I am looking for a dataset, where I could use NLP techniques to estimate target value using regression. For example, I could be give a few sentences that describe an accident, and the target value would be the cost of an accident. Kaggle has quite a few datasets for NLP, but they all, as far as I can see, for classification.
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
I know of a good few NLP regression tasks, but these are usually not about a predicting an extrinsic value from the semantic description, but rather predict some numerical property of the text itself:
- MT quality estimation: predict the quality of a translated sentence, usually as a score. Data available at the WMT QE Shared Task series. There are probably other similar datasets about predicting the quality of a sentence in general.
- Author profiling: predicting the age of the author of a text. Some datasets exist in particular through the PAN workshop series.
- Predicting a rating based on a comment - there's data everywhere on social media, Amazon etc. You could even use StackExchange posts and predict votes :)
I'm not aware of any data similar to your example (btw it would be a very challenging task). I know that there are some works on detecting fraud from the text of an insurance claim, so maybe there are datasets that you could use in this area.
