Hi Mirko,

Please, check Stanford website:

https://nlp.stanford.edu/sentiment/

https://stanfordnlp.github.io/CoreNLP/

Thanks

Osman

There is no best supervised algorithm for text classification. The performance of the classifier is measures based on the features that you can obtain from the text corpus. This is mainly depends on two methods. Corpus based and Dictionary based methos..

Try to focus more on collecting the features based on your requirement. Accuracy reflects the collected features strength.

The easiest model for sentiment analysis with an accepted result is using BOW concept .Then you can train a classifier of your choice which supports linear classifications. This is not SoA model, however, it works at a certain level without considering the semantic information of the sentence.  You might consider word2vec and deep learning techniques to achieve the SoA results.

The harmony search algorithm is the most efficient algorithm, try it.