Dan Vodislav

Continuous Top-k Processing of Social Network Information Streams: A Vision

With the huge popularity of social networks, publishing and consuming content through information streams is nowadays at the heart of the new Web. Top-k queries over the streams of interest allow limiting results to relevant content, while continuous processing of such queries is the most effective approach in large scale systems. Current systems fail in combining continuous top-k processing with rich scoring models including social network criteria. We present in this paper our vision on the possible features of a social network of information streams, with a rich scoring model compatible with continuous top-k processing.

Continuous Top-k Queries in Social Networks

Information streams provide today a prevalent way of publishing and consuming content on the Web, especially due to the great success of social networks. Top-k queries over the streams of interest allow limiting results to the most relevant content, while continuous processing of such queries is the most effective approach in large scale systems. However, current systems fail in combining continuous top-k processing with rich scoring models including social network criteria. We present here the SANTA algorithm, able to handle scoring functions including content similarity, but also social network criteria and events in a continuous processing of top-k queries. We propose a variant (SANTA+) that accelerates the processing of interaction events in social networks. We compare SANTA/SANTA+ with an extension of a state-of-the-art algorithm and report a rich experimental study of our approach.

RDF Graph Summarization Based on Approximate Patterns

The Linked Open Data (LOD) cloud brings together information described in RDF and stored on the web in (possibly distributed) RDF Knowledge Bases (KBs). The data in these KBs are not necessarily described by a known schema and many times it is extremely time consuming to query all the interlinked KBs in order to acquire the necessary information. But even when the KB schema is known, we need actually to know which parts of the schema are used. We solve this problem by summarizing large RDF KBs using top-K approximate RDF graph patterns, which we transform to an RDF schema that describes the contents of the KB. This schema describes accurately the KB, even more accurately than an existing schema because it describes the actually used schema, which corresponds to the existing data. We add information on the number of various instances of the patterns, thus allowing the query to estimate the expected results. That way we can then query the RDF graph summary to identify whether the necessary information is present and if it is present in significant numbers whether to be included in a federated query result.

Diversity-Aware Continuous Top- k Queries in Social Networks

We consider here the problem of adding diversity requirements for the results of continuous top-k queries in a large scale social network, while preserving an efficient, continuous processing. We propose the DA-SANTA algorithm, which smoothly adds content diversity to the continuous processing of top-k queries at the social network scale. The experimental study demonstrates the very good properties in terms of effectiveness and efficiency of this algorithm.

Summarizing Linked Data RDF Graphs Using Approximate Graph Pattern Mining.

The Linked Open Data (LOD) cloud brings together information described in RDF and stored on the web in (possibly distributed) RDF Knowledge Bases (KBs). The data in these KBs are not necessarily described by a known schema and many times it is extremely time consuming to query all the interlinked KBs in order to acquire the necessary information. To tackle this problem, we propose a method of summarizing large RDF KBs using approximate RDF graph patterns and calculating the number of instances covered by each pattern. Then we transform the patterns to an RDF schema that describes the contents of the KB. Thus we can then query the RDF graph summary to identify whether the necessary information is present and if so its size, before deciding to include it in a federated query result.

Exact and Approximate Generic Multi-criteria Top-k Query Processing

Many algorithms for multi-criteria top-k query processing with ranking predicates have been proposed, but little effort has been directed toward genericity, i.e. supporting any type of access to the lists of predicate scores (sorted and/or random), or any access cost settings. In this paper we propose a general approach to exact and approximate generic top-k processing. To this end, we propose a general framework (GF) for generic top-k processing, able to express any top-k algorithm and present within this framework a first comparison between generic algorithms. In previous work, we proposed BreadthRefine (BR), a generic algorithm that considers the current top-k candidates as a whole instead of focusing on the best candidate for score refinement, then we compared it with specific top-k algorithms. In this paper, we propose two variants of existing generic strategies and experimentally compare them with the BR breadth-first strategy, showing that BR leads to better execution costs. We also extend the notion of θ-approximation to the GF framework and present a first experimental study of the approximation potential of top-k algorithms on early stopping.

Multi-criteria search algorithm: An efficient approximate k-NN algorithm for image retrieval

We propose a new method for approximate k-NN search in large scale image databases, based on top-k multi-criteria search techniques. The method defines a simple index structure based on sorted lists, which provides a good compromise between fast retrieval, storage requirements and update cost. The search algorithm delivers approximate results with guarantees about false negatives, with fast emergence of good approximations, monotonically improved and leading if necessary to an exact result. Experiments with the on-disk implementation show that our method produces very good approximate results several times faster than the Baseline method.

Generic Top-k Query Processing with Breadth-First Strategies

Many algorithms for top-k query processing with ranking predicates have been proposed, but little effort has been directed toward genericity, i.e. supporting any type sorted and/or random or cost settings for the access to the lists of predicate scores. In previous work, we proposed BreadthRefine BR, a generic algorithm that considers the current top-k candidates as a whole instead of focusing on the best candidate, then we compared it with specific top-k algorithms. In this paper, we compare the BR breadth-first strategy with other existing generic strategies and experimentally show that BR leads to better execution costs. To this end, we propose a general framework GF for generic top-k processing, able to express any top-k algorithm and present within this framework a first comparison between generic algorithms. We also extend the notion of i¾?-approximation to the GF framework and present a first experimental study of the approximation potential of top-k algorithms on early stopping.