Pith. sign in

REVIEW 4 major objections 6 minor 20 references

Mining Tweets to Predict Future Bitcoin Price

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Tweets can predict Bitcoin's next-day price, paper argues

desk verdict A clearly written student project whose central claim—that tweets add predictive power beyond price history—is never tested, because no price-only baseline appears anywhere. read the letter →

arxiv 2412.02148 v1 pith:BOABJC55 submitted 2024-12-03 cs.AI

Ashutosh Hathidara , Gaurav Atavale , Suyash Chaudhary This is my paper · ORCID
classification cs.AI
keywords BitcoinpricepredictionTwittersentimentanalysistweetvolumerandomforestclassificationridgeregressioncryptocurrencysocialmediaanalyticstimeseriesforecasting
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that what people post about Bitcoin on Twitter carries predictive information about the cryptocurrency's future price. The authors aggregate roughly three years of tweets into daily features — tweet volume, likes, retweets, and counts of positive/negative/neutral sentiment — and use them, together with the previous day's price, to forecast next-day price and direction. Their best classifier, a random forest, reaches 62% accuracy and an F1 score of 0.75 on up/down movement, and ridge regression yields the lowest test error among the regression models. If the claim holds, social media activity could serve as a practical supplement to price-history forecasts for Bitcoin and other cryptocurrencies.

What carries the argument

The central mechanism is the day-level aggregation of tweet metadata into a fixed feature vector: daily tweet volume, counts of tweets exceeding like/retweet thresholds, counts of tweets by sentiment, one-hot encodings of day of week and hour, and the previous day's Bitcoin price. These features feed the regression and classification pipelines; the random forest classifier with balanced class weights and the ridge regressor are the best-performing models. The aggregation step is what converts noisy, irregular social media posts into a tabular time series usable by standard machine learning tools.

What would settle it

Retrain the best regression and classification models using only the previous-day Bitcoin price (plus day/hour one-hot features) and compare test MSE and accuracy against the tweet-inclusive models; if the price-only model matches or beats them, the claim that tweets add predictive signal is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that user-generated tweets affect Bitcoin price fluctuations, and that features extracted from day-level aggregation of tweets allow prediction of both the direction of movement and, to a degree, the exact price. To show this, the authors build a dataset of daily tweet statistics from 16M tweets, run sentiment analysis with two libraries, and train eight classifiers and several regressors. They report the random forest classifier as best for direction (62% accuracy, 0.75 F1) and ridge regression as best for price level (lowest test MSE). The authors acknowledge the price field's extreme variance yet conclude that tweet-derived features make the prediction task tractable.

Load-bearing premise

The load-bearing premise is that tweet-derived features add predictive power beyond the previous day's price alone; the paper never tests this by comparing against a price-only model, so its central claim is unverified.

Editorial extensions

If this is right

  • If tweet features truly improve Bitcoin price forecasts, monitoring Twitter volume and engagement in real time becomes a practical input to trading and risk-management systems.
  • The reported 62% direction accuracy suggests that even a modest edge over chance can be obtained from public social media data, though profitability depends on costs and timing.
  • Extending the same day-level feature pipeline to other cryptocurrencies is straightforward, provided hashtag variants are accounted for during data collection.
  • The finding that over 90% of tweets are neutral implies that sentiment polarity alone is a weak signal; tweet volume and engagement may be carrying the predictive weight.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the regression models include previous-day price and no price-only baseline is reported, the low test MSE may be entirely due to price autocorrelation; an ablation removing tweet features would isolate their contribution.
  • The causal language in the conclusion ("tweets affect the price") is not supported by the correlational design; a natural experiment around a scheduled event, such as a celebrity tweet, could test whether spikes in tweet activity precede or follow price moves.
  • The day-level aggregation discards intraday ordering; aggregating over shorter windows and testing lead-lag relationships could reveal whether tweets lead price changes or merely react to them.
  • The authors' own skepticism about clustering suggests that grouping users into price-affecting categories is not yet reliable; a more robust test would be to compute per-user tweet history and price movement association.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. This paper analyzes roughly 16 million Bitcoin-related tweets from 2016–2019, aggregates tweet volume, engagement, and sentiment features at daily level, and uses these together with lagged Bitcoin price to predict next-day price (regression) and price direction (classification). The authors report ridge regression as the best regression model and random forest as the best classifier (62% accuracy, 0.75 F1), and conclude that tweet features affect Bitcoin price. The manuscript is a concise project-style report that includes EDA, sentiment analysis, clustering, and standard supervised models.

Significance. Establishing that tweet-derived features carry incremental predictive information for Bitcoin price would be a valuable contribution to the literature on social media and financial markets. The paper's strengths are the use of a large public dataset (16M tweets) and a transparent description of the preprocessing pipeline. However, the central claim is not supported by the reported experiments: no baseline or ablation isolates tweet features from the lagged-price feature, and the authors themselves report no correlation between sentiment and price movement. The contribution therefore reduces to an application of standard ML methods without a demonstration of added value.

major comments (4)
  1. [§2.4, §3] The regression and classification models include the previous day's Bitcoin price as a feature (explicitly stated in §2.4: 'We will also include current-day bitcoin prices as part of this step'), yet no experiment is reported that trains or evaluates a model using price history alone. Since Bitcoin prices are strongly autocorrelated, the low test MSE and 62% classification accuracy could be entirely attributable to the lagged-price feature. The manuscript must include a price-only baseline and a feature ablation (e.g., tweet features only, price+tweet vs. price-only) to isolate the incremental predictive value of tweet-derived features. Without this, the conclusion in §4 that 'user related tweets affect the price fluctuations' does not follow from the reported results.
  2. [§3 (Sentiment Analysis)] The paper states 'We didn't find any correlation between sentiment and movement of price.' This null result directly undermines the causal claim in the Conclusion that tweets affect Bitcoin price fluctuations. The authors need to reconcile this contradiction, either by restricting their claim to tweet volume/engagement features, or by providing evidence for a mechanism linking tweet features to price. As written, the paper's own results argue against the central thesis.
  3. [§2.4–2.5] All predictive results are based on a single 90:10 temporal split, with about 110 days in the test set, and no rolling-window evaluation, multiple seeds, or confidence intervals are reported. Given the nonstationary price series (range $364–$19,497), the test period may correspond to a specific market regime, and the reported metrics may not be stable. The authors should report results across multiple temporal folds and include standard deviations or error bars to assess robustness.
  4. [Table 1] The classification evaluation reports accuracy, recall, precision, and F1 without a majority-class baseline. Several models (KNN, Naive Bayes, Kernel SVM, XGBoost) achieve recall near 1.0 with lower precision, indicating that they predict the majority class almost always. Without a baseline such as 'always predict up' and class-wise confusion matrices, the 62% accuracy of the random forest cannot be interpreted as evidence of predictive skill. The F1 of 0.75 is also balance-sensitive and should be accompanied by stratified metrics.
minor comments (6)
  1. [§2.1, §2.4] The feature description is inconsistent: §2.1 mentions 'price of bitcoin on the previous day' while §2.4 refers to 'current-day bitcoin prices.' Clarify the feature definition to avoid any look-ahead bias in the prediction setup.
  2. [§2.5] The list of classifiers includes 'Decision Tree Classification model' twice; the duplicate should be removed.
  3. [References] Reference [20], cited for logistic regression, is identical to reference [18] (artificial neural networks); the citation does not correspond to the intended method.
  4. [§3] The text says 'DSCAN' instead of 'DBSCAN' in the clustering results discussion.
  5. [§2.4] Typo: 'StandardScalar' should be 'StandardScaler'.
  6. [§3] The regression results are presented only graphically (Figure 5); the numeric MSE values for each model should be stated in the text or in a table for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No definitional or fitted-parameter circularity; the core weakness is a missing price-only baseline, which is an experimental gap rather than a circular step.

full rationale

I find no circular step in the paper. The methods are standard machine-learning pipelines with no formal derivation, and no equation defines a prediction target in terms of itself or in terms of a fitted parameter that is then relabeled as a prediction. Including "current-day bitcoin prices" as a feature for next-day price is an autoregressive modeling choice, not a definitional identity: the model could in principle fail on the chronological 90/10 test split, and no target value is copied directly from the training set into the test set. The paper contains no load-bearing self-citations, no imported uniqueness theorem, and no ansatz smuggled in through prior work. The paper's own statement that sentiment did not correlate with price movement and its skepticism about clustering results weaken the concluding causal claim, as does the absence of a price-only baseline or ablation isolating the incremental value of tweet-derived features. Those are real evidentiary and design limitations, but they are not circularity: the reported predictions do not reduce by construction to the inputs that generate them.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The paper relies on standard ML assumptions plus several hand-chosen modeling decisions (language filter, aggregation thresholds, K for clustering, hyperparameters) and an untested assumption that tweet features add predictive information beyond price history. There are no invented physical or conceptual entities.

free parameters (4)
  • Random forest hyperparameters = n_estimators=100, max_features='log2', min_samples_split=2, class_weight='balanced', criterion='gini'
    Chosen via GridSearch on the training data; reported as the best configuration. This selection is part of the model fitting process and is not independently validated.
  • K in K-means clustering = k=3
    Selected from an elbow curve of inertia and distortion, a heuristic choice with no external validation.
  • Tweet feature aggregation thresholds = likes >0, >10, >100, >1000; retweets >0, >100
    Hand-chosen binning thresholds for tweet engagement features, set without justification or sensitivity analysis.
  • Train/test split ratio = 90:10 temporal split
    Chosen by the authors; a single split is used with no repeated cross-validation or bootstrap, so results may depend heavily on this choice.
assumptions (4)
  • domain assumption English-language tweets are representative of the entire Bitcoin conversation
    The authors filter to the 80% of tweets detected as English by langdetect and use only those for all analysis, assuming they capture the predictive signal.
  • domain assumption TextBlob sentiment labels are sufficiently accurate
    The authors state TextBlob labeled tweets 'more accurately' than VADER but provide no quantitative validation of either tool on this dataset.
  • ad hoc to paper Day-level aggregation preserves predictive information
    All tweet features are aggregated to daily counts, discarding intraday structure and tweet-level context. This is a modeling choice with no justification that it retains the signal.
  • domain assumption Previous-day Bitcoin price is a legitimate predictive feature
    The models include price history as a feature, implicitly assuming autocorrelation is a valid signal, but no baseline isolates whether tweet features add to this signal.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mining Tweets to Predict Future Bitcoin Price." pith.science (2026). https://pith.science/paper/BOABJC55

@misc{pith2026241202148,
  author       = {Pith},
  title        = {Pith review of: Mining Tweets to Predict Future Bitcoin Price},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BOABJC55}},
  note         = {Machine review of arXiv:2412.02148}
}
read the original abstract

Bitcoin has increased investment interests in people during the last decade. We have seen an increase in the number of posts on social media platforms about cryptocurrency, especially Bitcoin. This project focuses on analyzing user tweet data in combination with Bitcoin price data to see the relevance between price fluctuations and the conversation between millions of people on Twitter. This study also exploits this relationship between user tweets and bitcoin prices to predict the future bitcoin price. We are utilizing novel techniques and methods to analyze the data and make price predictions.

Figures

Figures reproduced from arXiv: 2412.02148 by the authors.

Figure 1
Figure 1. Raw data extracted from Kaggle [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Results extracted from EDA 4 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Results extracted from Sentiment Analysis [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: K-Means clustering analysis that all 3 models linear regression, ridge regression, and lasso regression perform really well in predicting bitcoin price [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Results extracted from Regression Analysis [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Comparing ROC curve of different classification models [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 18 canonical work pages

  1. [5]

    Cryptocurrency Price Prediction Using Tweet Volumes and Sentiment Analysis.SMU Data Science Review, 1(3):1–, 12 2017

    Jethin Abraham, Daniel Higdon, John Nelson, and Juan Ibarra. Cryptocurrency Price Prediction Using Tweet Volumes and Sentiment Analysis.SMU Data Science Review, 1(3):1–, 12 2017

    work page 2017

  2. [1]

    Bitcoin tweets - 16m tweets, 11 2019

    work page 2019

  3. [3]

    Patodkar Vaibhavi N and I.R S. Sheikh. Twitter as a Corpus for Sentiment Analysis and Opinion Mining. IJARCCE, 5(12):320–322, 12 2016

    work page 2016

  4. [4]

    Johan Bollen, H

    J. Johan Bollen, H. Huina Mao, and X. Xiaojun Zeng. Twitter mood predicts the stock market.Journal of Computational Science, 2(1):1–8, 3 2011

    work page 2011

  5. [6]

    tweet-preprocessor, 5 2020

    work page 2020

  6. [7]

    vaderSentiment, 5 2020

    work page 2020

  7. [9]

    Hartigan J. A. and Wong M. A. Algorithm AS 136: A K-Means Clustering Algorithm.Applied Statistics, 28(1):100, 1979

    work page 1979

  8. [10]

    Hierarchical Clustering: Objective Functions and Algorithms.Proceedings of the Twenty-Ninth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 378–397, 1 2018

    Cohen V., Kanade V., Mallmann-Trenn F., and Mathieu C. Hierarchical Clustering: Objective Functions and Algorithms.Proceedings of the Twenty-Ninth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 378–397, 1 2018

    work page 2018

Show all 20 references
  1. [11]

    Martin Ester, Hans-Peter Kriegel, Jörg Sander, and Xiaowei Xu. A density-based algorithm for discovering clusters a density-based algorithm for discovering clusters in large spatial databases with noise.Knowledge Discovery and Data Mining, pages 226–231, 8 1996

    1996

  2. [12]

    Scikit-learn: Machine Learning in Python.Le Centre pour la Communication Scientifique Directe - HAL - memSIC, 1 2011

    Fabian Pedregosa, Gaël Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, Jake Vanderplas, Alexandre Passos, David Cournapeau, Matthieu Brucher, Matthieu Perrot, and Edouard Duchesna...

    2011

  3. [13]

    D., Montgomery D

    Sampson R., Assuncao andP. D., Montgomery D. C., and Peck E. A. Introduction to Linear Regression Analysis. Journal of the American Statistical Association, 88(421):383, 3 1993

    1993

  4. [14]

    Hoerl A. E. and Kennard R. W. Ridge Regression: Biased Estimation for Nonorthogonal Problems. Technometrics, 12(1):55–67, 2 1970

    1970

  5. [15]

    Regression Shrinkage and Selection Via the Lasso.Journal of the Royal Statistical Society: Series B (Methodological), 58(1):267–288, 1 1996

    Tibshirani R. Regression Shrinkage and Selection Via the Lasso.Journal of the Royal Statistical Society: Series B (Methodological), 58(1):267–288, 1 1996

    1996

  6. [16]

    D., Breiman L., Friedman J

    Gordon A. D., Breiman L., Friedman J. H., Olshen R. A., and Stone C. J. Classification and Regression Trees. Biometrics, 40(3):874, 9 1984

    1984

  7. [17]

    C., Sheridan R

    Svetnik V., Liaw A., Tong C., Culberson J. C., Sheridan R. P., and Feuston B. P. Random Forest: A Classification and Regression Tool for Compound Classification and QSAR Modeling.Journal of Chemical Information and Computer Sciences, 43(6):1947–1958, 11 2003

    1947

  8. [19]

    k-nearest neighbour classifiers: 2nd edition (with python examples)

    Padraig Cunningham and Sarah Jane Delany. k-nearest neighbour classifiers: 2nd edition (with python examples). CoRR, abs/2004.04523, 2020. 7 Mining Tweets to Predict Future Bitcoin Price

    2004 arXiv

  9. [20]

    Application of artificial neural networks and multiple linear regression on local bond stress equation of UHPC and reinforcing steel bars

    Amini Pishro A., Zhang S., Huang D., Xiong F., Li W., and Yang Q. Application of artificial neural networks and multiple linear regression on local bond stress equation of UHPC and reinforcing steel bars. Scientific Reports, 11(1), 7 2021

    2021

  10. [21]

    Bayes and naive bayes classifier.CoRR, abs/1404.0933, 2014

    Vikramkumar, Vijaykumar B, and Trilochan. Bayes and naive bayes classifier.CoRR, abs/1404.0933, 2014

    2014 arXiv

  11. [22]

    Kernel Support Vector Machines and Convolutional Neural Networks

    Jiang S., Hartley R., and Fernando B. Kernel Support Vector Machines and Convolutional Neural Networks. 2018 Digital Image Computing: Techniques and Applications (DICTA), 12 2018

    2018

  12. [23]

    Xgboost: A scalable tree boosting system.CoRR, abs/1603.02754, 2016

    Tianqi Chen and Carlos Guestrin. Xgboost: A scalable tree boosting system.CoRR, abs/1603.02754, 2016. 8

    2016 arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.