Where are the domain-specific LoRA fine-tunes for financial sentiment analysis \u2014 one adapter per asset (OIL, GOLD, COFFEE, BTC, EUR\/USD, etc.)?<\/p>\n
The problem: no labeled dataset exists that’s asset-specific. Generic FinBERT doesn’t know that “OPEC cuts production” is bearish for oil. So I built one. <\/p>\n
The pipeline: <\/p>\n
~17,500 headlines collected across 35+ securities from RSS, Google News, GDELT, YouTube transcripts, and FMP. <\/p>\n
Claude Haiku pre-labels everything with asset-specific context (known inversions, price drivers). Humans review and override. <\/p>\n
Why per-asset matters: <\/p>\n
Because standard sentiment models like FinBERT treat “Fed raises rates” as bearish across the board. <\/p>\n
Or “rising dollar boosts USD index to 3-month high” \u2192 <\/p>\n
FinBERT: bullish. In the actual gold market this is bearish<\/p>\n
Or “OPEC increases production” is it nice for your OIL Futures?
\u2022 FinBERT sees “increases”, “production up” \u2192 bullish (more output = growth = good)
\u2022 Actual oil market \u2192 bearish (more supply = price drops)<\/p>\n
Labeling methodology:<\/p>\n
\u2022 4 classes: bullish \/ bearish \/ neutral \/ irrelevant (per asset, not generic)
\u2022 AI seed labels \u2192 human consensus \u2192 LoRA training data
\u2022 Target: ~500 human consensus labels per security before fine-tuning <\/p>\n
What’s going on HuggingFace: <\/p>\n
\u2022 Inversion catalog already live: polibert\/sentimentwiki-catalog
\u2022 Labeled dataset + LoRA adapters: uploading as each security hits threshold
\u2022 First uploads: OIL, GOLD, EUR\/USD (most labeled) <\/p>\n
Data sources that actually work (and a few that don’t): <\/p>\n
Works: OilPrice RSS, FXStreet, CoinDesk, GDELT, YouTube (Bloomberg\/Reuters\/Kitco), FMP (only paid one)
Doesn’t: S&P Global Platts (paywalled), USDA AMS (PDFs only), ICO coffee (Cloudflare-blocked) <\/p>\n
If you work in financial NLP and want to contribute labels or suggest assets: sentimentwiki.io (http:\/\/sentimentwiki.io\/<\/a>) \u2014 contributions welcome<\/p>\n<\/div>\n submitted by \/u\/Poli-Bert <\/a>
[link]<\/a><\/span> [comments]<\/a><\/span><\/p>
