Citation Burst

The beginning of a blue line depicts when an article is published. The beginning of  a red segment marks the beginning of a period of burst, whereas the end of the red segment marks the end of the burst period.

Burst Detection

A burst refers to a frequency surge of a particular type of events, for example, a surge of citations to a Nobel Prize winning publication. 
CiteSpace supports burst detection on several types of events:

  • Single or multi-word phrases from the title, abstract, or other parts of a publication

  • The number of citation counts of cited references over time

  • The frequencies of keyword appearances over time

  • The number of publications by an author, an institution, or a country

A detailed description of the original algorithm can be found in Kleinber's 2002 publication.
In CiteSpace, users may adjust the burst detection parameters in the Burstness tab in the Control Panel. For example, to find more items of burst, i.e. increase the sensitivity of the burst detection, reduce the gamma value. To reduce the number of items of burst to be identified, increase the minimum duration.

State Transition Cost α

The parameter α controls the cost of switching between burst states, i.e., how difficult it is to move from one state to a higher burst state: with a small value of α (e.g., α=0.5), state changes are easier, burst detection would find more items with burst; in contrast, with a larger α value (e.g., α=2), only very strong changes in frequency would qualify as bursts so you end up with fewer burst items.

Growth Rate γ

In burst detection, each burst level corresponds to an exponentially increasing event rate. Gamma controls the growth rate between burst levels.

Intuitively, a small gamma corresponds to small differences between burst levels, whereas a large gamma separates burst levels further apart.

Adjustment strategies:

  • Too many bursts detected → increase α

  • Burst intensity differences unclear → increase γ

  • No emerging topics detected → decrease α

References
Kleinberg, J. Bursty and hierarchical structure in streams. In Proceedings of the 8th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (Edmonton, Alberta, Canada, 2002), ACM Press, 2002, 91-101.