01-29-2017 02:34 PM - edited 01-29-2017 02:37 PM
I am new user of RapidMiner. I am following some tutorial from rapidminer users in youtube to build a twitter sentiment analysis process. It was run well but the result was not really good. For example, when I used decision tree, it will give all positive sentiment (no negative or neutral). When I used SVM, the results will all negative. My trainning data was mixed positive and negative. Can anaybody advice. Thanks.
02-01-2017 03:51 PM
It's very difficult to say without knowing the details of the data and looking at the model output. It is possible that your data does not contain sufficient attributes with predictive power to generate a differentiated model. So, for example, your decision tree may not have any splits at all and just contains the root class, which simply predicts the majority class.
02-01-2017 04:47 PM
So this highlights the issue that some algorithms are better than others for a particular task. If this is a standard binonmal (2 labels) classification task, you could try using the ROC operator and embed different algorithms in to do a "model bakeoff" and see which one gives you the best general AUC.