Using an auxiliary dataset to improve emotion estimation in users’ opinions

摘要

Sentimental analysis of social networking data is an economically affordable and effective way to track and evaluate public viewpoints that are critical for decision making in different areas. Predicting the users’ future opinions is crucial for companies and services; if companies understand users’ sentiments in considered time frames, they can do much better by knowing where exactly users are satisfied or unsatisfied. Utilizing an auxiliary dataset, this study uses the opinions of users on the Twitter social network expressed in the form of short text, and presents the Auxiliary Dataset-Latent Dirichlet Allocation (AD-LDA) model to improve the learning of users’ emotions around a specific topic. The proposed model considers the emotions –as predefined sentiments with a wide sentimental outlook– to estimate users’ feelings and sentiments about a particular subject or event. Coherence score evaluation results for the four studied hashtags showed an average 64.15% improvement compared to the conventional LDA model. The average Weighted-F1 criteria for studied hashtags was 79.83% for the accuracy of learning. Experimental and evaluation results show that our proposed model can effectively learn the emotions of words which leads to a better understanding of users’ feelings.