The topological face of recommendation: models and application to bias detection

Recommendation plays a key role in e-commerce and in the entertainment industry. We propose to consider successive recommendations to users under the form of graphs of recommendations. We give models for this representation. Motivated by the growing interest for algorithmic transparency, we then propose a first application for those graphs, that is the potential detection of introduced recommendation bias by the service provider. This application relies on the analysis of the topology of the extracted graph for a given user; we propose a notion of recommendation coherence with regards to the topological proximity of recommended items (under the measure of items’ k-closest neighbors, reminding the ”small-world” model by Watts & Stroggatz). We finally illustrate this approach on a model and on Youtube crawls, targeting the prediction of ”Recommended for you” links (i.e., biased or not by Youtube). The output of recommender systems are benchmarked by researchers and practitioners based on their precision and recall performances on test datasets [11]. Yet, while those metrics have proven useful for assessing the performances of recommenders, we find that the graph data-structure has not been applied for studying and learning about the recommendations made to users (i.e., the recommenders’ outputs). We argue that graph theory and the wide spectrum of graph algorithms available for data mining complex networks can be as well leveraged for complementing studies about recommender results. Our proposal is to represent the recommendations to users in either a global graph of recommendations, available by the service provider at a given point in time, or as a user-graph of recommendations that only captures the recommendation space to a single user, and that can also be observed at the service or by the user herself through the crawling of the service recommendation interface. The extracted graph topology is thus to be leveraged for analysis. One application we target is related to the field of algorithmic transparency. Some major service providers, such as Youtube, comment on the high level implementation of their recommender, without specifying details that would allow a transparent use by the public [4]. Recently, there has been an increase in the will for accountability of the service provided by those systems, that can be viewed as black-boxes operating in the cloud, and that a user interacts with 1 ar X iv :1 70 4. 08 99 1v 1 [ cs .S I] 2 8 A pr 2 01 7