IDS IUCL: Investigating Feature Selection and Oversampling for GermEval2017

Zeeshan Ali Sayyed, Daniel Dakota, Sandra Kübler

Abstract: We present the IDS IUCL contribution to the GermEval 2017 shared task on “Aspect- based Sentiment in Social Media Customer Feedback”. We choose to compete in both subtasks A & B. Our system focuses on handling the imbalance in the data sets, by focusing on feature selection and oversam- pling of the minority classes. We achieve 0.916 micro-F for relevance (task A) and 0.781 for polarity (Task B) on our devel- opment set. For task A, we reach the best scores among the submitted systems, for task B the 5th best results for timestamp1 and the best result for timestamp2.