Referral reward embedded bi-phase information diffusion technique for social networks

Abstract

Social networks provide an effective marketing platform to enhance the visibility of a business by reaching out to a wide base of potential customers cutting across gaps of geographic locations, age groups, and socio-economic differences. Implementing a successful viral marketing campaign through social media is one of the key strategies of any company that seeks to acquire more prospective customers. These companies usually set aside a certain budget for such a campaign, so as to reach out to and influence as many potential customers as possible. The problem of maximizing the spread of influence with a limited budget has been a central theme of research in social media analytics. Most approaches to this problem in the existing literature spend the entire budget in one single installment by triggering the influence spread at carefully selected seed nodes. This thesis investigates the effect of splitting the budget across two sequential phases, each phase corresponding to a different way of spreading influence. The key idea of our proposed approach is as follows: In Phase 1, we adopt the classical approach of triggering influence spread at a selected set of seed nodes by using a certain fraction of the available budget. In Phase 2, we use the remaining budget in the form of referral incentives offered for every successful referral (and hence addition of a new customer). Using a popular and well-accepted model for influence spread - the Independent Cascade Mode - we formulate an objective function for the proposed two-phase influence maximization problem and investigate its properties. With the help of detailed experimentation on synthetic and real-world datasets, we determine an effective budget split between the two phases. The principal findings from our study are as follows: Abstract

When the available budget is low, it is prudent to use it entirely for Phase 1, while when the budget is moderate to high, it is preferable to use much of the budget for Phase 1 and offer small individual incentives in Phase 2. In the presence of moderate to strict temporal constraints, Phase 2 is not warranted, while if the temporal constraints are low or absent, Phase 2 yields a decisive improvement in influence spread.

The problem formulation and the significant applicability of results in this thesis leave wide scope for further theoretical studies as well as implementing referral schemes in practice.