The rapid increase in research publications, particularly in computer science, presents significant challenges for researchers, reviewers, and institutions in performing comprehensive literature reviews. Traditional citance analysis, which mainly focuses on sentiment and citance intent, often misses important details. To address this, we introduce a new approach that combines three key aspects: Intent, Polarity, and Semantics. Using data from Semantic Scholar we classify citances based on this approach to create CitAn dataset which has 1,165 manually labeled citances from 80 subfields in Computer Science, Health Sciences, Bioinformatics, and Social Sciences and Humanities. Finally, we fine-tune state-of-the-art language models using this dataset, which yields promising results and paves the way for efficient citance analysis across various academic fields.