UIT-VSFC: Vietnamese Students’ Feedback Corpus for Sentiment Analysis

Abstract

Students' feedback is a vital resource for the interdisciplinary research combining of two fields: sentiment analysis and education. To strengthen the sentiment analysis of the Vietnamese language which is a low-resource language, we build a Vietnamese Students' Feedback Corpus (UIT-VSFC), a free and high-quality corpus for research on two different tasks: sentiment-based and topic-based classifications. In this paper, we present the methods of building annotation guidelines and ensure the annotation accuracy and consistency of this corpus. The resource consists of over 16,000 sentences which are human-annotated on the two tasks. To assess the quality of our corpus, we measure the inter-annotator agreements and classification accuracies on our UIT-VSFC. As a result, we achieved 91.20% of the inter-annotator agreement for the sentiment-based task and 71.07% of that for the topic-based task. In addition, the best results are of baseline model as the Maximum Entropy classifier with 87.94% and 84.03% of the overall F1-score of the sentiment-based and topic-based tasks respectively. These results illustrate that the corpus is reliable and helpful resource for research.