Comparison of Long Short-Term Memory Networks and Temporal Convolutional Networks for Sentiment Analysis

Abstract

The use of AI to detect human emotion is growing rapidly with the need for human-like customer service without constant human interaction, conducting market research, monitoring opinions on social media, and so on. One model that is often used for this sentiment detection is a recurrent neural network with long short-term memory. Another, newer model that can be used is a temporal convolutional network, but the differences between these two models are under-researched. The goal of this project is to apply these models and compare their performance when detecting sentiment. This will provide some guidance to programmers, allowing them to be more aware of the implications of implementing one model over the other. We find that overall, a temporal convolutional network outperforms the recurrent neural network with long short-term memory for sentiment analysis; our temporal convolutional network achieves 72% accuracy when detecting sentiment.