A Multi-Task Approach to Open Domain Suggestion Mining (Student Abstract)

Abstract

Consumer reviews online may contain suggestions useful for improving the target products and services. Mining suggestions is challenging because the field lacks large labelled and balanced datasets. Furthermore, most prior studies have only focused on mining suggestions in a single domain. In this work, we introduce a novel up-sampling technique to address the problem of class imbalance, and propose a multi-task deep learning approach for mining suggestions from multiple domains. Experimental results on a publicly available dataset show that our up-sampling technique coupled with the multi-task framework outperforms state-of-the-art open domain suggestion mining models in terms of the F-1 measure and AUC.

Open Domain Suggestion Mining Leveraging Fine-Grained Analysis

“…The same test-train split was used as provided in the original dataset, the details of which can be found in Table II. Similar to the prior work including Negi et al [2] and Jain et al [8], we use F1 score for the comparison of the various models experimented upon. The results of the various experiments performed are summarised in Table III.…”

“…This work was further extended by [8] where they perform suggestion mining as a multi-task approach to classify the nature as well as the domain of the given reviews. Our work follows the footsteps of [8] in modelling the nature of our downstream task.…”

“…In the case of random oversampling, all the above approaches work in the euclidean space and hence, interpretability is compromised which is not acceptable in Natural Language Analysis tasks. This problem was effectively tackled by Jain et al [8] where they propose LMOTE (Language Model-based Oversampling Technique) which leverages a Hidden Markov Model to oversample minority classes. Following this, Zhang et al [17] used a discourse marker-based technique to perform oversampling on the Sentiment Analysis of imbalanced datasets to yield significant better performance in an interpretable fashion.…”

“…Building on this work, Negi et al (c) [7] used an LSTMbased architecture to induce distant supervision in the Open-Domain Suggestion Mining task. This work was further extended by [8] where they perform suggestion mining as a multi-task approach to classify the nature as well as the domain of the given reviews. Our work follows the footsteps of [8] in modelling the nature of our downstream task.…”

“…• Training Baseline Classifier: The first step of phase 1 involves training a baseline sentiment analysis classifier. We use the architecture proposed by [8] for the same. • Discourse Marker Enhancement: For each review in the domain d i , we check for the presence of traditional discourse markers like "and", "but", "because".…”