An approach for suggestion mining based on deep learning techniques

Abstract

An organization often uses forums and social media channels for getting feedback from costumers or users. The ratings of products on rating platforms are a useful feedback to make a product better. The feedback from a customer is in the form of a suggestion which appears in a rating text or is directly asked from the customer. Suggestion mining is a binary classification problem that labels sentences as Suggestion or Non-suggestion. The suggestion mining is similar to sentiment analysis which is associated with common linguistic properties and challenges irrespective of the domain and application. Most of the previous works of suggestion mining proposed rule based methods and a very few developed statistical classifiers by using manually identified features. Recently, several researchers paid attention on deep learning technique based solutions to suggestion mining where features are automatically learned. In this work, various deep learning techniques like RNN, LSTM, Attention based LSTM and GRU are used in the experimentation of suggestion mining. The experiment carried out on the dataset provided in SemEval 2019 suggestion mining competition. The Attention based LSTM achieved best accuracies for suggestion mining when compared with other deep learning techniques.

Enhancing Suggestion Detection in Online User Reviews through Integrated Information Retrieval and Deep Learning Approaches

“…Reddy et al [28] applied various deep learning techniques such as RNN, LSTM, attention-based LSTM, and GRU for suggestion mining. The findings indicate that attention-based LSTM achieved the best accuracies for suggestion mining.…”

Automatic Suggestion Detection from Online User Reviews with Information Retrieval and Deep Learning

“…Fatyanosa et al [19] used several machine learning approaches, such as logistic Regression, Random Forest, Multinomial Naive Bayes, Linear Support Vector Classification, Sublinear Support Vector Classification, Variable Length Chromosome Genetic Algorithm-Naive Bayes, and CNN. They compared these approaches on the software and hotel datasets and obtained F 1 Score 0.47 and 0.37, respectively [20].…”