-
Notifications
You must be signed in to change notification settings - Fork 0
Baselines
What other readily available option would a company or private customer have to do twitter sentiment analysis?
Based on (apparently) first paper on twitter sentiment analysis by some standford people. One needs to register and can then send a .txt file to be classified from the command line.
Should be one of the better available API's for general sentiment analysis. Hard to gauge how much time it would take to score our test dataset with it. Probably boils down to creating a google cloud account and writing a short script.
I think it's the newer Sentiment 140. Looks pretty nice.
What other approaches have been proposed in the literature?
Compute glove embeddings for each token, average tokens of one tweet. Classify the obtained feature vector using logistic regression, SVM. Cloud probably be done in flair to reduce work and possibility of bugs.
As a "harder" baseline we could use the simple pretrained bert classifier. Would distinguish more clearly between our implemented improvements and the existing state of the art.
Sources:
- http://help.sentiment140.com/home
- https://cs.stanford.edu/people/alecmgo/papers/TwitterDistantSupervision09.pdf
- https://cloud.google.com/natural-language/docs/sentiment-tutorial
- http://www.da.inf.ethz.ch/teaching/2020/CIL/files/exercises/exercise06.pdf
- https://monkeylearn.com/blog/sentiment-analysis-of-twitter/
- https://huggingface.co/transformers/model_doc/bert.html#bertforsequenceclassification