Study of citation pattern classification based on transfer learning
Chuqing Feng · 2022
In the field of citation analysis, citation patterns can effectively reflect the changes of citation frequency in the time dimension, thus enhancing researchers’ knowledge and understanding of the citation pattern of literature and assisting them in the subsequent mining and utilization of literature value. However, there is currently no unified and complete citation analysis framework in the academic community, which can classify all citations to reflect different citation durations. In previous methods, the idea of feature extraction followed by classifier classification was used to identify the peak position and overall trend of the curve; However, subjective factors in this method interfere greatly with the setting of feature thresholds. Therefore, based on the analysis of a large amount of data, combined with previous research, this paper divides reference curves into the following categories: “left-biased” citation curves, “normal” citation curves, “right-biased” citation curves, bimodal citation curves, exponential growth citation curves, waveform citation curves, and “sleeping beauty” citation curves. In this framework, an end-to-end category recognition model, the Inception-v3 migration classification model, is constructed for efficient and accurate classification of large volume citation curves. The model finally achieves high accuracy and category recognition, and can perform the curve classification task well.