Aspect-Based-Sentiment-Analysis

This repository tries to implement Aspect-Based-Sentiment-Analysis on SemEval 2014 dataset. It is a part of course project of the course CS671: Introduction to Natural Language Processing. The approach followed here is mainly dependent on the paper “Aspect extraction for opinion mining with a deep convolutional neural network” published by Soujanya Poria, Erik Cambria and Alexander Gelbukh in Knowledge-Based Systems.

References

To be updated soon.

Results

CNN + GloVe WE + POS:

Precision: 51.64% Recall: 46.32% F1_score: 48.84%

CNN + GloVe WE :

Precision: 47.95% Recall: 43.01% F1_Score: 45.34%

CNN + Google WE + POS :

Precision: 46.75% Recall: 66.18% F1_Score: 54.79%