Hi everybody,

I am looking for a sentiment analysis tool with high accuracy especially focused on short text such as tweets (but it's not mandatory). I searched a bit and I found a Quora question: https://www.quora.com/What-are-the-most-powerful-open-source-sentiment-analysis-tools

From their answers, I presume that there is no open source highly accurate system.

However, I haven't tried couple of promising systems yet:

http://www.sentimentanalysisonline.com/ http://tationem.com:8080/TexSIE/#working http://help.sentiment140.com/api

If you've used these tools above, I appreciate if you give some feedback.

I tried a bunch of tools (some of them are not open source software) such as: Semantria (https://semantria.com/api), Stanford CoreNLP toolbox, Alchemist (http://goo.gl/NpQ8jy) and RapidMiner.

I tested them on a twitter dataset that I manually labeled with 3 sentiment classes (neutral, positive, negative). These are the scores: http://goo.gl/YH1oJq (I haven't add RapidMiner yet, sorry).

AlchemyAPI seems arguably the best among them but the performance is not well especially when predicting neutral tweets.

For the next step, I would like to evaluate their score on SemEval Sentimental Analysis dataset (http://alt.qcri.org/semeval2015/task10/) but I expect that the trend will be same. I also examined the best system's paper of SemEval 2014 SA task: http://goo.gl/OYjqux. I haven't replicated it and evaluated its score on the same data yet, though.

I am a bit sceptical to use an off-the-shelf solution since I can't optimise them or I can't shape it for my needs. I would like to start with an open source tool and improve it pre-processing/post-processing or introduce some features and so on.

Sentiment analysis is very popular subject not only for research but also for market. So I'll be very happy to see some recommendations, ideas or directions.

Sorry for such a long post.

Thank you