Improving Performance of Long Short-Term Memory Networks for Sentiment Analysis Using Multicore and GPU Architectures.

A massive amount of data is generated on the Internet through websites, blogs, and social networks, with all kinds of content, including comments about companies and their products. Sentiment analysis (SA) is the interpretation of emotions in texts. It is essential for different companies as it helps identify customer's opinions. It is also beneficial to understand people's responses to new content, providing audience insights to help make decisions. However, current technological advances enable us to efficiently store and retrieve these immense amounts of data to better insight into different application areas. Companies use this information to make marketing decisions. In response, we propose a performance optimization LSTM for SA using multicore and GPUs to keep the accuracy. To validate our proposal, we have applied it over a public database with 50,000 film records. The results showed a performance improvement of 3.17 times on the multicore and 12.15 on the GPU.