Why does the sentiment analysis classified a negative text as positive?

AmosGHAmosGH Member Posts: 7 Learner I
edited October 2019 in Help
I run some sentiment analysis but it gave me wrong analysis in the sense that a text that supposes to be negative was classified as positive. What accounts for this misinterpretation and what can be done to rectify it. Again how does the sentiment analysis from Aylien classified a text to be negative, positive and neutral?


  • Options
    tkaratasiostkaratasios Member Posts: 6 Contributor II
    Can someone guide me with some problems I have about performing a sentiment analysis? Reading all the previews I see that there are several extentions about rapidminer. Having downloaded these extensions means that they work simultaneously or do I have to choose the prefered one? I see the same problem as the AmosGH
    How do I perform this? In RapidMiner : 
     - Extract Sentiment operator from the Toolbox extension (to install from the MarketPlace)
    Can these two be added in rapidminer? 
     - TextBlob
     - NLTK (Natural Language ToolKit)

    Thank you for your time and forgive me for the number of questions I did
  • Options
    Pavithra_RaoPavithra_Rao Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 123 RM Data Scientist
    Hi @tkaratasios,

    Here's quick tutorial videos on how to use operators (as part of extensions) and build the workflow in RapidMiner.


    Also, each operator comes with sample tutorial processess to learn more about how to use a specific operator.

    Hope this helps!

  • Options
    tkaratasiostkaratasios Member Posts: 6 Contributor II
    edited February 2020
    Thank you Pavithra for the provided help. I will see your recomendations but I also have another question which need some help
    When I conduct a sentiment analysis using a phrase which would be estimated as positive instead of that rapidminer characterized it as negative ?
  • Options
    kaymankayman Member Posts: 662 Unicorn
    Classic sentiment analysis can be tricky as it (oversimplified) sums up negative and positive indicators and looks at the one who gets the highest count.

    So it will not recognize irony, sarcasm, tongue in cheek and all the other ways we humans are able to use nice words to make something sound bad or the other way around. 

    Apart from that the training data is very important, most if not all providers from commercial solutions trained their models on public and relatively generic data. So if your content does not match this well (because for instance you have a rather specific domain) your 'obvious giveaways' may not have been in the original data set and are therefore ignored for yours.

    So this leaves you with the option to train your own, and if you have enough data to train you can get pretty good results (but never perfect, blame our language flexibility for that...) 

    I personally like the Vader fork for NLTK, gave me good results and was rather easy to implement
  • Options
    tkaratasiostkaratasios Member Posts: 6 Contributor II
    edited February 2020
    Thank you for the support Kayman. Please give some details about your suggestion. Refering to Vader means that it can be applied in rapidminer? Moreover how can I do some training to the program?
Sign In or Register to comment.