Abstract

Suggestion mining is a relatively new area & is challenged by issues like the complexity in a task or manual formulation, the knowledge of sentence-level semantics, figurative sentences, handling long & complex words, context dependence, & also very imbalanced class distribution. Deep learning is an industry that can be highly competitive in machine learning. We use the Random Multimodel Deep Learning (RMDL) approach in this paper to address the problem of suggestion mining using the SemEval-2019 Task 9 data sets. Though its data sets are very imbalanced and unstructured, we have utilized SMOTE techniques to extract class imbalance problems. To solve the imbalanced dataset problem, SMOTE (synthetic Minority oversampling technique) is a widely used over-sampling tool. Experimental findings show that the advantages of SMOTE to manage complex data and imbalanced data set are superior to our current SMOTE-RMDL (SMO-RMDL) model of the existing research process.

Key words: Suggestion Mining, Deep learning, CNN, RNN, DNN, RMDL, SMOTE