Vanja Josifovski

Distributed large-scale natural graph factorization

Natural graphs, such as social networks, email graphs, or instant messaging patterns, have become pervasive through the internet. These graphs are massive, often containing hundreds of millions of nodes and billions of edges. While some theoretical models have been proposed to study such graphs, their analysis is still difficult due to the scale and nature of the data.n We propose a framework for large-scale graph decomposition and inference. To resolve the scale, our framework is distributed so that the data are partitioned over a shared-nothing set of machines. We propose a novel factorization technique that relies on partitioning a graph so as to minimize the number of neighboring vertices rather than edges across partitions. Our decomposition is based on a streaming algorithm. It is network-aware as it adapts to the network topology of the underlying computational hardware. We use local copies of the variables and an efficient asynchronous communication protocol to synchronize the replicated values in order to perform most of the computation without having to incur the cost of network communication. On a graph of 200 million vertices and 10 billion edges, derived from an email communication network, our algorithm retains convergence properties while allowing for almost linear scalability in the number of computers.

Top-k publish-subscribe for social annotation of news

Social content, such as Twitter updates, often have the quickest first-hand reports of news events, as well as numerous commentaries that are indicative of public view of such events. As such, social updates provide a good complement to professionally written news articles. In this paper we consider the problem of automatically annotating news stories with social updates (tweets), at a news website serving high volume of pageviews. The high rate of both the pageviews (millions to billions a day) and of the incoming tweets (more than 100 millions a day) make real-time indexing of tweets ineffective, as this requires an index that is both queried and updated extremely frequently. The rate of tweet updates makes caching techniques almost unusable since the cache would become stale very quickly. n nWe propose a novel architecture where each story is treated as a subscription for tweets relevant to the storys content, and new algorithms that efficiently match tweets to stories, proactively maintaining the top-k tweets for each story. Such top-k pub-sub consumes only a small fraction of the resource cost of alternative solutions, and can be applicable to other large scale content-based publish-subscribe problems. We demonstrate the effectiveness of our approach on realworld data: a corpus of news stories from Yahoo! News and a log of Twitter updates.

Taxonomy discovery for personalized recommendation

Personalized recommender systems based on latent factor models are widely used to increase sales in e-commerce. Such systems use the past behavior of users to recommend new items that are likely to be of interest to them. However, latent factor model suffer from sparse user-item interaction in online shopping data: for a large portion of items that do not have sufficient purchase records, their latent factors cannot be estimated accurately. In this paper, we propose a novel approach that automatically discovers the taxonomies from online shopping data and jointly learns a taxonomy-based recommendation system. Out model is non-parametric and can learn the taxonomy structure automatically from the data. Since the taxonomy allows purchase data to be shared between items, it effectively improves the accuracy of recommending tail items by sharing strength with the more frequent items. Experiments on a large-scale online shopping dataset confirm that our proposed model improves significantly over state-of-the-art latent factor models. Moreover, our model generates high-quality and human readable taxonomies. Finally, using the algorithm-generated taxonomy, our model even outperforms latent factor models based on the human-induced taxonomy, thus alleviating the need for costly manual taxonomy generation.

Latent factor models with additive and hierarchically-smoothed user preferences

Items in recommender systems are usually associated with annotated attributes: for e.g., brand and price for products; agency for news articles, etc. Such attributes are highly informative and must be exploited for accurate recommendation. While learning a user preference model over these attributes can result in an interpretable recommender system and can hands the cold start problem, it suffers from two major drawbacks: data sparsity and the inability to model random effects. On the other hand, latent-factor collaborative filtering models have shown great promise in recommender systems; however, its performance on rare items is poor. In this paper we propose a novel model LFUM, which provides the advantages of both of the above models. We learn user preferences (over the attributes) using a personalized Bayesian hierarchical model that uses a combination(additive model) of a globally learned preference model along with user-specific preferences. To combat data-sparsity, we smooth these preferences over the item-taxonomy using an efficient forward-filtering and backward-smoothing inference algorithm. Our inference algorithms can handle both discrete attributes (e.g., item brands) and continuous attributes (e.g., item prices). We combine the user preferences with the latent-factor models and train the resulting collaborative filtering system end-to-end using the successful BPR ranking algorithm. In our extensive experimental analysis, we show that our proposed model outperforms several commonly used baselines and we carry out an ablation study showing the benefits of each component of our model.

Scalable K-Means by ranked retrieval

The k-means clustering algorithm has a long history and a proven practical performance, however it does not scale to clustering millions of data points into thousands of clusters in high dimensional spaces. The main computational bottleneck is the need to recompute the nearest centroid for every data point at every iteration, aprohibitive cost when the number of clusters is large. In this paper we show how to reduce the cost of the k-means algorithm by large factors by adapting ranked retrieval techniques. Using a combination of heuristics, on two real life data sets the wall clock time per iteration is reduced from 445 minutes to less than 4, and from 705 minutes to 1.4, while the clustering quality remains within 0.5% of the k-means quality. The key insight is to invert the process of point-to-centroid assignment by creating an inverted index over all the points and then using the current centroids as queries to this index to decide on cluster membership. In other words, rather than each iteration consisting of points picking centroids, each iteration now consists of centroids picking points. This is much more efficient, but comes at the cost of leaving some points unassigned to any centroid. We show experimentally that the number of such points is low and thus they can be separately assigned once the final centroids are decided. To speed up the computation we sparsify the centroids by pruning low weight features. Finally, to further reduce the running time and the number of unassigned points, we propose a variant of the WAND algorithm that uses the results of the intermediate results of nearest neighbor computations to improve performance.

Up next: retrieval methods for large scale related video suggestion

The explosive growth in sharing and consumption of the video content on the web creates a unique opportunity for scientific advances in video retrieval, recommendation and discovery. In this paper, we focus on the task of video suggestion, commonly found in many online applications. The current state-of-the-art video suggestion techniques are based on the collaborative filtering analysis, and suggest videos that are likely to be co-viewed with the watched video. In this paper, we propose augmenting the collaborative filtering analysis with the topical representation of the video content to suggest related videos. We propose two novel methods for topical video representation. The first method uses information retrieval heuristics such as tf-idf, while the second method learns the optimal topical representations based on the implicit user feedback available in the online scenario. We conduct a large scale live experiment on YouTube traffic, and demonstrate that augmenting collaborative filtering with topical representations significantly improves the quality of the related video suggestions in a live setting, especially for categories with fresh and topically-rich video content such as news videos. In addition, we show that employing user feedback for learning the optimal topical video representations can increase the user engagement by more than 80% over the standard information retrieval representation, when compared to the collaborative filtering baseline.

Focused matrix factorization for audience selection in display advertising

Audience selection is a key problem in display advertising systems in which we need to select a list of users who are interested (i.e., most likely to buy) in an advertising campaign. The users past feedback on this campaign can be leveraged to construct such a list using collaborative filtering techniques such as matrix factorization. However, the user-campaign interaction is typically extremely sparse, hence the conventional matrix factorization does not perform well. Moreover, simply combining the users feedback from all campaigns does not address this since it dilutes the focus on target campaign in consideration. To resolve these issues, we propose a novel focused matrix factorization model (FMF) which learns users preferences towards the specific campaign products, while also exploiting the information about related products. We exploit the product taxonomy to discover related campaigns, and design models to discriminate between the users interest towards campaign products and non-campaign products. We develop a parallel multi-core implementation of the FMF model and evaluate its performance over a real-world advertising dataset spanning more than a million products. Our experiments demonstrate the benefits of using our models over existing approaches.

Towards a robust modeling of temporal interest change patterns for behavioral targeting

Modern web-scale behavioral targeting platforms leverage historical activity of billions of users to predict user interests and inclinations, and consequently future activities. Future activities of particular interest involve purchases or transactions, and are referred to as conversions. Unlike ad-clicks, conversions directly translate to advertisers revenue, and thus provide a very concrete metric for return on advertising investment. A typical behavioral targeting system faces two main challenges: the web-scale amounts of user histories to process on a daily basis, and the relative sparsity of conversions (compared to clicks in a traditional setting). These challenges call for generation of effective and efficient user profiles. Most existing works use the historical intensity of a users interest in various topics to model future interest. In this paper we explore how the change in user behavior can be used to predict future actions and show how it complements the traditional models of decaying interest and action recency to build a complete picture about the user interests and better predict conversions. Our evaluation over a real-world set of campaigns indicates that the combination of change of interest, decaying intensity, and action recency helps in: 1) scoring significant improvements in optimizing for conversions over traditional baselines, 2) substantially improving the targeting efficiency for campaigns with highly sparse conversions, and 3) highly reducing the overall history sizes used in targeting. Furthermore, our techniques have been deployed to production and scored a substantial improvement in targeting performance while imposing a negligible overhead in terms of overall platform running time.