Rudolf Mayer

Unsupervised Learning and Clustering

Unsupervised learning is very important in the processing of multimedia content as clustering or partitioning of data in the absence of class labels is often a requirement. This chapter begins with a review of the classic clustering techniques of k-means clustering and hierarchical clustering. Modern advances in clustering are covered with an analysis of kernel-based clustering and spectral clustering. One of the most popular unsupervised learning techniques for processing multimedia content is the self-organizing map, so a review of self-organizing maps and variants is presented in this chapter. The absence of class labels in unsupervised learning makes the question of evaluation and cluster quality assessment more complicated than in supervised learning. So this chapter also includes a comprehensive analysis of cluster validity assessment techniques.

Combination of audio and lyrics features for genre classification in digital audio collections

In many areas multimedia technology has made its way into mainstream. In the case of digital audio this is manifested in numerous online music stores having turned into profitable businesses. The widespread user adaption of digital audio both on home computers and mobile players show the size of this market. Thus, ways to automatically process and handle the growing size of private and commercial collections become increasingly important; along goes a need to make music interpretable by computers. The most obvious representation of audio files is their sound - there are, however, more ways of describing a song, for instance its lyrics, which describe songs in terms of content words. Lyrics of music may be orthogonal to its sound, and differ greatly from other texts regarding their (rhyme) structure. Consequently, the exploitation of these properties has potential for typical music information retrieval tasks such as musical genre classification; so far, there is a lack of means to efficiently combine these modalities. In this paper, we present findings from investigating advanced lyrics features such as the frequency of certain rhyme patterns, several parts-of-speech features, and statistic features such as words per minute (WPM). We further analyse in how far a combination of these features with existing acoustic feature sets can be exploited for genre classification and provide experiments on two test collections.

Visualising class distribution on self-organising maps

The Self-Organising Map is a popular unsupervised neural network model which has been used successfully in various contexts for clustering data. Even though labelled data is not required for the training process, in many applications class labelling of some sort is available. A visualisation uncovering the distribution and arrangement of the classes over the map can help the user to gain a better understanding and analysis of the mapping created by the SOM, e.g. through comparing the results of the manual labelling and automatic arrangement. In this paper, we present such a visualisation technique, which smoothly colours a SOM according to the distribution and location of the given class labels. It allows the user to easier assess the quality of the manual labelling by highlighting outliers and border data close to different classes.

Using Ontologies for Enterprise Architecture Integration and Analysis

Enterprise architecture facilitates the alignment between different domains, such as business, applications and information technology. These domains must be described with description languages that best address the concerns of its stakeholders. However, current model-based enterprise architecture techniques are unable to integrate multiple descriptions languages either due to the lack of suitable extension mechanisms or because they lack the means to maintain the coherence, consistency and traceability between the representations of the multiple domains of the enterprise. On the other hand, enterprise architecture models are often designed and used for communication and not for automated analysis of its contents. Model analysis is a valuable tool for assessing the qualities of a model, such as conformance and completeness, and also for supporting decision making. This paper addresses these two issues found in model-based enterprise architecture: (1) the integration of domain description languages, and (2) the automated analysis of models. This proposal uses ontology engineering techniques to specify and integrate the different domains and reasoning and querying as a means to analyse the models. The utility of the proposal is shown through an evaluation scenario that involve the analysis of an enterprise architecture model that spans multiple domains.

Using ontologies to capture the semantics of a (business) process for digital preservation

IT-supported business processes and computationally intensive science (called e-science) have become increasingly ubiquitous in the last decades. Along with this trend comes the need to make at least the most important of these processes available for the long term, to allow later analysis of their execution, or even a re-execution. As such, the preservation of scientific experiments and their results enables others to reproduce and verify the results as well as build on the result of earlier work. All but the simplest processes require to be described by a multitude of information objects, as well as their interconnections and relations, to be successfully preserved. To enable a semantic description of these objects in a structured manner, we developed a formal meta-model that can be utilised in the digital preservation of a process. The meta-model describes classes of elements and their relations, in the form of ontologies, with a core ontology describing the generic concepts, and extension mechanisms to map supplementary ontologies describing more specific aspects. In this paper, we present the overall architecture and individual ontologies, and motivate their usefulness via the application to use cases from different domains.

Preserving scientific processes from design to publications

Digital Preservation has so far focused mainly on digital objects that are static in their nature, such as text and multimedia documents. However, there is an increasing demand to extend the applications towards dynamic objects and whole processes, such as scientific workflows in the domain of E-Science. This calls for a revision and extension of current concepts, methods and practices. Important questions to address are e.g. what needs to be captured at ingest, how do the digital objects need to be described, which preservation actions are applicable and how can the preserved objects be evaluated. In this paper we present a conceptual model for capturing the required information and show how this can be linked to evaluating the re-invocation of a preserved process.

Combination of feature selection methods for text categorisation

Feature selection plays a vital role in text categorisation. A range of different methods have been developed, each having unique properties and selecting different features. We show some results of an extensive study of feature selection approaches using a wide range of combination methods. We performed experiments on 18 test collections and report a subset of the results.

Improving Scientific Conferences by Enhancing Conference Management Systems with Information Mining Capabilities

In this paper we identify tasks in the field of conference management where methods from the domain of information retrieval, information management and information organization can assist the organizer, the program committee member and the participants. In particular, we focus on tasks, where (1) the quality of the conference can be increased by assisting in the creation of an improved review process by better matching the reviewers expertise with the paper topics, (2) the conference participants profit by allowing them better access to the wealth of information accumulated throughout a conference series and at the same time (3) reducing the workload of conference organizer by partially automating tedious tasks, such as review assignment and the creation of review plans. We report on case studies from a medium-sized (around 400 participants) as well as a large (more than 700 participants) conference in computer science as well as in medical domains.

Feature selection in a cartesian ensemble of feature subspace classifiers for music categorisation

In this paper, we evaluate the impact of feature selection on the classification accuracy and the achieved dimensionality reduction, which benefits the time needed on training classification models. Our classification scheme therein is a Cartesian ensemble classification system, based on the principle of late fusion and feature subspaces. These feature subspaces describe different aspects of the same data set. We use it for the ensemble classification of multiple feature sets from the audio and symbolic domains. We present an extensive set of experiments in the context of music genre classification, based on Music IR benchmark datasets. We show that while feature selection does not benefit classification accuracy, it greatly reduces the dimensionality of each feature subspace, and thus adds to great gains in the time needed to train the individual classification models that form the ensemble.

The Metro Visualisation of Component Planes for Self-Organising Maps

The self-organising map is a popular unsupervised neural network model which has successfully been used for clustering various kinds of data. To help in understanding the influence of single variables or components on clusterings, we introduce a novel method for the visualisation of component planes for SOMs. The approach presented is based on the discretisation of the components and makes use of the well-known metro map metaphor. It depicts consistent values and their ordering across the map for discretisations of various components and their correlations in terms of directions on the map. In our approach component lines are drawn for each component of the data, allowing the combination of numerous component planes into one plot. We also propose a method to further aggregate these component lines, by grouping highly correlated variables, i.e. similar lines on the map. To show the applicability of our approach we provide experimental results for two popular machine learning data sets.