Emergence of scaling in human-interest dynamics

Human behaviors are often driven by human interests. Despite intense recent efforts in exploring the dynamics of human behaviors, little is known about human-interest dynamics, partly due to the extreme difficulty in accessing the human mind from observations. However, the availability of large-scale data, such as those from e-commerce and smart-phone communications, makes it possible to probe into and quantify the dynamics of human interest. Using three prototypical “big data” sets, we investigate the scaling behaviors associated with human-interest dynamics. In particular, from the data sets we uncover power-law scaling associated with the three basic quantities: (1) the length of continuous interest, (2) the return time of visiting certain interest, and (3) interest ranking and transition. We argue that there are three basic ingredients underlying human-interest dynamics: preferential return to previously visited interests, inertial effect, and exploration of new interests. We develop a biased random-walk model, incorporating the three ingredients, to account for the observed power-law scaling relations. Our study represents the first attempt to understand the dynamical processes underlying human interest, which has significant applications in science and engineering, commerce, as well as defense, in terms of specific tasks such as recommendation and human-behavior prediction.

💡 Research Summary

This paper pioneers the quantitative study of human‑interest dynamics by exploiting three large‑scale online activity datasets: Douban (cultural content), Taobao (e‑commerce), and Mobile‑Phone Reading (MPR). Together they cover over 70,000 users and up to 18 months of click‑stream data. The authors define three fundamental observables: (1) the length l of a continuous session within the same interest category, (2) the return interval τ measured as the number of clicks between two visits to the same category, and (3) the rank r of each interest based on its visitation frequency for a given user. Empirical analysis reveals robust power‑law scaling for all three quantities across all datasets: P(l) ∝ l^−α, P(τ) ∝ τ^−β, and P(r) ≈ r^−γ exp(−r/S). The heavy tails of l and τ indicate that users can stay unusually long in a single interest and can also revisit an old interest after very long gaps, contradicting a simple Markovian (exponential) assumption. Transition‑probability matrices p(i,j) further show (i) high probabilities for switches among top‑ranked interests and (ii) large diagonal entries p(i,i), evidencing a preferential return mechanism and an inertial effect that keeps users within a currently explored category.

Guided by these observations, the authors propose a phenomenological biased random‑walk model. At each “hop” a user either explores a brand‑new interest with probability ρ_n − λ (ρ∈(0,1], λ>0, n = number of distinct interests already visited) or returns to a previously visited interest with complementary probability. The return choice follows a preferential attachment rule proportional to past visitation frequencies, capturing the preferential return. Once an interest is selected, an “inertial” phase is modeled as an excited random walk (ERW), which generates the observed long sequences of clicks within the same category. Analytical treatment (provided in the Supplementary Information) and numerical simulations reproduce the empirical exponents α, β, and γ, demonstrating that the three identified mechanisms—preferential return, inertia, and exploration—are sufficient to generate the observed scaling laws.

The model, however, does not predict the exact exponent values and omits finer cognitive factors such as individual memory decay, attention limits, and detailed content semantics. The authors acknowledge these limitations and suggest extensions incorporating memory kernels, cognitive load, or category hierarchies.

Beyond theory, the paper quantifies predictability of user behavior using entropy and Fano’s inequality, finding that despite apparent randomness, user interest trajectories are highly predictable. The findings have immediate implications for recommendation systems, targeted advertising, behavioral forecasting, and even medical or security contexts where understanding evolving human interests is critical. In sum, the study establishes that human‑interest dynamics exhibit universal scaling patterns rooted in a simple set of stochastic rules, opening a new quantitative avenue for both scientific inquiry and practical applications.