{"paper_id":"2611.23057","title":"Integrating market sentiments for stock price prediction: A comparative study of Bi-LSTM and multilayer perceptions","authors":["Ahmet Akusta"],"paper_date":"2026-11-17","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2611.23057","page_url":"https://thequant.space/flowcharts/integrating-market-sentiments-for-stock-price-prediction-a-comparative-study-of/","scores":{"empirical_rigor":6,"math_complexity":4,"quadrant":"Street Traders"},"proposes":{"status":"available","statement":"The paper proposes integrating sentiment scores from financial news headlines, derived using FinBERT and TextBlob, into predictive frameworks based on bidirectional LSTM and multilayer perceptron models for Nvidia stock price forecasting.","quote":"This study investigates the integration of sentiment analysis with machine learning models to forecast stock price movements using the Nvidia Corporation as a case study. These scores were integrated into predictive frameworks based on bidirectional long short-term memory (Bi-LSTM) networks and multilayer perceptrons (MLPs) developed alongside historical stock price data.","summary":"This study investigates the integration of sentiment analysis with machine learning models to forecast stock price movements using the Nvidia Corporation as a case study. Sentiment scores were derived from Nvidia-related financial news headlines using two sentiment analysis tools: FinBERT, a…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"An 80/20 train‑test split is used for model evaluation.","type":"data","stated":"explicit","quote":"The train,test split was performed using an 80/20 ratio.","support":"full"},{"assumption":"Shuffling is applied to the time‑series data while preserving temporal relationships.","type":"model","stated":"explicit","quote":"Because the data is a time series dataset, shuffling was applied to ensure randomness while preserving the integrity of the temporal relationships within the dataset.","support":"full"},{"assumption":"Market conditions can be categorized into bullish, stagnation, and strong bullish phases for phase‑specific analysis.","type":"market","stated":"explicit","quote":"This study assesses the predictive performance over the entire observation period and across distinct market phases: bullish, stagnation, and strong bullish conditions. Akusta, 2025 Bi,LSTM (Market Phases): Bi,LSTM models are independently trained on data from each defined phase to investigate the impact of different market conditions.","support":"full"},{"assumption":"Sentiment scores derived from news headlines are relevant predictors of stock price movements.","type":"statistical","stated":"explicit","quote":"Feature engineering In this study, we derived sentiment scores from financial news headlines to integrate market sentiments into Nvidia’s stock price predictions. Sentiment analysis of the collected news headlines was conducted using two distinct approaches: TextBlob and FinBERT.","support":"full"},{"assumption":"Historical price data and sentiment features are sufficient inputs for the Bi‑LSTM and MLP models without additional macro or order‑book information.","type":"data","stated":"implied","quote":"The first dataset, obtained from Yahoo Finance, comprises 390 daily historical stock price records for Nvidia Corporation, encompassing metrics, such as opening, closing, highest, and lowest prices, and trading volume. The second dataset includes 9,400 news headlines about Nvidia during the same timeframe, measuring market sentiment toward the company.","support":"full"},{"assumption":"The models are evaluated on a single stock (Nvidia) and results are assumed to reflect the effect of sentiment integration for that asset.","type":"other","stated":"implied","quote":"This study investigates the integration of sentiment analysis with machine learning models to forecast stock price movements using the Nvidia Corporation as a case study. The first dataset, obtained from Yahoo Finance, comprises 390 daily historical stock price records for Nvidia Corporation, encompassing metrics, such as opening, closing, highest, and lowest prices, and trading volume.","support":"full"}]},"datasets":{"status":"available","items":[{"name":"Nvidia daily historical stock price records","provider":"Yahoo Finance","kind":"market-prices","access":"public","period":null,"universe":"Nvidia Corporation","quote":"The first dataset, obtained from Yahoo Finance, comprises 390 daily historical stock price records for Nvidia Corporation, encompassing metrics, such as opening, closing, highest, and lowest prices, and trading volume.","support":"full","known_id":"yahoo-finance"},{"name":"Nvidia news headlines","provider":"Nasdaq.com","kind":"text-news","access":"public","period":null,"universe":"Nvidia Corporation","quote":"The second dataset includes 9,400 news headlines about Nvidia during the same timeframe, measuring market sentiment toward the company. These data were obtained from Nasdaq.com.","support":"full"}],"known":[{"id":"yahoo-finance","name":"Yahoo Finance","provider":"Yahoo","kind":"market-prices","access":"public","quote":"The first dataset, obtained from Yahoo Finance, comprises 390 daily historical stock price records for Nvidia Corporation, encompassing metrics, such as opening, closing, highest, and lowest prices, and trading volume.","mentions":1,"used_in":1}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"daily"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R1","candidates":{"front_matter":"n/a","evidence_card":"no","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"forecast-accuracy","findings":[{"finding":"Sentiment features improve predictive accuracy during stagnation phases, with TextBlob outperforming FinBERT in those contexts.","kind":"predictive-accuracy","quote":"The findings indicate that sentiment features enhance predictive accuracy in specific contexts, particularly during stagnation phases, with TextBlob demonstrating superior performance to FinBERT in specific scenarios. In contrast, sentiment features gained prominence during low,volatility or stagnation phases, indicating the context,dependent value of both specialized (FinBERT) and general,purpose (TextBlob)","support":"full"},{"finding":"Bi‑LSTM models consistently achieve superior performance over MLPs in capturing temporal dependencies.","kind":"performance","quote":"In addition, Bi-LSTM models exhibit consistently superior performance in capturing temporal dependencies compared to MLPs. In this study, Bi,LSTM models generally outperformed MLPs in terms of capturing temporal dependencies and sequential patterns, particularly during the Stagnation Phase and the overall period.","support":"full"},{"finding":"TextBlob sentiment scores yield the lowest MAE, MSE, and RMSE among all model configurations.","kind":"performance","quote":"TextBlob scores led to the lowest MAE, MSE, and RMSE, along with a higher R² Score.","support":"full"},{"finding":"Feature importance analysis shows that the daily closing price is the dominant predictor across all market conditions.","kind":"statistical","quote":"Additionally, the feature importance findings reveal that “Daily Close” consistently remains the single most influential predictor, especially when market momentum is strong. Feature importance analysis revealed that “Daily Close” remained the dominant predictor across all market condi, tions, particularly during strong directional trends.","support":"full"},{"finding":"A large train‑test MAE gap (e.g., Model 3: 1.60 vs. 30.19) indicates potential overfitting, though test R² remains high (0.93).","kind":"negative","quote":"The substantial discrepancy between the train and test MAE (e.g., Model 3: 1.60 vs. 30.19) raises concerns about potential overfitting; however, the maintained high test R² (0.93) suggests that while generalization may be limited, the model retains explanatory power on unseen data.","support":"full"}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"available","supports":[{"paper":{"paper_id":"2402.06698","title":"FNSPID: A Comprehensive Financial News Dataset in Time Series","paper_date":"2024-02-09","authors":["Zihan Dong","Xinyu Fan","Zhiyuan Peng"],"empirical_rigor":8.5,"page_url":"https://thequant.space/flowcharts/fnspid-a-comprehensive-financial-news-dataset-in-time-serie/"},"question":"Does incorporating sentiment scores improve stock price prediction performance?","this_paper_finding":"TextBlob sentiment scores yield the lowest MAE, MSE, and RMSE among all model configurations.","other_paper_finding":"Adding sentiment scores modestly enhances performance on the transformer-based model.","other_paper_quote":"adding sentiment scores modestly enhances performance on the transformer-based model","note":"Finding A reports a modest performance boost for a transformer model, and Finding B shows TextBlob sentiment yields the lowest MAE, MSE, and RMSE, supporting A.","confidence":0.595}],"contradicts":[],"qualifies":[],"compared_with":32,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":false,"code_url":null,"note":"No repository presented by the authors as their own was found in the paper. Absence here is not proof that no code exists."},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":3,"extracted_at":"2026-11-18T11:39:11Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}