Learning representations for sentiment classification using Multi-task framework

Hardik Meisheri,Harshad Khadilkar

doi:10.18653/v1/w18-6244

Abstract

Most of the existing state of the art sentiment classification techniques involve the use of pre-trained embeddings. This paper postulates a generalized representation that collates training on multiple datasets using a Multi-task learning framework. We incorporate publicly available, pre-trained embeddings with Bidirectional LSTM’s to develop the multi-task model. We validate the representations on an independent test Irony dataset that can contain several sentiments within each sample, with an arbitrary distribution. Our experiments show a significant improvement in results as compared to the available baselines for individual datasets on which independent models are trained. Results also suggest superior performance of the representations generated over Irony dataset.

Highlights

Sentiment analysis has attracted substantial research interest, especially in the field of social media, owing to the growing number of data and active users
We train and evaluate our model on sentiment classification SemEval dataset obtained through shared task and affect emotion dataset from SemEval-2018
We compare the results with sentiment specific word embeddings (Tang et al, 2014), where we use Fully connected layers along with attention as the downstream model

Summary

Introduction

Sentiment analysis has attracted substantial research interest, especially in the field of social media, owing to the growing number of data and active users. The research community has gravitated towards a pragmatic characterization of language with the division into (and quantification of) specific emotions for sentiment analysis. This approach has come to prominence in recent times as a large number of enterprises (not just social media corporations) rely on understanding customer sentiments for defining product and marketing strategies (Pang and Lee, 2004; Socher et al, 2012). It is conceivable that an automated system quickly alerting the management about the rate and depth of negative sentiments due to the incident, would have enabled them to produce a more amelioratory response from the outset

Methods

Results

Conclusion