Instrumetriq/crypto-market-sentiment-observations

Instrumetriq — Crypto Market Activity & Sentiment Context Dataset

Time-aligned observational snapshots of crypto market activity and social sentiment across 270+ assets, designed to contextualize market structure, liquidity, and attention dynamics.

Observational data only. No trading advice, predictions, or signal generation.

Dataset Description

This dataset provides weekly Sunday snapshots from Instrumetriq's continuous monitoring pipeline. Each snapshot captures synchronized market and sentiment observations for 270+ crypto assets.

This repository currently contains 14 weekly samples (2025-12-21 through 2026-03-22).

• Homepage: instrumetriq.com
• Repository: github.com/SiCkGFX/instrumetriq-public
• Interactive demo: Open in Google Colab
• Archival releases (DOI): doi.org/10.5281/zenodo.18508636

Data Collection

Spot market data - sourced from Binance via public REST API

• Mid prices, bid–ask spreads, liquidity percentiles
• Sampled at 10-second intervals, aggregated per observation window

Social sentiment data - sourced from X (Twitter)

• Posts are filtered for crypto relevance using a dedicated BERTweet-based classifier
• Relevant posts are scored by a two-model sentiment pipeline (BERTweet primary, DistilBERT referee with confidence calibration)
• Sentiment is exposed only in aggregated form (counts and averages per window)
• No raw posts, usernames, or personally identifiable information are included

Observation structure

• Each asset is monitored in ~2-hour rolling cycles (~120–130 minutes)
• One observation row per asset per cycle
• Approximately ~2,500 observations per day across all tiers

Methodology Versions

The sentiment pipeline was updated in February 2026. Records include a methodology_regime field for programmatic identification:

	V1 (before Feb 16, 2026)	V2 (after Feb 16, 2026)
Primary model	DistilBERT-based	BERTweet-based
Referee model	DistilBERT-based	DistilBERT-based (updated)
Relevance filter	None	BERTweet-based crypto filter
methodology_regime	"v1"	"v2"
sentiment_model_version	"v1.0"	"v2.0"

Cutover timestamps:

• Phase 1 (2026-02-16T05:14Z) - Updated sentiment models
• Phase 2 (2026-02-17T06:03Z) - Crypto relevance filter activated

Dataset Tiers

All tiers contain the same observations (same rows, same timestamps). They differ only in column depth.

Tier 1 - Explorer

• 19 flat columns (Parquet + CSV)
• Aggregated sentiment counts and averages
• Spot prices, spreads, liquidity, and quality scores
• Best for: dashboards, general analysis, lightweight inspection

Tier 2 - Analyst

• 8 top-level columns including nested structures
• Extends Tier 1 with detailed sentiment aggregates, author statistics, and engagement metrics
• Best for: behavioral analysis, cross-sectional studies

Tier 3 - Researcher

• 12 top-level columns including deeply nested structures
• Extends Tier 2 with 700+ spot price samples per observation window (10-second resolution)
• Multi-window sentiment, diagnostics, and futures-derived context
• Best for: research, validation, microstructure analysis

Note: High-frequency (10-second) spot price samples are available only in Tier 3.

File Structure

Each week is stored in a folder named by date:

week_YYYY-MM-DD/
  YYYY-MM-DD_tier1.parquet
  YYYY-MM-DD_tier1.csv
  YYYY-MM-DD_tier2.parquet
  YYYY-MM-DD_tier3.parquet

Tier 2 and Tier 3 are provided in Parquet only due to nested structures.

Schema documentation is in schema_documentation/.

Intended Uses

• Market structure research
• Behavioral and sentiment analysis
• Liquidity and execution context studies
• Exploratory and descriptive analytics
• Academic research and reproducibility

Limitations & Ethics

• Observational data only - not suitable for backtesting trading strategies or generating signals
• No individual social media posts or personal data are included
• All sentiment data is aggregated and anonymized
• Covers X (Twitter) only - no other social platforms
• No predictions, no trading advice, no implied causation

Citation

If you use this dataset in your research, please cite:

@dataset{instrumetriq_2026,
  author    = {Instrumetriq},
  title     = {Instrumetriq Crypto Market and Social Sentiment Dataset},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.18508636},
  url       = {https://doi.org/10.5281/zenodo.18508636}
}

Access & Subscriptions

Free weekly samples are available in this repository and on GitHub.

Full continuous access to all tiers is available via subscription at instrumetriq.com/access.