Important Citation Identification by Exploiting the Optimal In-text Citation Frequency

Research is always based on previously done work. To acknowledge the worthy work of the predecessors of the field, researchers do citations. Citations are factors that are used for measuring the impact factor of journals, to rank the researchers, to find out latest research topics, for allocating research grants etc. In current epoch the research community has turned their focus towards citations and is of the view that all citations are not equally important. To find out important citations, researchers used different approaches such as context based, cue word based, metadata based, frequency based, textual based etc. Among proposed methodologies, frequency based approach was extensively used. The citation with high frequency was considered as important, but it is yet unclear that what should be the frequency cut off value of citation for considering it important. This research explored the significance of applying Threshold value over Frequency count for binary classification. We identified optimal threshold value of frequency count and further applied this to classify the citations into important and non-important ones. To evaluate the proposed approach a benchmark data set annotated by two domain experts was used that consisted of 465 citation pairs. The results were compared with state of the art precision value of 0.72. While the experiment showed increased value of 0.75 in terms of precision.