Frequency Decomposition Model of Popularity Evolution in Online Social Media

The effective analysis of information diffusion popularity in online social media plays an important role in accurate mastering the development trend of public opinion and maintaining the normal order of society. The existing quantitative methods are mainly based on time-domain, which fails to comprehensively analyze the network dynamics, the contributing factor with the generalization. To this end, an assumption of waveform synthesis in the frequency-domain is presented, that is, the waveform of popularity evolution in the time-domain without noise is a linear combination of many sinusoidal waveforms with different frequencies and a single period via stretching and translation transforming. The transformation parameters can be utilized to analyze contributing factor and their variety corresponds to the network dynamic because such parameters are relative to the motive power and burst-time. Therefore, a waveform decomposition algorithm is designed for actual popularity evolution in this paper. The data of the diffusion waveform is standardized and normalized, then the greedy algorithm is utilized to decompose the waveform, and the results include two parts: the parameters of sub-waveforms and the value of SNR(Signal-to-Noise Ratio). Finally, the standardization and normalization benchmarks are gained via the Baseline Dataset which is the popularity evolution dataset in Weibo, and three different types of granularity are chosen to decompose the diffusion waveform in both the Baseline Dataset and the SpecialNews Dataset which includes entertainment news only. The results are compared to prove the generalization potential and feasibility of the model.