The increasing use of complex “black-box” models for financial news sentiment analysis presents a challenge in high-stakes settings where transparency and trust are paramount. This study introduces and validates a finance-focused, fully reproducible, open-source workflow for building, explaining, and evaluating sector-specific sentiment models mapped to standard market taxonomies and investable proxies. We benchmark interpretable and transformer-based models on public datasets and a newly constructed, manually annotated gold-standard corpus of 1500 U.S. sector-tagged financial headlines. While a zero-shot FinBERT establishes a reasonable baseline (macro F1 = 0.555), fine-tuning on our gold data yields a robust macro F1 = 0.707, a substantial uplift. We extend explainability to the fine-tuned FinBERT with Integrated Gradients (IG) and LIME and perform a quantitative faithfulness audit via deletion curves and AOPC; LIME is most faithful (AOPC = 0.365). We also quantify the risks of weak supervision: accuracy drops (−21.0%) and explanations diverge (SHAP rank ρ = 0.11) relative to gold-label training. Crucially, econometric tests show the sentiment signal is reactive, not predictive, of next-day returns; yet it still supports profitable sector strategies (e.g., Technology long-short Sharpe 1.88). Novelty lies in a finance-aligned, sector-aware, trustworthiness blueprint that pairs fine-tuned FinBERT with audited explanations and uncertainty checks, all end-to-end reproducible and tied to investable sector ETFs.

Sentiment analysis plays a crucial role in domains such as financial news, user reviews, and public opinion monitoring, yet existing approaches face challenges when dealing with long and domain-specific texts due to semantic dilution, insufficient context modeling, and dispersed emotional signals. To address these issues, a multi-granularity attention-based sentiment analysis model built on a transformer backbone is proposed. The framework integrates sentence-level and document-level hierarchical modeling, a different-dimensional embedding strategy, and a cross-granularity contrastive fusion mechanism, thereby achieving unified representation and dynamic alignment of local and global emotional features. Static word embeddings combined with dynamic contextual embeddings enhance both semantic stability and context sensitivity, while the cross-granularity fusion module alleviates sparsity and dispersion of emotional cues in long texts, improving robustness and discriminability. Extensive experiments on multiple benchmark datasets demonstrate the effectiveness of the proposed model. On the Financial Forum Reviews dataset, it achieves an accuracy of 0.932, precision of 0.928, recall of 0.925, F1-score of 0.926, and AUC of 0.951, surpassing state-of-the-art baselines such as BERT and RoBERTa. On the Financial Product User Reviews dataset, the model obtains an accuracy of 0.902, precision of 0.898, recall of 0.894, and AUC of 0.921, showing significant improvements for short-text sentiment tasks. On the Financial News dataset, it achieves an accuracy of 0.874, precision of 0.869, recall of 0.864, and AUC of 0.895, highlighting its strong adaptability to professional and domain-specific texts. Ablation studies further confirm that the multi-granularity transformer structure, the different-dimensional embedding strategy, and the cross-granularity fusion module each contribute critically to overall performance improvements.

Background: The accurate prediction of stock prices remains a significant challenge due to market volatility, non-linear patterns, and the influence of diverse factors like economic indicators and investor sentiment. Traditional models like ARIMA and shallow machine learning models often fail to capture these complexities. Methods: This study develops and evaluates a Stacked Long Short-Term Memory (LSTM) neural network model for stock market forecasting. The model integrates heterogeneous data sources, including technical indicators, macroeconomic variables, and sentiment scores derived from financial news headlines using FinBERT. The architecture employs two LSTM layers, dropout regularization, and Bayesian hyperparameter tuning. We trained the model on a diverse dataset of four major indices (S&P 500, NASDAQ, FTSE 100, Nikkei 225) from 2010 to 2024 and evaluated its performance in a zero-shot setting on the unseen Shanghai Stock Exchange (SSE) Composite index. Results: The proposed Stacked LSTM model significantly outperformed traditional benchmarks (ARIMA, GARCH, SVM). It achieved an R² of 0.956 and the lowest RMSE (105.18) on the test set. In the critical zero-shot cross-market evaluation on the SSE, the model demonstrated strong generalization with an R² of 0.93, whereas ARIMA and GARCH failed entirely (R² ≈ 0). The incremental addition of technical indicators and sentiment features progressively improved predictive accuracy, with sentiment reducing short-term prediction errors by approximately 15% during volatile periods. Conclusion: The results confirm that a deep Stacked LSTM architecture, enriched with multi- source features and robust regularization, provides a superior and generalizable framework for stock market forecasting, offering substantial improvements over conventional methods.

ABSTRACT A systematic review of experimental research on automated journalism was conducted to examine the dominant theoretical and methodological trends across 27 experimental studies from a larger database search (n = 1,523). The analysis revealed that nearly half of coded studies did not include a manipulation check; as a result, half of the published studies did not offer evidence that their manipulation was recalled successfully by participants. A second limitation was that most studies did not justify their desired sample size with a power analysis. The coded literature was internationally diverse, mostly used validated measures for outcome variables, and frequently tested the effects of automation across multiple news topics. Common theoretical frameworks included the MAIN model/machine heuristic and expectation confirmation/violation theory. More than 25 different outcome variables were coded, most of which focused on evaluations of news writing, with a particular emphasis on credibility. Sports and finance news were most commonly tested as the context for automation effects. Mediation analysis was uncommon in the literature, although tests of moderation effects were more common. Directions for future research based on the results of the systematic review are provided.

Forecasting market movements in stocks, gold, and crude oil requires a deep understanding of how financial news sentiment influences asset prices. Analyzing news sentiment is crucial for understanding market dynamics and forecasting price fluctuations. However, creating accurate financial news datasets, particularly in terms of proper labeling and sourcing, continues to be a significant challenge. This paper presents a comprehensive literature review on financial news sentiment analysis and its application in market trend prediction.By reviewing articles in reputable journals from 2018–2025, we consolidate key findings, including techniques for dataset creation, labeling, and sourcing, as well as the use of advanced methods such as Natural Language Processing (NLP) and deep learning models. This review contributes to the growing literature on sentiment analysis in the context of the relationship between stocks and commodities, especially gold, crude oil, and the role of global and market specific news sentiments in determining the assets prices. The study focuses on issues that concern researchers in this regard; it also compares the relative success of various prediction models and discusses the criteria for assessing their effectiveness.We propose solutions to current challenges and outline future research directions to improve sentiment analysis in financial markets.

ABSTRACT Headlines are crucial for attracting click-through rates, and several factors influence audience preferences. While certain factors affect general audiences, there is a gap in the literature regarding specific factors that impact click-through rates among professional audiences. This study uses an econometric model to analyze a real-world time-series dataset (N = 8,179) of headlines from a financial news outlet catering to a global professional audience. The evidence reveals that headlines using simple words are preferred by audiences of financial professionals. Further testing on stipulated writing rules shows that some protocols enhance click-through rates, while others are less effective. These results suggest distinct audience preferences regarding headline readability and writing rules, providing valuable insights for journalists and editors targeting this specific group of readers.

Environmental, Social, and Governance (ESG) investing has gained unprecedented momentum in global financial markets, driving the need for sophisticated analytical frameworks that can process vast amounts of unstructured information. This research presents a comprehensive investigation into the application of natural language processing techniques for ESG news sentiment analysis and its subsequent impact on investment portfolio performance. The study develops a multi-dimensional sentiment analysis model that extracts ESG-related information from financial news sources, incorporating advanced text mining algorithms to quantify sentiment scores across environmental, social, and governance dimensions. Through empirical analysis of portfolio performance metrics, the research demonstrates that ESG sentiment-driven investment strategies yield superior risk-adjusted returns compared to traditional approaches. The methodology integrates real-time news processing capabilities with portfolio optimization algorithms, enabling dynamic allocation decisions based on sentiment-derived ESG signals. Experimental results indicate a 50.8% improvement in Sharpe ratio and 17.3% reduction in portfolio volatility when incorporating ESG sentiment analysis. The findings contribute to the advancement of sustainable finance technology and provide practical insights for institutional investors seeking to enhance portfolio performance through alternative data integration.

Innovative Paths of Financial Data News Visualization in the AI Era: A Case Study of DT Finance

ABSTRACT This study investigates the relationship between the forecasting power of web-based financial news sentiments and market capitalisation by analysing large, mid, and small-cap Indian stocks at both the index and sectoral levels. The analysis is further extended to two emerging Asian economies, Malaysia, and Vietnam. Given the dominant contribution of the finance sector, the study analyses the impact of financial news sentiments within the financial sector across each market cap. For this, 1,54,448 news headlines are extracted from an online news website, where financial news is identified using a TFIDF-GRU model and a keyword search with 187 financial terms. Sentiments are computed using a hybrid Doc2Vec-TFIDF feature extraction technique and an SVM classifier. The study employs a hybrid BiLSTM-GRU model incorporating web-based financial news sentiments alongside technical and macroeconomic indicators such as 10-year bond yield, exchange rate, gold price, crude oil price, and S&P500 closing price. Findings reveal that the forecasting power of web-based financial news sentiments varies significantly with market cap, with a strong impact on large-cap and mid-cap stocks. The study holds significant economic and policy implications, offering actionable insights for stakeholders across financial markets, regulatory bodies, and government.

We propose a two-stage statistical learning framework to investigate how financial news headlines posted over weekends affect stock returns. In the first stage, Natural Language Processing (NLP) techniques are used to extract sentiment features from news headlines, including FinBERT sentiment scores and Impact Probabilities derived from Logistic Regression models (Binomial, Multinomial, and Bayesian). These Impact Probabilities estimate the likelihood that a given headline influences the stock’s opening price on the following trading day. In the second stage, we predict over-weekend log returns using various sets of covariates: sentiment-based features, traditional financial indicators (e.g., trading volumes, past returns), and headline counts. We evaluate multiple statistical learning algorithms—including Linear Regression, Polynomial Regression, Random Forests, and Support Vector Machines—using cross-validation and two performance metrics. Our framework is demonstrated using financial news from MarketWatch and stock data for Apple Inc. (AAPL) from 2014 to 2023. The results show that incorporating sentiment features, particularly Impact Probabilities, improves predictive accuracy. This approach offers a robust way to quantify and model the influence of qualitative financial information on stock performance, especially in contexts where markets are closed but news continues to develop.

Managing personal finances remains a significant challenge in developing economies, where access to financial advisory services is limited and financial literacy is often low. In response to this, we propose «ECO InvestMind AI», an intelligent mobile application designed to democratise wealth management through the integration of advanced machine learning (ML) and natural language processing (NLP) techniques. Our system leverages a dual-stream hybrid architecture combining a fine-tuned BERT model for sentiment analysis of financial news and an MLP-based time series model for stock price forecasting. The application provides personalised stock recommendations by analysing market trends, Environmental, Social and Governance (ESG) scores and sentiment signals, offering users real-time, behaviourally adaptive financial guidance. To ensure scalability and maintainability, the app is built using Clean Architecture principles and Kotlin Multiplatform, enabling seamless cross-platform performance. A relational database structure supports the integration of diverse datasets, including historical stock prices, ESG metrics and sentiment-labelled news. Evaluation results demonstrate high predictive accuracy across multiple Kazakhstani companies and indices. This paper illustrates how AI-powered tools like «ECO InvestMind AI» can bridge the gap between traditional investment tools and modern financial advisory systems, empowering users with actionable insights and promoting inclusive participation in financial markets. The proposed hybrid algorithm achieved a total system accuracy of 96.87%, confirming the effectiveness of combining sentiment analysis and time series forecasting for intelligent investment decision-making.

This article examines how Polish business journalism framed wartime economic developments during the Russia–Ukraine conflict (2022–mid-2024). Drawing on a mixed-methods analysis of 395 headlines from major financial news outlets, the authors combine thematic categorization with sentiment analysis to trace how the media represented key sectors such as energy, agriculture and finance. The findings reveal an early dominance of negative sentiment, followed by a gradual shift toward neutral and notably ambivalent framing – especially in response to complex issues like the grain import crisis and transport sector disruptions. The study highlights ambivalence as a strategic framing device that resists binary evaluations, and reflects the layered economic realities of prolonged conflict. The authors argue that business journalism plays a critical role not only in conveying economic information, but also in shaping public interpretation and institutional response during periods of systemic uncertainty.

Stock price prediction is a challenging task with dynamic trends and volatile markets due to opinion and sentiment forces in the market. The conventional Autoregressive Integrated Moving Average (ARIMA) and Support Vector Regression (SVR) methods only take into account historical numerical values and ignore the influence of current financial news. Thus, they do not model the interdependence between historical stock prices and opinion data and are plagued by lower precision and prediction power. To overcome these drawbacks, this study proposes a Hybrid Sentiment-Aware Stock Prediction Model (HSASP) that integrates Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) with an Attention Mechanism (AM). The CNN captures spatial relations from Tesla's past stock history, and BiLSTM uses opinion-based information from Reddit News to capture temporal relations. The AM selects the most significant features by assigning weights to valuable details, enhancing the predictability of the model. The suggested HSASP model improves accuracy by 18%, precision by 16%, and trend consistency by 20%, being successful for stock price prediction. With the integration of price- and opinion-based information, the suggested model provides a strong option for decision-making with high efficiency and precision.

Major sell-side institutions have begun embedding large-language-model (LLM) “desk copilots” such as Bank of America’s Maestro and Goldman Sachs’ GS AI Assistant into sales-and-trading workflows to synthesize internal research, client flow data, and market-microstructure signals in real time (Financial News London, 2024; Reuters, 2024). This review paper surveys the emerging body of academic, regulatory, and practitioner literature on generative-AI trade assistants (GATAs), framing their potential to reshape pre-trade analytics across equities, foreign exchange, and derivatives markets. We synthesize findings on three core dimensions—information asymmetry, order-routing efficiency, and conduct-risk controls—and propose a conceptual evaluation framework to guide regulators and market participants. The paper concludes by identifying open research questions around model governance, fairness, and systemic risk propagation.

Stock price forecasting remains a persistent challenge in time series analysis due to complex inter-stock relationships and dynamic textual signals such as financial news. While Graph Neural Networks (GNNs) can model relational structures, they often struggle with capturing higher-order dependencies and are sensitive to noise. Moreover, sentiment signals are typically aggregated using fixed time windows, which may introduce temporal bias. To address these issues, we propose DASF-Net (Diffusion-Aware Sentiment Fusion Network), a multimodal framework that integrates structural and textual information for robust prediction. DASF-Net leverages diffusion processes over two complementary financial graphs—one based on industry relationships, the other on fundamental indicators—to learn richer stock representations. Simultaneously, sentiment embeddings extracted from financial news using FinBERT are aggregated over an empirically optimized window to preserve temporal relevance. These modalities are fused via a multi-head attention mechanism and passed to a temporal forecasting module. DASF-Net integrates daily stock prices and news sentiment, using a 3-day sentiment aggregation window, to forecast stock prices over daily horizons (1–3 days). Experiments on 12 large-cap S&P 500 stocks over four years demonstrate that DASF-Net outperforms competitive baselines, achieving up to 91.6% relative reduction in Mean Squared Error (MSE). Results highlight the effectiveness of combining graph diffusion and sentiment-aware features for improved financial forecasting.

Investor attitude is an important factor in driving the behavior of the stock market, usually dwarfing conventional financial markets. Although previous study has investigated the mental side of investing, this has been concentrated mostly on the developed world and has had no real-time behavioral incorporation. This study fulfills this requirement by investigating the roles played by investor mentality, media reporting, and macroeconomic triggers in determining stock market volatility during bear markets with a special emphasis on Kathmandu. The goal is to measure the effect of sentiment on market performance based on both structured survey data and unstructured social media and news text. A hybrid methodological design is used, combining statistical modeling (Chi-Square test) and a machine learning-based sentiment classification approach utilizing a fine-tuned BERT (Bidirectional Encoder Representations from Transformers) model. Social media and financial news information are both drawn from a 2022–2025 Kaggle Stock Market Sentiment dataset, and NVivo provides thematic analysis of investor interviews. The following BERT model, coded through Python's Hugging Face Transformers package, resulted in a 9.2% improvement on accuracy compared to standard sentiment analysis models such as logistic regression and SVM, providing richer contextual knowledge of investor emotion. Empirical findings indicate a statistically significant relationship between sentiment and investment behavior (χ² = 23.250, p = 0.006), validating market change behavior. This study not only emphasizes sentiment as an important predictive variable but also proposes a scalable, real-time predictive framework for risk-sensitive investment strategy. The application of this model can improve the early identification of volatility and guide sentiment-based trading systems for investors, analysts, and policymakers in developing markets.

This study explores whether sentiment extracted from financial news using large language models (LLMs) can predict abnormal intraday stock returns following dividend announcements. Drawing on 4,682 news items linked to 1,258 announcements from 394 S&P 500 companies (January 2023–January 2024), we use ChatGPT to extract sentiment polarity scores and we apply different models to forecast cumulative abnormal returns (CARs) in 30-minute intervals. Our findings reveal that sentiment – especially when captured immediately after news releases – has significant predictive power over intraday price movements. Strategies based on ChatGPT-derived sentiment consistently outperform benchmark models, particularly within the first two hours of trading. These results remain robust across alternative specifications and placebo tests, highlighting the value of LLMs for real-time market prediction. This research advances the literature on sentiment analysis and behavioral finance by linking emotion-driven news interpretation to high-frequency trading performance.

This study explores the integration of artificial intelligence (AI) in the Indian stock market, focusing on its impact on equity trading and investment decisions. The research adopts a mixed-methods approach, utilising surveys and interviews to analyse traders’ profiles, social media influence and the effectiveness of AI-driven tools. Findings reveal that most traders are young investors allocating less than 25% of their portfolios to equities. Social media platforms, particularly Telegram and Instagram, play a significant role in shaping investment decisions, with stock analysis and recommendations being the most consumed content. AI tools, such as stock screeners and financial news aggregators, have moderately improved decision-making efficiency, though concerns regarding their reliability persist. The study underscores the need for enhanced educational initiatives for traders, improved AI tool functionality and stricter regulatory frameworks to ensure transparency and trust in AI-driven trading. By addressing these challenges, stakeholders can harness AI’s potential to foster a more efficient and competitive financial ecosystem in India.

Stock price prediction remains a complex challenge in financial markets due to the dynamic interplay of economic indicators, global events, and investor sentiment. This study explores the integration of sentiment analysis into stock price forecasting using a FinBERT-LSTM model. By leveraging financial news data and market indicators, we aimed to enhance predictive accuracy. Sentiment features, such as sentiment intensity and daily sentiment ratios, were extracted using the FinBERT model and combined with traditional market data in an LSTM framework. Comparative analysis demonstrated that the sentiment-enhanced model significantly outperformed the baseline LSTM model, particularly during periods of high market volatility. These findings highlight the critical role of sentiment in market dynamics, providing a foundation for more robust predictive models. Future research directions include the incorporation of additional sentiment sources and advanced model architectures to further improve performance and adaptability in diverse market conditions.

This research explored how artificial intelligence (AI) (and behavioral insights) could be used to improve the process of determining investments in the stock market. The study used mixed-method research design to analyze data containing historical market data and sentiment data collected in social media and financial news and assessing predictive models, such as Long Short-Term Memory (LSTM) networks, random forests, and gradient boosting machines. The findings showed that the AI-driven models were highly effective in comparison to conventional methods in prediction of stock price variations and that LSTM models are the most accurate. The addition of sentiment analysis also enhanced the level of prediction in all models, which shows substantial importance of investor sentiment and market action in determining the price of an asset. Moreover, AI models that combined behavioral knowledge attained better risk neutral returns, as well as, lower portfolio volatility. In spite of these benefits, issues surrounding model interpretability, data privacy, and regulatory compliance remained the problem, with the overall lack of significant adoption. Findings of the case study have also identified that major investment firms had high implementation costs besides finding it difficult to balance predictive power and transparency. The research has come to the conclusion that, although AI has a transformative nature in investment strategies, ethics and regulatory compliance should be at the forefront of focus. Future study in these aspects should include further interpretation of explainable and fair AI framework, tests of cross-market soundness, and further placements of viewpoints of behavioral finance to enhance that of sustainable and responsible investment.