What is karma? Quantifying online influence and credibility

Many online communities explicitly publicize the concept of karma or reputation for users, computed as a sum of positive votes by other members of the community. These explicit measures are often seen as proxies for more intangible notions such as influence (the ability of a member to persuade others) and credibility (the trustworthiness of the user as an member of the community). In this paper, we describe the range of network characteristics, interactions, and temporal factors that affect the accumulation of karma points. In order to confirm our intuitions about influential network characteristics, we evaluate our features on datasets that contain an explicit measure of karma. However, our primary motivation in developing a model that enables generalization to private communities without a concept of “voting” for users but with an implied network structure. For example, our findings can be applied to e-mail interaction graphs to determine a private “karma” score for members of an organization, and to subsequently identify users who have high influence and credibility. Our paper is organized as follows: we begin by describing prior studies of interaction networks in Section 2. Afterwards, we discuss our collection of data from two disparate online communities: Hacker News and Super User, and the conversion into an interaction graph structure. In Section 4 we examine different properties of these networks and how they relate to karma and reputation. Using our results, we train classifiers in Section 5 that are able to identify high-karma individuals with an improvement over our baseline measure. Finally, we conclude with future directions of research in Section 6.