Extracting reference text from citation contexts

Information from the textual context of citations in scientific articles has been studied and used in many applications by the research community. For example, it has been used in topic modeling, sentiment analysis, scientific paper summarization and information retrieval. However, these applications suffer the problem of right identification of citation context window and alternately use the text in a fixed size window around the citation mention. In this way, citation contexts may contain terms or other text that is not used for describing the citation and should not be included in the citation context. Identifying such nonreference text in the citation context is a non-trivial task, yet significant. In this paper, it is attempted to identify and remove the non-reference text from the citation contexts by developing a heuristic algorithm based on pruning the transition-based dependency parse tree. Evaluating the accuracy of our algorithm, results showed 77% macro-precision, 83% macro-recall and 80% F-macro for 88 research articles of testing dataset having varying number of citations. Additionally, we find that for many of the cited articles in our testing dataset, the number of objective citation contexts is more than subjective ones.