Just want to understand the limitation of KNIME text processing.
Can KNIME classify the sentiment (+ve/-ve) based on context. Here i mean context is based on some political decision - Suppose i want to cluster two group of people opinion, one who favors the company current policies and other who doesnt. Customer express their likeness and dislikeness with more or less same word.
Suppose group A in favor of current policy and they dont express any happiness towards current policy because they are happy i.e. they do not lodge any complain or praied the current policy.
Group B who is against of current policy and they express dislikeness towards current policy and they lodge complaint agaist current policy and they express their emotion by text as follows.
"This is disgusting i am not going to give you my business and want to cancel my contract ASAP".
if you revert your policy in favor of group A then they express the same concerns as group B
"This is very pathetic step , you should keep yourself away from such polices , if you dont stay away from such policy then i am going to terminate my contract from your service"
How can we cluster those type of customer cause the policies had been changed numerous time in past.
the sentiment analysis described here: https://www.knime.org/blog/sentiment-analysis is using terms as feature based on which a model will be trained. context i.t.o. opinion or favor is not considered here. However, you can of course add new cols as features representing opinion on specific topics or favor. That you be an interesting approach to add additional context information. Adding cols could be done by simply joining those features to the document vectors.
There is already one extra column which mention very generic feature as 'Policy' relate to above both contexts. Problem is that how can i discriminate +ve to -ve based on existing previous data.
Do you have labeled previous data i.t.o. +ve and -ve values? If so you could build a classifier.
i dont have label , but i have created the labels myself manually by reading whole text for both +ve and -ve.
Ok, that means you have a labeled data set and you can use a supervised learning approach. The article https://www.knime.org/blog/sentiment-analysis shows how to build a classifier for positive and negatove labels. You can adat this workflow to create one learning to predict you labels.
This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.