| | --- |
| | language: |
| | - en |
| | license: cc-by-nc-4.0 |
| | size_categories: |
| | - 10M<n<100M |
| | task_categories: |
| | - text-generation |
| | pretty_name: Distributed Ledger Technology (DLT) / Blockchain Tweets |
| | dataset_info: |
| | features: |
| | - name: timestamp |
| | dtype: large_string |
| | - name: tweet |
| | dtype: large_string |
| | - name: language |
| | dtype: string |
| | - name: total_tokens |
| | dtype: int64 |
| | - name: sentiment_class |
| | list: |
| | - name: label |
| | dtype: string |
| | - name: score |
| | dtype: float64 |
| | - name: sentiment_label |
| | dtype: string |
| | - name: sentiment_score |
| | dtype: float64 |
| | - name: confidence_level |
| | dtype: string |
| | - name: year |
| | dtype: string |
| | splits: |
| | - name: train |
| | num_bytes: 6772592588.26901 |
| | num_examples: 22033090 |
| | download_size: 3584763878 |
| | dataset_size: 6772592588.26901 |
| | configs: |
| | - config_name: default |
| | data_files: |
| | - split: train |
| | path: data/train-* |
| | tags: |
| | - Blockchain |
| | - DLT |
| | - Cryptocurrencies |
| | - Bitcoin |
| | - Ethereum |
| | - Crypto |
| | --- |
| | |
| | # DLT-Tweets |
| |
|
| | <p align="center"> |
| | <a href="https://huggingface.co/papers/2602.22045"><b>[Paper]</b></a> • |
| | <a href="https://github.com/dlt-science/DLT-Corpus"><b>[Code]</b></a> |
| | </p> |
| |
|
| | ## Dataset Description |
| |
|
| | ### Dataset Summary |
| |
|
| | DLT-Tweets is a large-scale corpus of social media posts related to Distributed Ledger Technology (DLT). This dataset is part of the larger DLT-Corpus collection, designed to support NLP research, social computing studies, and public discourse analysis in the DLT domain. It was introduced in the paper [DLT-Corpus: A Large-Scale Text Collection for the Distributed Ledger Technology Domain](https://huggingface.co/papers/2602.22045). |
| |
|
| | The dataset contains **22.03 million social media documents** with **1,120 million tokens** (1.12 billion tokens), spanning posts from **2013 to mid-2023**. All documents are in English and have been processed to protect user privacy. |
| |
|
| | This dataset is part of the DLT-Corpus collection. For the complete corpus including scientific literature and patent data, see: https://huggingface.co/collections/ExponentialScience/dlt-corpus-68e44e40d4e7a3bd7a224402 |
| |
|
| | ### Languages |
| |
|
| | English (en) |
| |
|
| | ## Dataset Structure |
| |
|
| | ### Data Fields |
| |
|
| | Each post in the dataset contains the following fields: |
| |
|
| | - **tweet**: Full text content of the social media post |
| | - **timestamp**: Date and time when the post was created |
| | - **year**: Year the post was created |
| | - **language**: Language code (all entries are 'en' for English) |
| | - **total_tokens**: Total number of tokens in the tweet |
| | - **sentiment_class**: Classified sentiment category (e.g., positive, negative, neutral) |
| | - **sentiment_label**: Detailed sentiment label |
| | - **sentiment_score**: Numerical sentiment score |
| | - **confidence_level**: Confidence level of the sentiment classification |
| | |
| | **Privacy Protection:** All usernames have been removed from the tweet text to protect user privacy. |
| | |
| | ### Data Splits |
| | |
| | This is a single corpus without predefined splits. Users should create their own train/validation/test splits based on their specific research needs. Consider temporal splits for time-series analyses or studying how discourse evolves over time. |
| | |
| | ## Dataset Creation |
| | |
| | ### Curation Rationale |
| | |
| | DLT-Tweets was created to address the lack of large-scale social media corpora for studying public discourse, sentiment, and information diffusion in the Distributed Ledger Technology domain. Social media data provides unique insights into: |
| | |
| | - Public perception and sentiment toward DLT technologies |
| | - Real-time reactions to DLT events (market movements, hacks, regulations) |
| | - Information spreading patterns and viral content |
| | - Community dynamics and influencer networks |
| | - The gap between technical development and public understanding |
| | |
| | ### Source Data |
| | |
| | #### Data Collection |
| | |
| | Social media posts were aggregated from: |
| | 1. **Previously published academic datasets** focused on cryptocurrency and blockchain topics |
| | 2. **Publicly available industry sources** that collected DLT-related social media data |
| | |
| | All data was collected **before Twitter/X's 2023 API restrictions** that limited academic research access. The collection complies with platform Terms of Service that were in effect at the time of collection, which explicitly permitted academic research use. |
| | |
| | #### Data Processing |
| | |
| | The collection process involved: |
| | |
| | 1. **Aggregation**: Combining multiple sources while tracking provenance |
| | 2. **Username removal**: Removing all @username mentions to protect privacy |
| | 3. **Duplicate detection**: Identifying and removing duplicate posts |
| | 4. **Language filtering**: Filtering for English-language posts using language detection |
| | 5. **Sentiment analysis**: Adding sentiment labels using automated classification |
| | 6. **Quality filtering**: Removing extremely short posts |
| | 7. **Privacy protection**: Ensuring no identifying information remains |
| | |
| | ### Annotations |
| | |
| | The dataset includes automated sentiment annotations generated using sentiment analysis models. These provide: |
| | - Sentiment class (positive, negative, neutral) |
| | - Sentiment scores and confidence levels |
| | |
| | #### Annotation Process |
| | |
| | Sentiment was determined using automated sentiment analysis models trained on social media text. |
| | |
| | ### Personal and Sensitive Information |
| | |
| | **Privacy Measures:** |
| | |
| | - **All usernames have been removed** from the text content |
| | - No profile information, user IDs, or biographical data is included |
| | - Only the text content and basic engagement metrics are retained |
| | - Posts are not linkable back to specific individuals |
| | |
| | **Residual Considerations:** |
| | |
| | - Some posts may contain self-disclosed information in the text itself |
| | - Famous individuals or organizations might be identifiable through context |
| | - Users should not attempt to re-identify individuals from this dataset |
| | |
| | ## Considerations for Using the Data |
| | |
| | ### Social Impact of Dataset |
| | |
| | This dataset can enable: |
| | |
| | - **Positive impacts**: Understanding public discourse, detecting misinformation, studying information diffusion, analyzing sentiment trends, improving public communication about DLT |
| | - **Potential negative impacts**: Could be misused for market manipulation, creating targeted misinformation campaigns, or developing manipulative trading systems |
| | |
| | **Researchers should implement appropriate safeguards when working with this data.** |
| | |
| | ### Discussion of Biases |
| | |
| | Potential biases include: |
| | |
| | - **Platform bias**: Only Twitter/X data is included; other social platforms are not represented |
| | - **User bias**: Social media users are not representative of the general population |
| | - **Language bias**: Only English-language posts are included |
| | - **Temporal bias**: More recent years have more posts due to platform growth and increased DLT interest |
| | - **Topic bias**: Certain events (market crashes, hacks) may generate disproportionate discussion |
| | - **Bot bias**: Despite filtering, some bot-generated content may remain |
| | - **Geographic bias**: English-speaking regions are over-represented |
| | |
| | ### Other Known Limitations |
| | |
| | - **Temporal gap**: No posts after mid-2023 due to platform API restrictions |
| | - **Context loss**: Username removal eliminates ability to analyze user behavior or influence |
| | - **Incomplete threads**: Some conversation threads may be incomplete due to filtering |
| | - **Sarcasm and irony**: Social media posts often use language that is difficult for NLP models to interpret |
| | - **Misinformation**: Dataset may contain false or misleading claims about DLT |
| | - **Market sensitivity**: Discussions may reflect market manipulation or coordinated campaigns |
| | - **Evolving terminology**: DLT terminology evolves rapidly; older posts may use outdated terms |
| | |
| | ## Additional Information |
| | |
| | ### Dataset Curators |
| | |
| | Walter Hernandez Cruz, Peter Devine, Nikhil Vadgama, Paolo Tasca, Jiahua Xu |
| | |
| | ### Licensing Information |
| | |
| | **CC-BY-NC 4.0** (Creative Commons Attribution-NonCommercial 4.0 International) |
| | |
| | This dataset is released under CC-BY-NC 4.0 for **research purposes only**. |
| | |
| | **Key terms:** |
| | - **Attribution required**: You must give appropriate credit to the dataset creators |
| | - **Non-commercial use**: Commercial use is not permitted under this license |
| | - **Academic research**: The dataset is intended for academic and non-profit research |
| | |
| | **Legal basis:** Data was collected before changes in Twitter/X's Terms of Service in 2023, under terms that explicitly permitted academic research use. See: |
| | - https://x.com/en/tos/previous/version_18 |
| | - https://x.com/en/tos/previous/version_17 |
| | |
| | For more information on CC-BY-NC 4.0, see: https://creativecommons.org/licenses/by-nc/4.0/ |
| | |
| | ### Citation Information |
| | |
| | ```bibtex |
| | @misc{hernandez2026dlt-corpus, |
| | title={DLT-Corpus: A Large-Scale Text Collection for the Distributed Ledger Technology Domain}, |
| | author={Walter Hernandez Cruz and Peter Devine and Nikhil Vadgama and Paolo Tasca and Jiahua Xu}, |
| | year={2026}, |
| | eprint={2602.22045}, |
| | archivePrefix={arXiv}, |
| | primaryClass={cs.CL}, |
| | url={https://arxiv.org/abs/2602.22045}, |
| | } |
| | ``` |
