Asset-Specific Sentiment Inversions in Financial NLP
Standard sentiment models assign polarity to text in isolation. In financial NLP, that's often wrong. The same phrase can be bullish for one asset and bearish for another. swik catalogs these systematic inversions โ instances of context-dependent polarity โ so inference engines can correct for them. This is the core challenge of Aspect-Based Sentiment Analysis (ABSA) applied to financial markets.
The Inversion Problem
A sentiment inversion occurs when the naive polarity of a financial headline differs from the actual directional impact on a specific asset's price. These are not random errors โ they are systematic inversions rooted in supply chain relationships, macro linkages, and domain knowledge that general language models don't encode.
Solving this requires context-dependent polarity: the same phrase scored differently depending on which asset you're analyzing. This is the financial NLP equivalent of ABSA โ and it's why asset-specific sentiment signal correction matters for any serious application in trading or quantitative research.
Active inversion catalog
Live data from the swik community catalog. Updated as entries are confirmed.
How swik corrects for inversions
swik uses a two-layer inference architecture. A base language model provides an initial sentiment reading. The swik inversion catalog then applies asset-specific sentiment signal correction โ overriding the naive reading when a known phrase inversion applies. This is what makes swik an asset-specific news analyzer rather than a generic sentiment classifier.
The catalog is community-maintained and open. Every inversion entry includes the phrase, naive polarity, actual directional impact, and economic reasoning. Entries move from hypothesis to active status through community voting: minimum 3 confirmations, 2:1 confirm-to-reject ratio required.
Open data
The full inversion catalog is released under Creative Commons Attribution 4.0 and available on GitHub and HuggingFace. Use it for fine-tuning models, cross-asset spillover analysis, backtesting sentiment signals, or building your own asset-specific sentiment APIs.