answering
A benchmark reveals that language models answer financial questions above 90 percent with clarification but only 28.9 percent when they must elicit it themselves, exposing model ambiguity resolution.
arXiv
- 25 Sep 20260cites
FinRankGRPO: Optimizing LLMs for Listwise Financial Asset Ranking via Group Relative Policy Optimization
Develops a two-stage framework that fine-tunes language models for listwise asset ranking using Spearman rank correlation rewards, achieving a Sharpe ratio of 0.636 on asset allocation.
arXiv
- 25 Sep 20263fanfare
LLM-Based Semantic Surprises in FOMC Communication: Asset Prices and Financial-Market Stress
Semantic surprises extracted from Federal Reserve statements predict subsequent financial-stress dynamics and reduce forecast error by up to 23%, particularly when initial stress is high or during recessions.
SSRN
- 16 Apr 20264cites
Transforming the Voice of the Customer: Large Language Models for Identifying Customer Needs
Large Language Models are streamlining the process of identifying customer needs, letting analysts concentrate on more valuable work while still delivering precise insights.
arXiv
- 28 Dec 2025105shares
Twitter Sentiment and Financial Trends
A new financial sentiment index derived from Twitter data shows strong links to market conditions and can forecast stock market returns, particularly in response to changes in U.S. monetary policy.
SSRNFeatured 2×