Abstract
dc:descriptionThe subject of this doctoral dissertation is the study of filtering and data mining problems in heterogeneous information objects. More specifically, the research focuses on how various filtering and data mining techniques can be used to solve significant problems arising from the rapid expansion of the internet, the growing demand for improved and personalized services for users, and the need to maintain the software systems that support this global explosion of information usage. Although these problems appear to be heterogeneous, they all stem from the overarching need to support and manage the rapid global spread of web-based information. Data mining techniques, with their broad applicability across different types of data, present an ideal solution to the aforementioned challenges. In recent years, substantial research has been conducted on how to apply these techniques to web usage data as well as XML data. Furthermore, there is growing interest in their application to software evaluation, where the source code files of a software system serve as the data source, in order to detect and address issues early. XML (eXtensible Markup Language) has become the standard for data representation on the Web. The explosive growth of data represented in XML created a need for efficient search within the hierarchical structure of XML documents to locate specific information. This requirement, along with the need for fast access to XML tree nodes, led to the development of specialized indexes. These indexes must be dynamic to adapt to the evolving nature of the data. Additionally, due to the need to retrieve specific information, it is essential to filter XML datasets using predefined patterns and rules to identify data matching these criteria. At the same time, the clustering of large volumes of XML data for further processing necessitates intelligent and efficient grouping algorithms, which must be tailored to both the semi-structured nature of XML and the actual content it represents. Moreover, the vast amount of data to be processed makes distributed management in peer-to-peer (P2P) networks increasingly attractive, as it provides advantages such as load balancing and distributed storage. The internet's rapid expansion has also led to efforts to enhance the services offered by websites to better serve their users. Personalizing the presented content requires a thorough study of user behavior as reflected in website usage data (clickstreams). Discovering common behavior patterns among users can only be achieved through the application of suitable data mining algorithms. Additionally, the dimensions of internal, external, and usage quality of a software product evolve throughout its lifecycle. For example, quality at the start of the lifecycle emphasizes external and user-centric aspects, which differ from internal aspects such as design quality that concern software engineers. Data mining techniques aimed at achieving the necessary quality levels (e.g., quality assessment) must take these dimensions into account at each stage of the product lifecycle. In this context, the dissertation's research focuses on identifying the specific challenges posed by XML data and on designing efficient algorithms for filtering, clustering, and managing XML data in peer-to-peer networks. Furthermore, it examines how to process and extract knowledge from web usage data to enable the delivery of personalized information to users. This part of the research emphasizes effective user clustering based on browsing history and the definition of a similarity metric between users using their historical data. Lastly, the dissertation addresses the application of data mining techniques to source code metrics of a software system to assess their quality and identify potential issues early. Research in this area focused on designing a clustering algorithm for source code metrics, enabling the grouping of code components (packages, classes, functions) based on quality, and introducing a new methodology for the overall assessment of a complete software system. Key research contributions include: - Efficient XML data filtering algorithms applicable to publish-subscribe systems for delivering only user-relevant information. -Novel clustering algorithms for large volumes of XML data across different levels using data mining methods. -A new model for managing distributed XML data in peer-to-peer networks for optimal distributed search and retrieval. This model uses new XML clustering algorithms for optimal data distribution and distributed Bloom Filters for more efficient query routing.- A clustering algorithm focused on software quality evaluation metrics to assist system analysts in the early detection and resolution of source code quality issues.- Efficient data mining algorithms for web usage data utilizing multi-level data clustering techniques.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2011
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Antonellis, Panagiotis
- Αντωνέλλης, Παναγιώτης
Subjects
dc:subject × 15- Φιλτράρισμα δεδομένων
- Εξόρυξη δεδομένων
- XML
- Ποιότητα κώδικα
- Data mining
- Information filtering
- Code quality
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Επιστήμη ηλεκτρονικών υπολογιστών
- Επιστήμη ηλεκτρονικών υπολογιστών, θεωρία και μέθοδοι
- Computer and Information Sciences
- Natural Sciences
- Computer Science
- Computer science, theory and methods
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58871