Economic Twitter Narratives with LLM Analyses from 2007-2023

Two collections of economy-related X (Twitter) posts span 2007-2020 and 2021-2023, with LLM-generated analyses for the latter period. The data, curated using targeted keywords, supports research into macroeconomic narratives. The pre-pandemic dataset provides tweet IDs, while the post-2021 data includes corresponding LLM analyses.

Use Cases

• Compare economic narrative topics in pre-pandemic (2007-2020) and post-LLM cutoff (2021-2023) tweet ID collections.
• Analyze LLM-generated analyses of 2021-2023 posts to identify shifts in macroeconomic sentiment.
• Train models to detect narrative frames using the keyword-curated collections of economy-related posts.
• Study the evolution of economic discourse by linking tweet IDs from the two distinct time periods.

Strengths

• Covers a 17-year time range from 2007 to 2023, enabling longitudinal study.
• Provides two distinct collections for pre-pandemic and post-LLM training cutoff periods.
• Includes LLM-generated analyses for the 2021-2023 posts, adding an analytical layer.

Limitations

• The post-2021 dataset contains only LLM analyses, not the original tweet data due to platform restrictions.
• Data scope is limited to posts identified by targeted keywords, not the full platform discourse.
• The pre-pandemic collection consists only of tweet IDs, requiring hydration which may have data loss.

Provenance

Source

X (formerly Twitter) platform, curated by researchers.

Collection Method

Posts collected using targeted keywords; analyses generated by LLMs.

Time Range

2007-2020 (pre-pandemic) and 2021-2023 (post LLM training cutoff).

Freshness

Data last updated in February 2026, with coverage through 2023.

Geography

null