Łukasz Skonieczny

Hierarchical Document Clustering Using Frequent Closed Sets

FIHC is a novel data mining algorithm for hierarchical grouping of text documents. It uses frequent term sets to identify clusters. The approach fails when the number of frequent sets of terms is large. In order to overcome this problem, we propose a method for generating the FIHCs hierarchy of clusters by using only frequent closed sets. The new approach is validated on a number of Cluto package datasets. The experiments prove that our version of FIHC is faster up to two orders of magnitude than the original.

Faster Clustering with DBSCAN

Grouping data into meaningful clusters belongs to important tasks in the area of artificial intelligence and data mining. DBSCAN is recognized as a high quality scalable algorithm for clustering data. It enables determination of clusters of any shape and identification of noise data. In this paper, we propose a method improving the performance of DBSCAN. The usefulness of the method is verified experimentally both for indexed and non-indexed data.

Development of a University Knowledge Base

This paper summarizes deployment and a one-year-exploitation of repository platform for knowledge and scientific resources. The platform was developed under SYNAT project and designed for the purpose of Warsaw University of Technology. First we present functionalities of the platform with accordance to the SYNAT objectives and the requirements of Warsaw University of Technology. Those functionalities are then compared to other well established systems like dLibra, Fedora Commons Repository and DSpace. The architecture of the platform and the aspect of integration with other systems are also discussed. Finally we discuss how the platform can be reused for deployment for the purposes of any other university with respect to the specific resources it produces and archives.

AI Platform for Building University Research Knowledge Base

This paper is devoted to the 3-years research performed at Warsaw University of Technology, aimed at building of an advanced software for university research knowledge base. As a result, a text mining platform has been built, enabling research in the areas of text mining and semantic information retrieval. In the paper some of the implemented methods are tested from the point of view of their applicability in a real life system.

RDF Graph Partitions: A Brief Survey

The paper presents justifications and solutions for RDF graph partitioning. It uses an approach from the classical theory of graphs to deal with this problem. We present four ways to transform an RDF graph to a classical graph. We show how to apply solutions from the theory of graphs to RDF graphs. We also perform an experimental evaluation using the gpmetis algorithm (a recognized graph partitioner) on both real and synthetic RDF graphs and prove its practical usability.

University Knowledge Base: Two Years of Experience

This chapter is devoted to the 2-years development and exploitation of the repository platform built at Warsaw University of Technology for the purpose of gathering University research knowledge. The platform has been developed under the SYNAT project, aimed at building nation-wide scientific information infrastructure. The implementation of the platform in the form of the advanced information system is discussed. New functionalities of the knowledge base are presented.

Mining for Unconnected Frequent Graphs with Direct Subgraph Isomorphism Tests

In the paper we propose the algorithm which discovers both connected and unconnected frequent graphs from the graphs set. Our approach is based on depth first search candidate generation and direct execution of subgraph isomorphismtest over database. Several search space pruning techniques are also proposed. Due to lack of unconnected graph mining algorithms we compare our algorithm with two general techniques which make unconnected graph discovery possible by means of connected graph mining algorithms.We also perform undirected comparison of our algorithm with connected graph mining algorithms by comparing the number of discovered frequent subgraphs per second. Finally we derive a connected graph mining algorithm from our algorithm and show that it is competitive (though not winning) with popular connected graph mining algorithms.

Towards Robust Evaluation of Super-Resolution Satellite Image Reconstruction

Super-resolution reconstruction (SRR) consists in processing an image or a bunch of images to generate a new image of higher spatial resolution. This problem has been intensively studied, but seldom is SRR applied in practice for satellite data. In this paper, we briefly review the state of the art on SRR algorithms and we argue that commonly adopted strategies for their evaluation do not reflect the operational conditions. We report our study on assessing the SRR outcome, relying on new quantitative measures. The obtained results allow us to outline the most important research pathways to improve the performance of SRR.

Fast Discovery of Generalized Sequential Patterns

Knowledge in the form of generalized sequential patterns finds many applications. In this paper, we focus on optimizing GSP, which is a well-known algorithm for discovering such patterns. Our optimization consists in more selective identification of nodes to be visited while traversing a hash tree with candidates for generalized sequential patterns. It is based on the fact that elements of candidate sequences are stored as ordered sets of items. In order to reduce the number of visited nodes in the hash tree, we also propose to use not only parameters windowSize and maxGap as in original GSP, but also parameter minGap. As a result of our optimization, the number of candidates that require final time-consuming verification may be considerably decreased. In the experiments we have carried out, our optimized variant of GSP was several times faster than standard GSP.

B4MultiSR: A Benchmark for Multiple-Image Super-Resolution Reconstruction.

Super-resolution reconstruction (SRR) methods consist in processing single or multiple images to increase their spatial resolution. Deployment of such techniques is particularly important, when high resolution image acquisition is associated with high cost or risk, like for medical or satellite imaging. Unfortunately, the existing SRR techniques are not sufficiently robust to be deployed in real-world scenarios, and no real-life benchmark to validate multiple-image SRR has been published so far. As gathering a set of images presenting the same scene at different spatial resolution is not a trivial task, the SRR methods are evaluated based on different assumptions, employing various metrics and datasets, often without using any ground-truth data. In this paper, we introduce a new multi-layer benchmark dataset for systematic evaluation of multiple-image SRR techniques with particular reference to satellite imaging. We hope that the new benchmark will help the researchers to improve the state of the art in SRR, making it suitable for real-world applications.