MLRec 2015

In conjunction with 15th SIAM International Conference on Data Mining (SDM 2015)
May 2, 2015, Vancouver, British Columbia, Canada

The proceeding of MLRec'15 is now available for download.

This workshop focuses on applying novel as well as existing machine learning and data mining methodologies for improving recommender systems. There are many established conferences such as NIPS and ICML that focus on the study of theoretical properties of machine learning algorithms. On the other hand, the recent developed conference ACM RecSys focuses on different aspects of designing and implementing recommender systems. We believe that there is a gap between these two ends, and this workshop aims at bridging the recent advances of machine learning and data mining algorithms to improving recommender systems. Since many recommendation approaches are built upon data mining and machine learning algorithms, these approaches are deeply rooted in their foundations. As such, there is an urgent need for researchers from the two communities to jointly work on 1) what are the recent developed machine learning and data mining techniques that can be leveraged to address challenges in recommender systems, and 2) from challenges in recommender systems, what are the practical research directions in the machine learning and data mining community.

Topics of Interest

We encourage submissions on a variety of topics, including but not limited to:

• Novel machine learning algorithms for recommender systems, e.g., new content/context aware recommendation algorithms, new algorithms for matrix factorization handling cold-start items, tensor-based approach for recommender systems, and etc.
• Novel approaches for applying existing machine learning algorithms, e.g., applying bilinear models, (non-convex) sparse learning, metric learning, low-rank approximation/PCA/SVD, neural networks and deep learning, for recommender systems.
• Novel optimization algorithms and analysis for improving recommender systems, e.g., parallel/distributed optimization techniques and efficient stochastic gradient descent.
• Industrial practices and implementations of recommendation systems, e.g., feature engineering, model ensemble, and lessons from large-scale implementations of recommender systems.
• Machine learning methods for security and privacy aware recommendations, user-centric recommendations with emphasize on users’ interaction and engagement, Explore-Exploit approach, multi-armed bandits for recommendation, and etc.

Submission

Instructions

The workshop accepts long paper and short (demo/poster) papers. Short papers submitted to this workshop should be limited to 4 pages while long papers should be limited to 8 pages. All papers should be formatted using the SIAM SODA macro. Authors are required to submit their papers electronically in PDF format to the submission site by 11:59pm MDT, Janurary 12 Feb 2, 2015. The site has started to accept manuscrips.

Important Dates

• Paper Submission: February 2, 2015
• Author Notification: February 10, 2015
• Camera Ready Paper Due: February 16, 2015
• Workshop: May 2, 2015

Slides from speakers are now available in the Program.

Invited Speakers

Chih-Jen Lin, National Taiwan University
Title: MF (Matrix Factorization) and FM (Factorization Machines) for Recommender Systems
Abstract: MF (Matrix Factorization) and FM (Factorization Machines) are both effective methods for Recommender Systems. In the first part of this talk, we focus on MF that assumes the ratings from users to items are the only given information. We discuss our recent efforts in developing a parallel package LIBMF for shared-memory systems. Currently, stochastic gradient (SG) method is one of the most popular algorithms to solve the optimization problem for MF. However, as a sequential approach, SG is difficult to be parallelized. We carefully reduce the cache-miss rate and address the load balance of threads to have an effective parallel SG implementation. Further, because the performance of SG highly depends on the setting of learning rates, we develop some adaptive approaches to achieve fast and stable convergence. Experiments show that our implementation outperforms available parallel matrix factorization packages. In the second part of the talk, we consider situations where some user or item features are also available. FM (Factorization Machines) is a useful model in such situations. Recently an extension of FM called FFM (Field-aware FM) has been shown to be very effective for CTR predictions in computational advertising. We discuss our group members' winning approaches for two Kaggle competitions on CTR predictions. Our work has been released in another package LIBFFM for public use.
Bio: Chih-Jen Lin is currently a distinguished professor at the Department of Computer Science, National Taiwan University. He obtained his B.S. degree from National Taiwan University in 1993 and Ph.D. degree from University of Michigan in 1998. His major research areas include machine learning, data mining, and numerical optimization. He is best known for his work on support vector machines (SVM) for data classification. His software LIBSVM is one of the most widely used and cited SVM packages. For his research work he has received many awards, including the ACM KDD 2010 and ACM RecSys 2013 best paper awards. He is an IEEE fellow, a AAAI fellow, and an ACM distinguished scientist for his contribution to machine learning algorithms and software design. More information about him can be found here.

Martin Ester, Simon Fraser University
Title: Probabilistic Graphical Models for Recommendation in Social Media
Abstract: In this talk, we will take a closer look at two key ingredients of social media, social networks and location-based services, and investigate how to model their properties for effective item and location recommendation. In the social sciences, the effects of social influence (friends tend to become more similar to each other), and homophily or selection (people tend to befriend similar people) have been identified as drivers of the dynamics of social networks. In the first part of the talk we will discuss methods for exploiting these effects for improving the accuracy of item recommendation in social networks. In the second main part of the talk, we will explore another key aspect of social media, namely location-based social networks (LBSN). A key effect in LBSN is geographical influence, which means that nearby locations have more similar features than far away locations. Location recommendation is different in nature from traditional item recommendation, since the check-in at a physical location requires more commitment from a user than the adoption of an item such as a movie. We will present probabilistic graphical models to capture these properties for location recommendation. The talk will discuss related research with a focus on our own work. In the conclusion, we will outline interesting directions for future research in the fascinating field of recommendation in social media.
Bio: Martin Ester received a PhD in Computer Science from ETH Zurich, Switzerland, in 1990 with a thesis on knowledge-based systems and logic programming. He has been working for Swissair developing expert systems before he joined University of Munich as an Assistant Professor in 1993. Since November 2001, he has been an Associate Professor, now Full Professor at the School of Computing Science of Simon Fraser University, where he co-directs the Database and Data mining research lab. He has published extensively in the top conferences and journals of his field such as KDD, ICDM and TKDE, and his work has been very well-cited. His most famous paper on DBSCAN received more than 6600 citations, and his H-index is 47. He recently served as PC Co-Chair of ACM/IEEE ASONAM 2014 and ACM RecSys 2014. His current research interests include social network analysis, recommender systems, opinion mining, biological network analysis and high-throughput sequence data analysis.

Jieping Ye, University of Michigan
Title: Orthogonal Rank-One Matrix Pursuit for Low Rank Matrix Completion
Abstract: Low rank matrix learning has recently attracted significant attentions in machine learning and data mining due to its wide range of applications, such as collaborative filtering, dimensionality reduction, compressed sensing, and multi-task learning. In this talk, we consider the general form of low rank matrix completion: given a partially observed real-valued matrix Y, we aim to find a matrix X with minimum rank that best approximates the matrix Y on the observed elements. As it is intractable to minimize the matrix rank exactly in the general case, many approximate solutions have been proposed to attack the problem. A widely used convex relaxation of matrix rank is the trace norm or the nuclear norm. The resulting optimization problem is, however, computationally expensive for large matrices. In this talk, we present an efficient and scalable low rank matrix completion algorithm. The key idea is to extend the orthogonal matching pursuit method from the vector case to the matrix case. The algorithm achieves a linear convergence rate, which is significantly better than the previous known results. We demonstrate the efficiency and effectiveness of the proposed algorithm on several large-scale datasets, including the Netflix and MovieLens datasets.
Bio: Jieping Ye is an Associate Professor of Computational Science and Bioinformatics at the University of Michigan. He received his Ph.D. degree in Computer Science from University of Minnesota, Twin Cities in 2005. His research interests include machine learning, data mining, and biomedical informatics. He has served as Senior Program Committee/Area Chair/Program Committee Vice Chair of many conferences including NIPS, KDD, IJCAI, ICDM, SDM, ACML, and PAKDD. He serves as an Associate Editor of IEEE Transactions on Pattern Analysis and Machine Intelligence. He won the SCI Young Investigator of the Year Award at ASU in 2007, the SCI Researcher of the Year Award at ASU in 2009, and the NSF CAREER Award in 2010. His papers have been selected for the outstanding student paper at the International Conference on Machine Learning in 2004, the KDD best research paper honorable mention in 2010, the KDD best research paper nomination in 2011 and 2012, the SDM best research paper runner up in 2013, the KDD best research paper runner up in 2013, and the KDD best student paper award in 2014.

Fei Wang, University of Connecticut
Title: Recommendation in Biomedical Informatics
Abstract: Recommendation is a fundamental problem existing in a wide range of applications. In this talk I will introduce some recommendation problems in biomedical informatics and potential solutions, which include (1) personalized treatment recommendation, which aims to recommend the right drug to the right patient; (2) drug repositioning, which aims at identification of the right disease for the reference drug to treat; (3) prediction of drug-drug interactions. I will present the current methods, initial results, and point out the potential challenges and future research directions.
Bio: Fei Wang is currently an associate professor in Department of Computer Science and Engineering, University of Connecticut. Before his current position, he was a Research Staff Member in IBM T. J. Watson Research Center. His research interests include data mining and machine learning algorithms as well as their applications in health and social informatics. He has published more than 100 papers in related venues including KDD, ICML, SIGIR and SDM. He got best research paper honorable mention in ICDM 2010, best paper finalist in SDM 2011, as well as finalist for Marco Romani Best Research paper award in AMIA Translational Summit 2014.

Shenghuo Zhu, Alibaba Group
Title: Relational learning using bilinear models and its application in E-commerce
Abstract: In this talk, I will discuss a few machine learning problems arising from the e-commerce field. Many of those problems, including some recommendation problems, can be addressed by relational learning. I will talk about a simple large scale bilinear model to formulate those relational learning problems. To provide a possible solution to address the large scale computation issues, I will discuss an optimization approach using random projection.
Bio: Shenghuo Zhu joins Alibaba Group in July, 2014. He was a senior researcher at NEC Laboratories America. Before that, he was working on customer behavior research at Amazon.com. He received his Ph.D degree in Computer Science from University of Rochester in 2003. He is currently working on machine learning and e-commerce. In addition he is interested in the areas of computer vision, data mining, information retrieval, social computing, user modeling, machine translation, natural language processing, etc.

Sofus Macskassy, Facebook
Title: Machine Learning at Facebook
Abstract: I will in this talk provide an overview of some of the areas where machine learning is used at Facebook and the impact it is making. I will discuss, at a high level, some of the problems in some of our products, including newsfeed, friend suggestions and ads, and how we use machine learning to address some of these problems.I will then discuss in some greater detail how we use machine learning to predict missing profile fields of users.
Bio: Sofus A. Macskassy is part of the applied machine learning research team at Facebook. He's previously run user modeling at Facebook, been a research faculty at USC/ISI, and been the Director of Fetch Labs. He received his PhD in machine learning/information filtering at Rutgers University. He is passionate about learning about users to better serve them through better filtering, ranking and recommendation. He was the general chair of KDD-2014, serves on the editorial board of JAIR and ML, and is well published at top-tier conferences and journals.