Recovering missing values in time series is critical when performing time series analysis. And the blackouts issue studied in this paper, described as losing all the data during a certain period, is among the most urgent issues due to its devastating impact on service quality, and is challenging because of the absence of coevolving data sequences for reference. As a result, many existing approaches that rely on data from other coevolving sequences for missing value recovery are infeasible in handling blackouts. To address the issue, this work proposes a novel Hankel matrix factorization approach, HKMF-T, to recover missing values during blackouts for tagged time series, where a tagged time series consists of a data sequence and a corresponding tag sequence. Motivated by real-world observations, HKMF-T decomposes the data sequence into two components: 1) an internal, slowly-varying smooth trend, and 2) external impacts indicated by the tag sequence. By transforming a partially observed data sequence into a corresponding Hankel matrix, we learn the above two components and estimate the missing values under a unified framework of Hankel matrix factorization. Extensive experiments are conducted to evaluate the practical performance of HKMF-T on real-world data sets. And the results suggest HKMF-T outperforms the baseline approaches for blackouts with long durations.

Recovering missing values in time series is critical when performing time series analysis. And the blackouts issue studied in this paper, described as losing all the data during a certain period, is among the most urgent and challenging issues. While the existing approaches for missing value recovery in time series could not handle this issue properly, in this work, we proposes a Hankel matrix factorization-based approach for tagged time series called HKMF-T, following the idea of decomposing a data sequence into the smooth trend and the external impact components. By transforming the data sequence into its Hankel matrix form, HKMF-T models the smooth trend implied by high-order temporal correlations as the product of two low-rank matrices, and learns the external impacts indicated by a corresponding tag sequence. Through extensive experiments conducted on three real-world data sets, HKMF-T shows its effectiveness by outperforming all baseline methods for blackouts with durations longer than nine sampling intervals.

Latent factor models have become a prevalent method in recommender systems, to predict users' preference on items based on the historical user feedback. Most of the existing methods, explicitly or implicitly, are built upon the first-order rating distance principle, which aims to minimize the difference between the estimated and real ratings. In this paper, we generalize such first-order rating distance principle and propose a new latent factor model (HoORaYs) for recommender systems. The core idea of the proposed method is to explore high-order rating distance, which aims to minimize not only (i) the difference between the estimated and real ratings of the same (user, item) pair (i.e., the first-order rating distance), but also (ii) the difference between the estimated and real rating difference of the same user across different items (i.e., the second-order rating distance). We formulate it as a regularized optimization problem, and propose an effective and scalable algorithm to solve it. Our analysis from the geometry and Bayesian perspectives indicate that by exploring the high-order rating distance, it helps to reduce the variance of the estimator, which in turns leads to better generalization performance (e.g., smaller prediction error). We evaluate the proposed method on four real-world data sets, two with explicit user feedback and the other two with implicit user feedback. Experimental results show that the proposed method consistently outperforms the state-of-the-art methods in terms of the prediction accuracy.

In recent years, recommender system is one of indispensable components in many e-commerce websites. One of the major challenges that largely remains open is the cold-start problem, which can be viewed as a barrier that keeps the cold-start users/items away from the existing ones. In this paper, we aim to break through this barrier for cold-start users/items by the assistance of existing ones. In particular, inspired by the classic Elo Rating System, which has been widely adopted in chess tournaments, we propose a novel rating comparison strategy (RAPARE) to learn the latent profiles of cold-start users/items. The centerpiece of our RAPARE is to provide a fine-grained calibration on the latent profiles of cold-start users/items by exploring the differences between cold-start and existing users/items. As a generic strategy, our proposed strategy can be instantiated into existing methods in recommender systems. To reveal the capability of RAPARE strategy, we instantiate our strategy on two prevalent methods in recommender systems, i.e., the matrix factorization based and neighborhood based collaborative filtering. Experimental evaluations on five real data sets validate the superiority of our approach over the existing methods in cold-start scenario.

Recognizing human activities from sensor readings has recently attracted much research interest in pervasive computing due to its potential in many applications, such as assistive living and healthcare. This task is particularly challenging because human activities are often performed in not only a simple (i.e., sequential), but also a complex (i.e., interleaved or concurrent) manner in real life. Little work has been done in addressing complex issues in such a situation. The existing models of interleaved and concurrent activities are typically learning-based. Such models lack of flexibility in real life because activities can be interleaved and performed concurrently in many different ways. In this paper, we propose a novel pattern mining approach to recognize sequential, interleaved, and concurrent activities in a unified framework. We exploit Emerging Pattern-a discriminative pattern that describes significant changes between classes of data-to identify sensor features for classifying activities. Different from existing learning-based approaches which require different training data sets for building activity models, our activity models are built upon the sequential activity trace only and can be applied to recognize both simple and complex activities. We conduct our empirical studies by collecting real-world traces, evaluating the performance of our algorithm, and comparing our algorithm with static and temporal models. Our results demonstrate that, with a time slice of 15 seconds, we achieve an accuracy of 90.96 percent for sequential activity, 88.1 percent for interleaved activity, and 82.53 percent for concurrent activity.