I have to evaluate a sentiment lexicon but most of the sentiment lexicon are tested and evaluated on some datasets of sentiment analysis. I want to know about the most suitable metrics for lexicon evaluation
You could start with getting familiar how was it done in the lexicon era (I have a feeling it's coming back, e.g. via distant-supervision approaches, and I am looking forward to using your lexicon if it's planned to be published). For example reading Related Work section in the classic "Using SentiWordNet for multilingual sentiment analysis" by Denecke. If I understand correctly you already have a set of words in various languages, so you could modify the translation approach and use both translated evaluation text and words/phrases from the lexicon.
- Badges
- Science topic