Athanasios Papagelis

Algorithmic Aspects of Web Intelligent Systems

We discuss the algorithmic aspects of a Web intelligent system which demonstrate several challenges encountered during the development of such systems on the Web. The particular system is an online fun portal that adopts several key ideas spanning from site design to user walk-through and extended user-input analysis, in order to intelligently adapt to user interests and, consequently, improve user experience. Our algorithmic techniques build upon concepts from machine learning and statistics (naive Bayes analysis and user models), as well as several ideas for efficient and timely data manipulation (incremental algorithms and extended caching), and they allow us to efficiently handle and analyze the collected data.

Data set operations to hide decision tree rules

This paper focuses on preserving the privacy of sensitive patterns when inducing decision trees. We adopt a record augmentation approach for hiding sensitive classification rules in binary datasets. Such a hiding methodology is preferred over other heuristic solutions like output perturbation or cryptographic techniques - which restrict the usability of the data - since the raw data itself is readily available for public use. We show some key lemmas which are related to the hiding process and we also demonstrate the methodology with an example and an indicative experiment using a prototype hiding tool.

Enabling Social Navigation on the Web

For a place that gathers millions of people the Web seems pretty lonely at times. This is mainly due to the current predominant browsing scenario; that of an individual participating in an autonomous surfing session. We believe that people should be seen as an integral part of the browsing and searching activity towards a concept known as social navigation. In this work, we extend the typical Web browserpsilas functionality so as to raise awareness of other people having similar Web surfing goals at the current moment. We further present features and algorithms that facilitate online communication and collaboration towards common searching targets. The utility of our system is established by experimental studies. The extentions we present can be easily adopted in a typical Web browser.

Iclone: towards online social navigation

For a place that gathers millions of people the Web seems pretty lonely at times. This is mainly due to the current predominant browsing scenario; that of an individual participating in an autonomous surfing session. We believe that people should be seen as an integral part of the browsing and searching activity towards a concept known as social navigation. Based on this observation we present iClone (www.iclone.com), a social web browser that is able to raise awareness of other people surfing similar websites at the same time by utilizing temporal correlations of their web history logs and to facilitate online communication and collaboration.

Hiding decision tree rules by data set operations

This paper focuses on preserving the privacy of sensitive patterns in the context of inducing decision trees. The subject at hand is approached through a record augmentation approach for hiding sensitive classification rules in binary datasets. Such a hiding methodology is preferred over other heuristic solutions like output perturbation or cryptographic techniques - that restrict the usability of the data in different ways - since the raw data itself is readily available for public use. This methodology is based upon the unique characteristics of the induction of binary decision trees with binary-valued symbolic attributes and binary classes.

Managing the Decision Tree Life-Cycle with Components

Decision trees are one of the most successful Machine Learning paradigms. This paper presents a library of decision tree algorithms in Java. The basic components of a decision tree algorithm are described to support the design of the system architecture. The library can easily embody parts of conventional as well as novel algorithms. The system allows the non-expert user to conduct experiments with decision trees using components and visual tools that facilitate tree construction and manipulation, while the expert user can focus on algorithm design and comparison with few implementation details. The system has been successfully used as a workbench in a programming laboratory for junior computer science students, aiming at providing a solid introduction to object-oriented concepts and practices based on fundamental machine learning paradigm.

Searching the web through user information spaces

During the last years web search engines have moved from the simple but inefficient syntactical analysis (first generation) to the more robust and usable web graph analysis (second generation). Much of the current research is focussed on the so-called third generation search engines that, in principle, inject “human characteristics” on how results are obtained and presented to the end user. Approaches exploited towards this direction include (among others): an alteration of PageRank [1] that takes into account user specific characteristics and bias the page ordering using the user preferences (an approach, though, that does not scale well with the number of users). The approach is further exploited in [3], where several PageRanks are computed for a given number of distinct search topics. A similar idea is used in [6], where the PageRank computation takes into account the content of the pages and the query terms the surfer is looking for. In [4], a decomposition of PageRank to basic components is suggested that may be able to scale the different PageRank computations to a bigger number of topics or even distinct users. Another approach to web search is presented in [2], where a rich extension of the web, called semantic web, and the application of searching over this new setting is described.

CONSOLIDATING A HEURISTIC FOR INCREMENTAL DECISION TREE LEARNING THROUGH ASYMPTOTIC ANALYSIS

This paper addresses stability issues in incremental induction of decision trees. Stability problems arise when an induction algorithm must revise a decision tree very often and oscillations between similar concepts decrease learning speed. We review a heuristic that solves this problem and subsequently employ asymptotic analysis to approximate the basic parameters related to the estimation of computational effort in incremental learning of decision trees. We then use these approximations to simplify the heuristic, we deliver insight into its amortizing behavior and argue how they can also speed-up its execution and enhance its applicability, also providing experimental evidence to support these claims.