Sentiment Analysis - Better performance with Naive Bayes model

m_moertlm_moertl Member Posts: 11 Contributor I
edited June 2019 in Help

Hey there!


Just a general question. I'm trying to do SentimentAnalysis with Naive Bayes. For that i build a trainingset with about 2.5k facebookposts and gave them their polartiy manuely. When i run my process now, i get an accurency of about 44.7%. My question now is, how can i get a better performance?


I know, more or less, how Naive Bayes works. But i can't figure out how the Perfomance operator calculate it.


Thanks and kind regards



  • Options
    Thomas_OttThomas_Ott RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 1,761 Unicorn

    Hi @m_moertl,


    So getting better accuracy in sentiment models requires a bit more thinking. The main culprits, IMHO, is usually around the text processing. You may be destroying a lot of information if you just use the default 'non letters' tokenization. You might want to use 'specify characters' instead and choose what you are splitting on yourself. This is incredibly handy when dealing with Twitter Content and I speak about it in this video here: https://youtu.be/ia2iV5Ws3zo (give it Like and Subscribe!).


    Other reasons could include how much you prune or not prune and additional text processing setups. Check all that first to make sure you are not losing valuable information. 


    Another possible reason is that Naive Bayes might not be a good algo to use You should try a LinearSVM or even a Deep Learning algo.  


    If you do all this, you should in theory get a bump in your Accuracy. 

Sign In or Register to comment.