As the web gets involved in people’s everyday lives, mining keywords to determine patterns of online activities becomes essential in several fields, including sociology, behavior, and marketing. While existing mining techniques study the individual behavior of a target activity, this paper focuses on co-evolving characteristics of multiple activities, that is, keywords as well as individual behavior and seasonal dynamics. The underlying idea of this paper is that the behavior of keywords resembles the behavior of biological species because both consume limited resources; that is, the latter consumes natural resources such as food while the former consumes people’s attention, time, and money.

The main contribution of this paper is to propose a model that can forecast online keyword dynamics, in other words, the popularity change of a keyword, based on the competition among multiple keywords. The model consists of p, r, K, A, W, and B, each of which represents the initial popularity size of a set of keywords, the growth rate of each keyword, carrying capacity that covers all competing keywords, the interaction matrix for the competing keywords, and two seasonal component-related matrices. To find an appropriate parameter set for a given set of keywords, this paper proposes two algorithms: automatic seasonal component analysis and a multi-step fitting algorithm. The experimental results are quite bright. When the model is applied to seven sets of keywords, including video games, programming languages, apparel companies, and so on, it discovers competing keywords and captures seasonal activities, which closely matches human intuition.

The proposed model outperforms existing mining techniques in terms of accuracy and efficiency. The accuracy of the model proves that the co-evolving concept of keywords captures the characteristics of online activities correctly. However, it is worthwhile to note that the model is applicable to keywords that have a sufficient amount of activity history since the model parameters are obtained based on stochastic analysis of the keywords.