The strength distribution and combined duration prediction of online collective actions: Big data analysis and BP neural networks

Abstract

Unveiling the patterns and mechanisms of human behaviors is the core task of collective action studies. In terms of the strength (impacts or power) of collective actions, some of them have bigger political influences or societal impacts than others, and the success chance or probability therefore varies a great deal. Some online collective actions have little effects or favorable outcomes, while others have successfully changed policies or decisions made by local or central governments; and others even overthrow the governments. Hence, the indicator of strength is applied to measure the powers, pressures, attentions, concerns, and impacts generated by certain collective actions. The strength of collective action is defined and calculated by the life function, i.e. it is defined as the summation or integral of life function (viability or total participation) divided by the durations or spans. There exists some regularity in terms of the strength’s distribution under both simulations and big data exploration. The peak model is utilized to simulate online collective actions, and the distribution of strength (N=1000) is close to normal distributions; it indicates by the observed big data cases (N=159) that it follows the lognormal distribution, which also holds true for subgroups. The introduction of strength paves the way for predicting the durations or spans of online collective actions. For the combined prediction with three factors (peak’s timing, peak’s heights, and viability), the accuracy of predicting durations or spans is close to 100% for both simulated data and observed big data. For separate predictions with single factor, the accuracy is closer to 100% as well.