Resumen
Automated citation analysis is becoming increasingly important in assessing the scientific quality of publications and identifying patterns of collaboration among researchers. However, little attention has been paid to analyzing the scientific content of the citation context. This study presents an unsupervised citation detection method that uses semantic similarities between citations and candidate sentences to identify implicit citations, determine their functions, and analyze their sentiments. We propose different document vector models based on TF-IDF weights and word vectors and compare them empirically to calculate their semantic similarity. To validate this model for identifying implicit citations, we used deep neural networks and LDA topic modeling on two citation datasets. The experimental results show that the F1 values for the implicit citation classification are 88.60% and 86.60% when the articles are presented in abstract and full-text form, respectively. Based on the citation function, the results show that implicit citations provide background information and a technical basis, while explicit citations emphasize research motivation and comparative results. Based on the citation sentiment, the results showed that implicit citations tended to describe the content objectively and were generally neutral, while explicit citations tended to describe the content positively. This study highlights the importance of identifying implicit citations for research evaluation and illustrates the difficulties researchers face when analyzing the citation context.