In matrix factorization, available graph side-information may not be well suited for the matrix completion problem, having edges that disagree with the latent-feature relations learnt from the incomplete data matrix. We show that removing these contested edges improves prediction accuracy and scalability. We identify the contested edges through a highly-efficient graphical lasso approximation. The identification and removal of contested edges adds no computational complexity to state-of-the-art graph-regularized matrix factorization, remaining linear with respect to the number of non-zeros. Computational load even decreases proportional to the number of edges removed. Formulating a probabilistic generative model and using expectation maximization to extend graph-regularised alternating least squares (GRALS) guarantees convergence. Rich simulated experiments illustrate the desired properties of the resulting algorithm. On real data experiments we demonstrate improved prediction accuracy with fewer graph edges (empirical evidence that graph side-information is often inaccurate). A 300 thousand dimensional graph with three million edges (Yahoo music side-information) can be analyzed in under ten minutes on a standard laptop computer demonstrating the efficiency of our graph update.

In difficult information seeking tasks, the majority of top-ranked documents for an initial query may be non-relevant, and negative relevance feedback may then help find relevant documents. Traditional negative relevance feedback has been studied on document results; we introduce a system and interface for negative feedback in a novel exploratory search setting, where continuous-valued feedback is directly given to keyword features of an inferred probabilistic user intent model. The introduced system allows both positive and negative feedback directly on an interactive visual interface, by letting the user manipulate keywords on an optimized visualization of modeled user intent. Feedback on the interactive intent model lets the user direct the search: Relevance of keywords is estimated from feedback by Bayesian inference, influence of feedback is increased by a novel propagation step, documents are retrieved by likelihoods of relevant versus non-relevant intents, and the most relevant keywords (having the highest upper confidence bounds of relevance) and the most non-relevant ones (having the smallest lower confidence bounds of relevance) are shown as options for further feedback. We carry out task-based information seeking experiments with real users on difficult real tasks; we compare the system to the nearest state of the art baseline allowing positive feedback only, and show negative feedback significantly improves the quality of retrieved information and user satisfaction for difficult tasks.