Ron Bekkerman

Disambiguating Web appearances of people in a social network

Say you are looking for information about a particular person. A search engine returns many pages for that persons name but which pages are about the person you care about, and which are about other people who happen to have the same name? Furthermore, if we are looking for multiple people who are related in some way, how can we best leverage this social network? This paper presents two unsupervised frameworks for solving this problem: one based on link structure of the Web pages, another using Agglomerative/Conglomerative Double Clustering (A/CDC)---an application of a recently introduced multi-way distributional clustering method. To evaluate our methods, we collected and hand-labeled a dataset of over 1000 Web pages retrieved from Google queries on 12 personal names appearing together in someones in an email folder. On this dataset our methods outperform traditional agglomerative clustering by more than 20%, achieving over 80% F-measure.

Scaling up machine learning: parallel and distributed approaches

This tutorial gives a broad view of modern approaches for scaling up machine learning and data mining methods on parallel/distributed platforms. Demand for scaling up machine learning is task-specific: for some tasks it is driven by the enormous dataset sizes, for others by model complexity or by the requirement for real-time prediction. Selecting a task-appropriate parallelization platform and algorithm requires understanding their benefits, trade-offs and constraints. This tutorial focuses on providing an integrated overview of state-of-the-art platforms and algorithm choices. These span a range of hardware options (from FPGAs and GPUs to multi-core systems and commodity clusters), programming frameworks (including CUDA, MPI, MapReduce, and DryadLINQ), and learning settings (e.g., semi-supervised and online learning). The tutorial is example-driven, covering a number of popular algorithms (e.g., boosted trees, spectral clustering, belief propagation) and diverse applications (e.g., recommender systems and object recognition in vision). The tutorial is based on (but not limited to) the material from our upcoming Cambridge U. Press edited book which is currently in production. Visit the tutorial website at http://hunch.net/~large_scale_survey/

On feature distributional clustering for text categorization

We describe a text categorization approach that is based on a combination of feature distributional clusters with a support vector machine (SVM) classifier. Our feature selection approach employs distributional clustering of words via the recently introducedinformation bottleneck method, which generates a more efficientword-clusterrepresentation of documents. Combined with the classification power of an SVM, this method yields high performance text categorization that can outperform other recent methods in terms of categorization accuracy and representation efficiency. Comparing the accuracy of our method with other techniques, we observe significant dependency of the results on the data set. We discuss the potential reasons for this dependency.

Multi-way distributional clustering via pairwise interactions

We present a novel unsupervised learning scheme that simultaneously clusters variables of several types (e.g., documents, words and authors) based on pairwise interactions between the types, as observed in co-occurrence data. In this scheme, multiple clustering systems are generated aiming at maximizing an objective function that measures multiple pairwise mutual information between cluster variables. To implement this idea, we propose an algorithm that interleaves top-down clustering of some variables and bottom-up clustering of the other variables, with a local optimization correction routine. Focusing on document clustering we present an extensive empirical study of two-way, three-way and four-way applications of our scheme using six real-world datasets including the 20 News-groups (20NG) and the Enron email collection. Our multi-way distributional clustering (MDC) algorithms consistently and significantly outperform previous state-of-the-art information theoretic clustering algorithms.

Multi-modal Clustering for Multimedia Collections

Most of the online multimedia collections, such as picture galleries or video archives, are categorized in a fully manual process, which is very expensive and may soon be infeasible with the rapid growth of multimedia repositories. In this paper, we present an effective method for automating this process within the unsupervised learning framework. We exploit the truly multi-modal nature of multimedia collections - they have multiple views, or modalities, each of which contributes its own perspective to the collections organization. For example, in picture galleries, image captions are often provided that form a separate view on the collection. Color histograms (or any other set of global features) form another view. Additional views are blobs, interest points and other sets of local features. Our model, called Comraf* (pronounced Comraf-Star), efficiently incorporates various views in multi-modal clustering, by which it allows great modeling flexibility. Comraf* is a light-weight version of the recently introduced combinatorial Markov random field (Comraf). We show how to translate an arbitrary Comraf into a series of Comraf* models, and give an empirical evidence for comparable effectiveness of the two. Comraf* demonstrates excellent results on two real-world image galleries: it obtains 2.5-3 times higher accuracy compared with a uni-modal k-means.

Learning to rank social update streams

As online social media further integrates deeper into our lives, we spend more time consuming social update streams that come from our online connections. Although social update streams provide a tremendous opportunity for us to access information on-the-fly, we often complain about its relevance. Some of us are flooded with a steady stream of information and simply cannot process it in full. Ranking the incoming content becomes the only solution for the overwhelmed users. For some others, in contrast, the incoming information stream is pretty weak, and they have to actively search for relevant information which is quite tedious. For these users, augmenting their incoming content flow with relevant information from outside their first-degree network would be a viable solution. In that case, the problem of relevance becomes even more prominent. In this paper, we start an open discussion on how to build effective systems for ranking social updates from a unique perspective of LinkedIn -- the largest professional network in the world. More specifically, we address this problem as an intersection of learning to rank, collaborative filtering, and clickthrough modeling, while leveraging ideas from information retrieval and recommender systems. We propose a novel probabilistic latent factor model with regressions on explicit features and compare it with a number of non-trivial baselines. In addition to demonstrating superior performance of our model, we shed some light on the nature of social updates on LinkedIn and how users interact with them, which might be applicable to social update streams in general.

High-precision phrase-based document classification on a modern scale

We present a document classification system that employs lazy learning from labeled phrases, and argue that the system can be highly effective whenever the following property holds: most of information on document labels is captured in phrases. We call this property near sufficiency. Our research contribution is twofold: (a) we quantify the near sufficiency property using the Information Bottleneck principle and show that it is easy to check on a given dataset; (b) we reveal that in all practical cases---from small-scale to very large-scale---manual labeling of phrases is feasible: the natural language constrains the number of common phrases composed of a vocabulary to grow linearly with the size of the vocabulary. Both these contributions provide firm foundation to applicability of the phrase-based classification (PBC) framework to a variety of large-scale tasks. We deployed the PBC system on the task of job title classification, as a part of LinkedIns data standardization effort. The system significantly outperforms its predecessor both in terms of precision and coverage. It is currently being used in LinkedIns ad targeting product, and more applications are being developed. We argue that PBC excels in high explainability of the classification results, as well as in low development and low maintenance costs. We benchmark PBC against existing high-precision document classification algorithms and conclude that it is most useful in multilabel classification.

Combinatorial markov random fields

A combinatorial random variable is a discrete random variable defined over a combinatorial set (e.g., a power set of a given set). In this paper we introduce combinatorial Markov random fields (Comrafs), which are Markov random fields where some of the nodes are combinatorial random variables. We argue that Comrafs are powerful models for unsupervised and semi-supervised learning. We put Comrafs in perspective by showing their relationship with several existing models. Since it can be problematic to apply existing inference techniques for graphical models to Comrafs, we design two simple and efficient inference algorithms specific for Comrafs, which are based on combinatorial optimization. We show that even such simple algorithms consistently and significantly outperform Latent Dirichlet Allocation (LDA) on a document clustering task. We then present Comraf models for semi-supervised clustering and transfer learning that demonstrate superior results in comparison to an existing semi-supervised scheme (constrained optimization).

Data weaving: scaling up the state-of-the-art in data clustering

The enormous amount and dimensionality of data processed by modern data mining tools require effective, scalable unsupervised learning techniques. Unfortunately, the majority of previously proposed clustering algorithms are either effective or scalable. This paper is concerned with information-theoretic clustering (ITC) that has historically been considered the state-of-the-art in clustering multi-dimensional data. Most existing ITC methods are computationally expensive and not easily scalable. Those few ITC methods that scale well (using, e.g., parallelization) are often outperformed by the others, of an inherently sequential nature. First, we justify this observation theoretically. We then propose data weaving - a novel method for parallelizing sequential clustering algorithms. Data weaving is intrinsically multi-modal - it allows simultaneous clustering of a few types of data (modalities). Finally, we use data weaving to parallelize multi-modal ITC, which results in proposing a powerful DataLoom algorithm. In our experimentation with small datasets, DataLoom shows practically identical performance compared to expensive sequential alternatives. On large datasets, however, DataLoom demonstrates significant gains over other parallel clustering methods. To illustrate the scalability, we simultaneously clustered rows and columns of a contingency table with over 120 billion entries.

Improving clustering stability with combinatorial MRFs

As clustering methods are often sensitive to parameter tuning, obtaining stability in clustering results is an important task. In this work, we aim at improving clustering stability by attempting to diminish the influence of algorithmic inconsistencies and enhance the signal that comes from the data. We propose a mechanism that takes m clusterings as input and outputs m clusterings of comparable quality, which are in higher agreement with each other. We call our method the Clustering Agreement Process (CAP). To preserve the clustering quality, CAP uses the same optimization procedure as used in clustering. In particular, we study the stability problem of randomized clustering methods (which usually produce different results at each run). We focus on methods that are based on inference in a combinatorial Markov Random Field (or Comraf, for short) of a simple topology. We instantiate CAP as inference within a more complex, bipartite Comraf. We test the resulting system on four datasets, three of which are medium-sized text collections, while the fourth is a large-scale user/movie dataset. First, in all the four cases, our system significantly improves the clustering stability measured in terms of the macro-averaged Jaccard index. Second, in all the four cases our system managed to significantly improve clustering quality as well, achieving the state-of-the-art results. Third, our system significantly improves stability of consensus clustering built on top of the randomized clustering solutions.