Rule-based Complaint Detection using RapidMiner

Salma Tayel; Matthias Reif; Andreas Dengel
In: Simon Fischer; Ingo Mierswa; JoĂŁo Mendes Moreira; Carlos Soares (Hrsg.). RCOMM 2013. RapidMiner Community Meeting and Conference (RCOMM-2013), August 27-30, Porto, Portugal, Pages 141-149, ISBN 978-3-8440-2145-5, Shaker-Verlag, Aachen, 8/2013.


Complaints are signs of user dissatisfaction from a service or product. Still, they are a good source of feedback for companies. Considering people reviews and complaints on the web can help them meet their customers’ expectations. Manually processing web articles related to some business to find out what people think of it is very time and effort consuming. Therefore, this task should be automated. Rule-based classifiers are very suitable for complaint detection. The generated classifiers are formed of rules that are comprehensible. This makes it easy for an employee to understand the criteria for classifying text as complaint. Our work compare five rule-based algorithms, OneR, ConjunctiveRule, Ridor, RIPPER, and PART using RapidMiner against complaint detection data sets of different domains. Results show that PART algorithm is the most suitable for complaint detection task. It achieves average 75% accuracy compared with about 60% for the other algorithms.

Weitere Links