DeepAggregation: A new approach for aggregating incomplete ranked lists using multi-layer graph embedding

Abstract

Preference aggregation, and specifically rank aggregation, is a well known problem in the fields of computational social choice and preference handling with broad application including web search and recommendation systems. Inspired by the recent advances in the area of deep neural representation learning, for the first time in the literature, in this paper we leverage unsupervised deep learning techniques - especially graph embeddings - for aggregating a collection of incomplete rank lists and accordingly we develop an algorithm called DeepAggregation. It takes as input a set of incomplete rank lists and constructs a multi-layer graph wherein the nodes are the alternatives that are ranked and the edges capture information contained in the incomplete rank lists. We then compute deep neural representation vectors (i.e. embeddings) for the nodes and then derive the aggregated order using these representation vectors. Our proposed algorithm can handle incomplete rank lists with or without ties. We conduct thorough empirical analysis of the proposed DeepAggregation algorithm using various real life data sets such as TripAdvisor reviews data. We empirically observe that DeepAggregation generates impressive results in comparison with a number of well-known state-of-the-art preference aggregation methods.