Online commercial app marketplaces serve millions of apps to billions of users in an efficient manner. Bandit optimization algorithms are used to ensure that the recommendations are relevant, and converge to the best performing content over time. However, directly applying bandits to real-world systems, where the catalog of items is dynamic and continuously refreshed, is not straightforward. One of the challenges we face is the existence of several competing content surfacing components, a phenomenon not unusual in large-scale recommender systems. This often leads to challenging scenarios, where improving the recommendations in one component can lead to performance degradation of another, i.e., “cannibalization”. To address this problem we introduce an efficient two-layer bandit approach which is contextualized to user cohorts of similar taste. We mitigate cannibalization at runtime within a single multi-intent content surfacing platform by formalizing relevant offline evaluation metrics, and by involving the cross-component interactions in the bandit rewards. The user engagement in our proposed system has more than doubled as measured by online A/B testings.

In the classical selection problem, the input consists of a collection of elements and the goal is to pick a subset of elements from the collection such that some objective function ƒ is maximized. This problem has been studied extensively in the data-mining community and it has multiple applications including influence maximization in social networks, team formation and recommender systems. A particularly popular formulation that captures the needs of many such applications is one where the objective function ƒ is a monotone and non-negative submodular function. In these cases, the corresponding computational problem can be solved using a simple greedy (1-1/e)-approximation algorithm.

Competition is a primary driver of player satisfaction and engagement in multiplayer online games. Traditional matchmaking systems aim at creating matches involving teams of similar aggregated individual skill levels, such as Elo score or TrueSkill. However, team dynamics cannot be solely captured using such linear predictors. Recently, it has been shown that nonlinear predictors that target to learn probability of winning as a function of player and team features significantly outperforms these linear skill-based methods. In this paper, we show that using final score difference provides yet a better prediction metric for competitive balance. We also show that a linear model trained on a carefully selected set of team and individual features achieves almost the performance of the more powerful neural network model while offering two orders of magnitude inference speed improvement. This shows significant promise for implementation in online matchmaking systems.

Teams that bring together experts with different expertise are important for solving complex problems. However, research shows that teaming up people simply based on their ability is not enough. Team members need to have clear roles, and they should mutually endorse and respect their teammates for the role they assume on the team. In this paper, we define the MaxMutualRespect problem, a novel team-formation problem that asks for a set of experts, each assigned to a distinct role, such that the total respect that the individuals receive by the rest of the team members for their assigned role is maximized. We show that the problem is NP-complete and we consider approximation and heuristic algorithms. Experiments with real datasets demonstrate that our problem definitions and algorithms work well in practice and yield intuitive results.

The proliferation of online social networks and the spread of smart mobile devices enable the collection of information related to a multitude of users' activities. These networks, where every node is associated with a type of action and a frequency, are usually referred to as activity networks. Examples of such networks include road networks, where the nodes are intersections and the edges are road segments. Each node is associated with a number of geolocated actions that users of an online platform took in its vicinity. In these networks, we define a prize-collecting subgraph to be a connected set of nodes, which exhibits high levels of activity, and is compact, i.e., the nodes are close to each other. The k-PCSubgraphs problem we address in this paper is defined as follows: given an activity network and an integer k, identify k non-overlapping and connected subgraphs of the network such that the nodes of each subgraph are close to each other, and the total number of actions they are associated with is high. Here, we define and study two new variants of the k-PCSubgraphs problem, where the subgraphs of interest are tours and paths. Since both these problems are NP-hard, we provide approximate and heuristic algorithms that run in time nearly-linear to the number of edges. In our experiments, we use real activity networks obtained by combining road networks and projecting on them user activity from Twitter and Flickr. Our experimental results demonstrate both the efficiency and the practical utility of our methods.