How Time‑Inhomogeneous Markov Chains Reveal Shifting Social Behaviors

Our daily actions appear patterned, yet social behavior is not as mechanical as a clock; it varies with time and environment. To capture this dynamics, we use time‑inhomogeneous Markov chains, which allow transition probabilities to change over time.

What is a Time‑Inhomogeneous Markov Chain?

1. Basics of Markov Chains

A Markov chain is a discrete stochastic process with the "memoryless" property: the future state depends only on the current state. For example, a user’s social‑media activity might depend only on yesterday’s activity, not on earlier days.

In a homogeneous Markov chain the transition probabilities are fixed, but real life is more complex; decisions are influenced by time (e.g., New Year enthusiasm vs. mid‑year fatigue). Hence we need a more flexible model—time‑inhomogeneous Markov chains—where probabilities are functions of time.

2. Introducing Time‑Inhomogeneity

In a time‑inhomogeneous Markov chain the probability of moving from one state to another varies with time. This allows us to capture how social behavior changes dynamically and to explore the impact of time, environment, and sudden events.

2. Modeling Social Behavior

1. Social Media Activity Example

Assume three user states:

State 1 (High activity) : multiple daily uses of the platform.

State 2 (Medium activity) : several uses per week.

State 3 (Low or inactive) : rarely uses the platform.

Behavior is not static; users may suddenly increase activity or lose interest. We aim to describe and predict these changes with a time‑inhomogeneous Markov chain.

2. Building the Model

State Transition Time Dependence

We let transition probabilities be time‑dependent. For example, the probability of moving from high to medium activity may increase over time as novelty fades; the probability of moving from medium to high may decrease; and the probability of moving from high to inactive may spike at specific moments (e.g., policy changes).

Specific functional forms can be defined for these probabilities (omitted for brevity).

These functions reflect how the likelihood of transitioning between states evolves with time.

Initial State and Evolution

Assume at the initial moment the distribution is 70% high activity, 20% medium, and 10% low. Using recursive formulas and the time‑varying transition matrix, we can predict future state distributions.

3. Evolution and Prediction

Simulating the model for the next 12 months with the initial distribution yields the projected activity changes shown below.

Through time‑inhomogeneous Markov chains we can deeply analyze how user behavior evolves over time, providing a theoretical foundation for platform optimization, personalized recommendation, and macro‑level social behavior analysis.