Opinion mining in higher education: a corpus-based approach

Student recruitment and retention heavily depend on student satisfaction and word of mouth voiced through surveys, social media, and review websites. The amount of online reviews makes their manual analysis intractable, revealing the need for automated approaches. The paper presents the first Serbian language corpus manually annotated for opinions in the domain of higher education. Agreement among annotators was calculated using the Fleiss' kappa and agr metrics. Statistical and linguistic analyses of the corpus revealed information useful for hand-crafted rules for sentiment analysis. Using developed corpus, dictionary- and machine learning-based approaches to sentiment analysis are benchmarked. Both exhibit high performance.