Toward Effective Suggestion Mining in Game Reviews: Introducing an Aspect-Based Multiclass Multilevel Dataset

This paper presents a comprehensive dataset for aspect-based multiclass multilevel suggestion mining from game reviews, addressing the challenges in capturing explicit and implicit suggestions along with their associated aspects. The dataset is carefully annotated by expert annotators, providing valuable insights into the nature of suggestions in game reviews. We describe the dataset’s development process and evaluate it using various deep learning models, including LSTM, TCN, DRC_Net, and GPT-3. The results demonstrate the dataset’s effectiveness in suggestion mining tasks. Furthermore, we combine TCN with machine learning models to refine subclass and aspect identification. The dataset fills the gap in existing datasets for aspect-based suggestion mining, enabling researchers to develop more robust suggestion mining models and gain a deeper understanding of suggestions in game reviews.