Erheng Zhong

Yue Shi, Erheng Zhong, Suju Rajan et al.

We tackle the blog recommendation problem in Tumblr for mobile users in this paper. Blog recommendation is challenging since most mobile users would suffer from the cold start when there are only a limited number of blogs followed by the user. Specifically to address this problem in the mobile domain, we take into account mobile apps, which typically provide rich information from the users. Based on the assumption that the user interests can be reflected from their app usage patterns, we propose to exploit the app usage data for improving blog recommendation. Building on the state-of-the-art recommendation framework, Factorization Machines (FM), we implement app-based FM that integrates app usage data with the user-blog follow relations. In this approach the blog recommendation is generated not only based on the blogs that the user followed before, but also the apps that the user has often used. We demonstrate in a series of experiments that app-based FM can outperform other alternative approaches to a significant extent. Our experimental results also show that exploiting app usage information is particularly effective for improving blog recommendation quality for cold start users.

Avradeep Bhowmik, Nathan Liu, Erheng Zhong et al.

While matrix factorisation models are ubiquitous in large scale recommendation and search, real time application of such models requires inner product computations over an intractably large set of item factors. In this manuscript we present a novel framework that uses the inverted index representation to exploit structural properties of sparse vectors to significantly reduce the run time computational cost of factorisation models. We develop techniques that use geometry aware permutation maps on a tessellated unit sphere to obtain high dimensional sparse embeddings for latent factors with sparsity patterns related to angular closeness of the original latent factors. We also design several efficient and deterministic realisations within this framework and demonstrate with experiments that our techniques lead to faster run time operation with minimal loss of accuracy.

Zhongqi Lu, Erheng Zhong, Lili Zhao et al.

Collaborative filtering (CF) aims to predict users' ratings on items according to historical user-item preference data. In many real-world applications, preference data are usually sparse, which would make models overfit and fail to give accurate predictions. Recently, several research works show that by transferring knowledge from some manually selected source domains, the data sparseness problem could be mitigated. However for most cases, parts of source domain data are not consistent with the observations in the target domain, which may misguide the target domain model building. In this paper, we propose a novel criterion based on empirical prediction error and its variance to better capture the consistency across domains in CF settings. Consequently, we embed this criterion into a boosting framework to perform selective knowledge transfer. Comparing to several state-of-the-art methods, we show that our proposed selective transfer learning framework can significantly improve the accuracy of rating prediction tasks on several real-world recommendation tasks.