Christoph Boden

Scalable similarity-based neighborhood methods with MapReduce

Similarity-based neighborhood methods, a simple and popular approach to collaborative filtering, infer their predictions by finding users with similar taste or items that have been similarly rated. If the number of users grows to millions, the standard approach of sequentially examining each item and looking at all interacting users does not scale. To solve this problem, we develop a MapReduce algorithm for the pairwise item comparison and top-N recommendation problem that scales linearly with respect to a growing number of users. This parallel algorithm is able to work on partitioned data and is general in that it supports a wide range of similarity measures. We evaluate our algorithm on a large dataset consisting of 700 million song ratings from Yahoo! Music.

Large-scale social-media analytics on stratosphere

The importance of social-media platforms and online communities - in business as well as public context - is more and more acknowledged and appreciated by industry and researchers alike. Consequently, a wide range of analytics has been proposed to understand, steer, and exploit the mechanics and laws driving their functionality and creating the resulting benefits. However, analysts usually face significant problems in scaling existing and novel approaches to match the data volume and size of modern online communities. In this work, we propose and demonstrate the usage of the massively parallel data processing system Stratosphere, based on second order functions as an extended notion of the MapReduce paradigm, to provide a new level of scalability to such social-media analytics. Based on the popular example of role analysis, we present and illustrate how this massively parallel approach can be leveraged to scale out complex data-mining tasks, while providing a programming approach that eases the formulation of complete analytical workflows.

Beyond search: Retrieving complete tuples from a text-database

A common task of Web users is querying structured information from Web pages. For realizing this interesting scenario we propose a novel query processor for systematically discovering instances of semantic relations in Web search results and joining these relation instances into complex result tuples with conjunctive queries. Our query processor transforms a structured user query into keyword queries that are submitted to a search engine, forwards search results to a relation extractor, and then combines relations into complex result tuples. The processor automatically learns discriminative and effective keywords for different types of semantic relations. Thereby, our query processor leverages the index of a search engine to query potentially billions of pages. Unfortunately, relation extractors may fail to return a relation for a result tuple. Moreover, user defined data sources may not return at least k complete result tuples. Therefore we propose an adaptive routing model based on information theory for retrieving missing attributes of incomplete result tuples. The model determines the most promising next incomplete tuple and attribute type for returning any-k complete result tuples at any point during the query execution process. We report a thorough experimental evaluation over multiple relation extractors. Our query processor returns complete result tuples while processing only very few Web pages.

Classification algorithms for relation prediction

Knowledge discovery from the Web is a cyclic process. In this paper we focus on the important part of transforming unstructured information from Web pages into structured relations. Relation extraction systems capture information from natural language text on Web pages, called Web text. However, extraction is quite costly and time consuming. Worse, many Web pages may not contain a textual representation of a relation that the extractor can capture. As a result many irrelevant pages are processed by relation extractors.

Benchmarking Data Flow Systems for Scalable Machine Learning

Distributed data flow systems such as Apache Spark or Apache Flink are popular choices for scaling machine learning algorithms in production. Industry applications of large scale machine learning such as click-through rate prediction rely on models trained on billions of data points which are both highly sparse and high-dimensional. Existing Benchmarks attempt to assess the performance of data flow systems such as Apache Flink, Spark or Hadoop with non-representative workloads such as WordCount, Grep or Sort. They only evaluate scalability with respect to data set size and fail to address the crucial requirement of handling high dimensional data. We introduce a representative set of distributed machine learning algorithms suitable for large scale distributed settings which have close resemblance to industry-relevant applications and provide generalizable insights into system performance. We implement mathematically equivalent versions of these algorithms in Apache Flink and Apache Spark, tune relevant system parameters and run a comprehensive set of experiments to assess their scalability with respect to both: data set size and dimensionality of the data. We evaluate the systems for data up to four billion data points and 100 million dimensions. Additionally we compare the performance to single-node implementations to put the scalability results into perspective. Our results indicate that while being able to robustly scale with increasing data set sizes, current state of the art data flow systems are surprisingly inefficient at coping with high dimensional data, which is a crucial requirement for large scale machine learning algorithms.

Extracting a Repository of Events and Event References from News Clusters

In this paper, we prose to build a repository of events and event references from clusters of news articles. We present an automated approach that is based on the hypothesis that if two sentences are a) found in the same cluster of news articles and b) contain temporal expressions that reference the same point in time, they are likely to refer to the same event. This allows us to group similar sentences together and apply open-domain Information Extraction (OpenIE) methods to extract lists of textual references for each detected event. We outline our proposed approach and present a preliminary evaluation in which we extract events and references from 20 clusters of online news. Our experiments indicate that for the largest part our hypothesis holds true, pointing to a strong potential for applying our approach to building an event repository. We illustrate cases in which our hypothesis fails and discuss ways for addressing sources or errors.

Fact-Aware Document Retrieval for Information Extraction

Exploiting textual information from large document collections such as the Web with structured queries is an often requested, but still unsolved requirement of many users. We present BlueFact, a framework for efficiently retrieving documents containing structured, factual information from a full-text index. This is an essential building block for information extraction systems that enable ad-hoc analytical queries on unstructured text data as well as knowledge harvesting in a digital archive scenario.Our approach is based on the observation that documents share a set of common grammatical structures and words for expressing facts. Our system observes these keyword phrases using structural, syntactic, lexical and semantic features in an iterative, cost effective training process and systematically queries the search engine index with these automatically generated phrases. Next, BlueFact retrieves a list of document identifiers, combines observed keywords as evidence for a factual information and infers the relevance for each document identifier. Finally, we forward the documents in the order of their estimated relevance to an information extraction service. That way BlueFact can efficiently retrieve all the structured, factual information contained in an indexed collection of text documents.We report results of a comprehensive experimental evaluation over 20 different fact types on the Reuters News Corpus Volume I (RCV1). BlueFact’s scoring model and feature generation methods significantly outperform existing approaches in terms of fact retrieval performance. BlueFact fires significantly fewer queries against the index, requires significantly less execution time and achieves very high fact recall across different domains.

The Berlin Big Data Center (BBDC)

Abstract The last decade has been characterized by the collection and availability of unprecedented amounts of data due to rapidly decreasing storage costs and the omnipresence of sensors and data-producing global online-services. In order to process and analyze this data deluge, novel distributed data processing systems resting on the paradigm of data flow such as Apache Hadoop, Apache Spark, or Apache Flink were built and have been scaled to tens of thousands of machines. However, writing efficient implementations of data analysis programs on these systems requires a deep understanding of systems programming, prohibiting large groups of data scientists and analysts from efficiently using this technology. In this article, we present some of the main achievements of the research carried out by the Berlin Big Data Cente (BBDC). We introduce the two domain-specific languages Emma and LARA, which are deeply embedded in Scala and enable declarative specification and the automatic parallelization of data analysis programs, the PEEL Framework for transparent and reproducible benchmark experiments of distributed data processing systems, approaches to foster the interpretability of machine learning models and finally provide an overview of the challenges to be addressed in the second phase of the BBDC.

PEEL: A Framework for Benchmarking Distributed Systems and Algorithms

During the last decade, a multitude of novel systems for scalable and distributed data processing has been proposed in both academia and industry. While there are published results of experimental evaluations for nearly all systems, it remains a challenge to objectively compare different system’s performance. It is thus imperative to enable and establish benchmarks for these systems. However, even if workloads and data sets or data generators are fixed, orchestrating and executing benchmarks can be a major obstacle. Worse, many systems come with hardware-dependent parameters that have to be tuned and spawn a diverse set of configuration files. This impedes portability and reproducibility of benchmarks. To address these problems and to foster reproducible and portable experiments and benchmarks of distributed data processing systems, we present PEEL, a framework to define, execute, analyze, and share experiments. PEEL enables the transparent specification of benchmarking workloads and system configuration parameters. It orchestrates the systems involved and automatically runs and collects all associated logs of experiments. PEEL currently supports Apache HDFS, Hadoop, Flink, and Spark and can easily be extended to include further systems.