Markus Endres

A preference-based recommender system

The installation of recommender systems in e-applications like online shops is common practice to offer alternative or cross-selling products to their customers. Usually collaborative filtering methods, like e.g. the Pearson correlation coefficient algorithm, are used to detect customers with a similar taste concerning some items. These customers serve as recommenders for other users. In this paper we introduce a novel approach for a recommender system that is based on user preferences, which may be mined from log data in a database system. Our notion of user preferences adopts a very powerful preference model from database systems. An evaluation of our prototype system suggests that our prediction quality can compete with the widely-used Pearson-based approach. In addition, our approach can achieve an added value, because it yields better results when there are only a few recommenders available. As a unique feature, preference-based recommender systems can deal with multi-attribute recommendations.

High Parallel Skyline Computation over Low-Cardinality Domains

A Skyline query retrieves all objects in a dataset that are not dominated by other objects according to some given criteria. Although there are a few parallel Skyline algorithms on multicore processors, it is still a challenging task to fully exploit the advantages of such modern hardware architectures for efficient Skyline computation. In this paper we present high-performance parallel Skyline algorithms based on the lattice structure generated by a Skyline query. We compare our methods with the state-of-the-art algorithms for multicore Skyline processing. Experimental results on synthetic and real datasets show that our new algorithms outperform state-of-the-art multicore Skyline techniques for low-cardinality domains. Our algorithms have linear runtime complexity and fully play on modern hardware architectures.

Scalagon: An Efficient Skyline Algorithm for All Seasons

Skyline queries are well-known in the database community and there are many algorithms for the computation of the Pareto frontier. The most prominent algorithms are based on a block-nested-loop style tuple-to-tuple comparison (BNL). Another approach exploits the lattice structure induced by a Skyline query over low-cardinality domains. In this paper, we present Scalagon, an algorithm which combines the ideas of the lattice approach and a BNL-style algorithm to evaluate Skylines on arbitrary domains. Since multicore processors are going mainstream, we also present a parallel version of Scalagon. We demonstrate through extensive experimentation on synthetic and real datasets that our algorithm can result in a significant performance advantage over existing techniques.

Complex preference queries supporting spatial applications for user groups

Our demo application demonstrates a personalized location-based web application using Preference SQL that allows single users as well as groups of users to find accommodations in Istanbul that satisfy both hard constraints and user preferences. The application assists in defining spatial, numerical, and categorical base preferences and composes complex preference statements in an intuitive fashion. Unlike existing location-based services, the application considers spatial queries as soft instead of hard constraints to determine the best matches which are finally presented on a map. The underlying Preference SQL framework is implemented on top of a database, therefore enabling a seamless application integration with standard SQL back-end systems as well as efficient and extensible preference query processing.

Skyline snippets

There is a strong demand for a deep personalization of search systems for many Internet applications. In this respect the proper handling of user preferences plays an important role. Here we focus on the efficient evaluation of the Pareto preference operator for structured data in very large databases. The result set of such a Pareto query, also known as the “skyline”, tends to become very large for higher dimensionalities. Often it is too time-consuming or just not necessary to compute the entire skyline, instead only some fraction of it, called a “snippet”, is sufficient. In this paper we contribute a novel algorithm for a fast computation of such skyline snippets. Our solutions do not rely on the availability of specialized pre-computed indexes, hence are generally applicable. We demonstrate the performance of our approach by several benchmarks studies. The presented results suggest that even for complex Pareto queries, yielding very large skylines, snippets can be computed sufficiently fast, and therefore can be integrated into online Web services.

Design and Implementation of a Framework for Context-Aware Preference Queries

In this paper we present a framework for a novel kind of context-aware preference query composition whereby queries for the Preference SQL system are created. We choose a commercial e-business platform for outdoor activities as a use case and develop a context model for this domain within our framework. The suggested model considers explicit user input, domain-specific knowledge, contextual knowledge and location-based sensor data in a comprehensive approach. Aside from the theoretical background of preferences, the optimization of preference queries and our novel generator based model we give special attention to the aspects of the implementation and the practical experiences. We provide a sketch of the implementation and summarize our user studies which have been done in a joint project with an industrial partner.

Beyond Skylines: Explicit Preferences

Skyline queries are well-known in the database community and there are many algorithms for the computation of the Pareto frontier. But users do not only think of finding the Pareto optimal objects, they often want to find the best objects concerning an explicit specified preference order. While preferences themselves often are defined as general strict partial orders, almost all algorithms are designed to evaluate Pareto preferences combining weak orders, i.e., Skylines. In this paper, we consider general strict partial orders and we present a method to evaluate such explicit preferences by embedding any strict partial order into a complete lattice. This enables preference evaluation with specialized lattice based algorithms instead of algorithms relying on tuple-to-tuple comparisons and therefore speed-ups their computation as can be seen in our experiments.

Advanced preference query processing for e-commerce

Travel and tourism represents the leading domain for applications in b2c e-commerce. Thus, it deserves highest attention. Due to insufficient search engines, however, arranging a trip on current online travel portals is often not as easy as it seems. We present a novel approach for an advanced processing of database queries dealing with individual as well as global preferences of customers. Our approach also comprises the deployment of existing preference search technology to the tourism domain of electronic commerce. Consequently, the tedious empty-result-effect is avoided. Thereby, a novel, personalized search process delivering custom-tailored products is enabled for tourism and in general for e-commerce, respectively. A first prototype implementing our advanced search has shown promising results.

The Structure of Preference Orders

Preferences are an important natural concept in real life and are well-known in the database and artificial intelligence community. Modeling preferences as strict partial orders closely matches people’s intuition. There are many algorithms for the evaluation of these strict partial orders. In particular some algorithms rely on the total order or the lattice structure constructed by a preference query. This paper provides an overview of the structure of preference orders. We present several measures of the different “better-than graphs” and give a deep insight into the structure of preferences. In fact, a careful analysis of the underlying “better-than graph” enables one to develop efficient algorithms for preference computation.

Algebraic optimization of grouped preference queries

SQL queries containing Group-by are common in data warehouse environments and OLAP. From this the concept of grouped Skyline queries emerged, wherein a Skyline of each group of tuples is requested. Grouped preference queries generalize this kind of Skyline queries. In this paper we present new algebraic transformation rules for grouped preference queries which are one of the most intuitive and practical type of queries. Our optimization laws reduce intermediate result sizes in the computation of joins, Cartesian products, and the preference selection. We have integrated these new rules into our rule-based Preference SQL query optimizer. Our performance benchmarks, building upon the well-known TPC-H and IMDB datasets, show that significant performance gains can be achieved.