This paper proposes a novel social media relevance feedback algorithm, based on multimedia content power (MCP). The algorithm estimates in a recursive manner, the similarity measure. This is accomplished by using a set of relevant/irrelevant samples, which are provided by the user, in order to adjust the system's response. In particular, the similarity measure is expressed in a parametric form of functional components. Another innovative point has to do with the estimation of MCP, which measures the influence of files over social media users. Toward this direction, user interactions (e.g., comments, likes, and shares) indicate that the file is influencing to them. The algorithm takes into consideration both the visual characteristics of multimedia files and their influence to retrieve information. The experimental results show that the proposed scheme offers several merits and future work is also discussed.