This manuscript introduces the idea of using Distributionally Robust Optimization (DRO) for the Counterfactual Risk Minimization (CRM) problem. Tapping into a rich existing literature, we show that DRO is a principled tool for counterfactual decision making. We also show that well-established solutions to the CRM problem like sample variance penalization schemes are special instances of a more general DRO problem. In this unifying framework, a variety of distributionally robust counterfactual risk estimators can be constructed using various probability distances and divergences as uncertainty measures. We propose the use of Kullback-Leibler divergence as an alternative way to model uncertainty in CRM and derive a new robust counterfactual objective. In our experiments, we show that this approach outperforms the state-of-the-art on four benchmark datasets, validating the relevance of using other uncertainty measures in practical applications.

Entropy regularization, smoothing of Q-values and neural network function approximator are key components of the state-of-the-art reinforcement learning (RL) algorithms, such as Soft Actor-Critic~\cite{haarnoja2018soft}. Despite the widespread use, the impact of these core techniques on the convergence of RL algorithms is not yet fully understood. In this work, we analyse these techniques from error propagation perspective using the approximate dynamic programming framework. In particular, our analysis shows that (1) value smoothing results in increased stability of the algorithm in exchange for slower convergence, (2) entropy regularization reduces overestimation errors at the cost of modifying the original problem, (3) we study a combination of these techniques that describes the Soft Actor-Critic algorithm.

We propose Meta-Prod2vec, a novel method to compute item similarities for recommendation that leverages existing item metadata. Such scenarios are frequently encountered in applications such as content recommendation, ad targeting and web search. Our method leverages past user interactions with items and their attributes to compute low-dimensional embeddings of items. Specifically, the item metadata is injected into the model as side information to regularize the item embeddings. We show that the new item representations lead to better performance on recommendation tasks on an open music dataset.

Expert finding is a task of finding knowledgeable people on a given topic. State-of-the-art expertise retrieval algorithms identify matching experts based on analysis of textual content of documents experts are associated with. While powerful, these models ignore social structure that might be available. In this paper, we develop a Bayesian hierarchical model for expert finding that accounts for both social relationships and content. The model assumes that social links are determined by expertise similarity between candidates. We demonstrate the improved retrieval performance of our model over the baseline on a realistic data set.

Clustering hypertext document collection is an important task in Information Retrieval. Most clustering methods are based on document content and do not take into account the hyper-text links. Here we propose a novel PageRank based clustering (PRC) algorithm which uses the hypertext structure. The PRC algorithm produces graph partitioning with high modularity and coverage. The comparison of the PRC algorithm with two content based clustering algorithms shows that there is a good match between PRC clustering and content based clustering.

Abstract In this article, we assess the effectiveness of Contextual Document Clustering (CDC) as a means of indexing within a dynamic and rapidly changing environment. We simulate a dynamic environment, by splitting two chronologically ordered datasets into time‐ordered segments and assessing how the technique performs under two different scenarios. The first is when new documents are added incrementally without reclustering [incremental CDC (iCDC)], and the second is when reclustering is performed [nonincremental CDC (nCDC)]. The datasets are very large, are independent of each other, and belong to two very different domains. We show that CDC itself is effective at clustering very large document corpora, and that, significantly, it lends itself to a very simple, efficient incremental document addition process that is seen to be very stable over time despite the size of the corpus growing considerably. It was seen to be effective at incrementally clustering new documents even when the corpus grew to six times its original size. This is in contrast to what other researchers have found when applying similar simple incremental approaches to document clustering. The stability of iCDC is accounted for by the unique manner in which CDC discovers cluster themes.