Local collaborative ranking

Recent online services rely heavily on automatic personalization to recommend relevant content to a large number of users. This requires systems to scale promptly to accommodate the stream of new users visiting the online services for the first time. In this work, we propose a content-based recommendation system to address both the recommendation quality and the system scalability. We propose to use a rich feature set to represent users, according to their web browsing history and search queries. We use a Deep Learning approach to map users and items to a latent space where the similarity between users and their preferred items is maximized. We extend the model to jointly learn from features of items from different domains and user features by introducing a multi-view Deep Learning model. We show how to make this rich-feature based user representation scalable by reducing the dimension of the inputs and the amount of training data. The rich user feature representation allows the model to learn relevant user behavior patterns and give useful recommendations for users who do not have any interaction with the service, given that they have adequate search and browsing history. The combination of different domains into a single model for learning helps improve the recommendation quality across all the domains, as well as having a more compact and a semantically richer user latent feature vector. We experiment with our approach on three real-world recommendation systems acquired from different sources of Microsoft products: Windows Apps recommendation, News recommendation, and Movie/TV recommendation. Results indicate that our approach is significantly better than the state-of-the-art algorithms (up to 49% enhancement on existing users and 115% enhancement on new users). In addition, experiments on a publicly open data set also indicate the superiority of our method in comparison with transitional generative topic models, for model- * Work done at Microsoft Research during the author's summer internship. ing cross-domain recommender systems. Scalability analysis show that our multi-view DNN model can easily scale to encompass millions of users and billions of item entries. Experimental results also confirm that combining features from all domains produces much better performance than building separate models for each domain.

show abstract

“…Recently, ranking based objective function has shown to be more effective in giving better recommendation as shown in [11].…”

Section: Related Workmentioning

confidence: 99%

A Multi-View Deep Learning Approach for Cross Domain User Modeling in Recommendation Systems

Elkahky

Song

2015

Proceedings of the 24th International Conference on World Wide Web

588

189

View full text Add to dashboard Cite

show abstract

“…Inspired by the analogy between query-document relations in IR and user-item relations in recommender systems, many CR approaches have been proposed over the past few years for personalized recommendations [5,20,22,29,34,35,36,38,39]. The majority of them optimize one of the (top-N) ranking evaluation metrics by exploiting advances in structured estimation [37] that minimize convex upper bounds of the loss functions based on these metrics [21].…”

Section: Related Workmentioning

confidence: 99%

“…For that, they used a novel approach of generating features for the LTR task using CF techniques, i.e., neighborhood-based and modelbased techniques, so that the problem can be transformed to a classic pairwise LTR problem. Recently, [22] introduced a method for Local Collaborative Ranking (LCR) where ideas of local low-rank matrix approximation were applied to the pairwise ranking loss minimization framework. A special case of this framework is Global Collaborative Ranking (GCR) where global low-rank structure is assumed, as is usually done in the ranking counterpart of Probabilistic Matrix Factorization (PMF)-type methods [1,25,31,33], although equal importance is given to the whole ranked list.…”

Section: Related Workmentioning

confidence: 99%

“…While existing CR models do optimize certain classical ranking metrics [34,35,36,39], most of these metrics are non convex, so one works with a relaxation of the original problem, e.g., CofiRank [39] optimizes a convex upper bound of a loss based on Normalized Discounted Cumulative Gain (NDCG) [17], CLiMF [34] optimizes a smoothed version of the Reciprocal Rank [4] via a lower bound, etc. Moreover, certain CR models [22] consider the entire list, rather than focusing on accuracy at the top of the list.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Collaborative Ranking with a Push at the Top

Christakopoulou

Banerjee

2015

Proceedings of the 24th International Conference on World Wide Web

View full text Add to dashboard Cite

The goal of collaborative filtering is to get accurate recommendations at the top of the list for a set of users. From such a perspective, collaborative ranking based formulations with suitable ranking loss functions are natural. While recent literature has explored the idea based on objective functions such as NDCG or Average Precision, such objectives are difficult to optimize directly. In this paper, building on recent advances from the learning to rank literature, we introduce a novel family of collaborative ranking algorithms which focus on accuracy at the top of the list for each user while learning the ranking functions collaboratively. We consider three specific formulations, based on collaborative p-norm push, infinite push, and reverse-height push, and propose e cient optimization methods for learning these models. Experimental results illustrate the value of collaborative ranking, and show that the proposed methods are competitive, usually better than existing popular approaches to personalized recommendation.

show abstract

“…Latent Factor Models (LFM) based on Matrix Factorization (MF) have gained great popularity because they usually have achieved stateof-the-art performance in some benchmark datasets [18]. A variety of MF algorithms have been proposed, such as Singular Value Decomposition (SVD) [7], Non-negative Matrix Factorization (NMF) [8], Max-Margin Matrix Factorization (MMMF) [14], Probabilistic Matrix Factorization (PMF) [11], and Local Matrix Factorization (LMF) [9]. They aim at learning latent factors from user-item matrices to make recommendation.…”

Section: Related Workmentioning

confidence: 99%

Personalized Recommendation of Item Category Using Ranking on Time-Aware Graphs

Chen

Hou

Nie

et al. 2015

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

SUMMARYRecommendation systems have been widely used in Ecommerce sites, social media and etc. An important recommendation task is to predict items that a user will perform actions on with users' historical data, which is called top-K recommendation. Recently, there is huge amount of emerging items which are divided into a variety of categories and researchers have argued or suggested that top-K recommendation of item category could be very beneficial for users to make better and faster decisions. However, the traditional methods encounter some common but crucial problems in this scenario because additional information, such as time, is ignored. The ranking algorithm on graphs and the increasingly growing amount of online user behaviors shed some light on these problems. We propose a construction method of time-aware graphs to use ranking algorithm for personalized recommendation of item category. Experimental results on real-world datasets demonstrate the advantages of our proposed method over competitive baseline algorithms.

show abstract

Local collaborative ranking

Cited by 112 publications

References 33 publications

A Multi-View Deep Learning Approach for Cross Domain User Modeling in Recommendation Systems

A Multi-View Deep Learning Approach for Cross Domain User Modeling in Recommendation Systems

Collaborative Ranking with a Push at the Top

Personalized Recommendation of Item Category Using Ranking on Time-Aware Graphs

Contact Info

Product

Resources

About