The need for effective software development and maintenance has long been identified
in the field of Software Engineering. Today, however, that new collaborative development
paradigms and open-source software initiatives have been introduced, the amount of software
engineering data that are available online is vast, thus the challenge is to effectively analyze
them in order to produce better software. And this challenge is actually a problem of reuse.
In this thesis, we provide a unified approach that includes applying data mining techniques
on software engineering data in order to enable reuse in different phases of the software
engineering life cycle. Our methodology offers solutions for the phases of requirements
elicitation and specification extraction, software design and development, and quality assurance
and testing.

Initially, we focus on requirements elicitation and specification extraction, where we
build ontologies for storing software requirements. Using a set of our designed tools, the
requirements engineer can instantiate our model from functional requirements written in
natural language, graphical storyboards and/or UML diagrams. Our modeling approach allows
validating the requirements, extracting specifications, as well as applying data mining
techniques for reusing requirements. Furthermore, we propose different methodologies for
recommending requirements, based on association rule extraction and matching techniques.

Concerning the phase of software development, we facilitate reuse at different levels. At
first, we design a code search engine that is oriented towards reuse. Our engine is based on an
index that allows searching for source code components, for code snippets, and for multiple
objects (at project level, e.g. for finding architectural patterns). After that, we propose a
source code recommendation system that encompasses test-driven reuse. Our system uses
information retrieval techniques and employs a syntax-aware mining model to retrieve components
that are relevant to the query of the developer. The results are also further examined
using tests to ensure that they cover the required functionality. The next challenge that we
face is that of connecting source code components, which is usually performed with snippets.
In this context, we propose a system that accepts queries in natural language and downloads
snippets, which are subsequently clustered according to the libraries used. The libraries and
the snippets are ranked according to their preference by the developer community. Finally,
in an effort to further improve on retrieved code (or generally any code written by the
developer), we propose a code validation methodology given information from questionanswering
services. We analyze the problem of finding similar question posts in these services,
and specifically recommend the use of source code for finding relevant snippets.

The third axis of this thesis involves evaluating the quality of the retrieved components.
To this end, we initially present a source code recommendation system that evaluates the
retrieved components both from a functional and from a reusability perspective. Our system
involves a syntax-aware matching mechanism and a model that assesses the reusability of
each component. After that, we design a more fine-grained reusability estimation model for
source code components, based on their preference/reuse rate by the developer community.
Our model, which is built using machine learning techniques, is capable of assessing the
quality of source code components as perceived by developers.

All in all, in this thesis we apply data mining techniques on software engineering data, in
order to provide a unified solution for practicing reuse in all phases of software development
and maintenance. This solution aims to create better software with minimal cost and effort,
thus affecting important economic and social aspects of everyday life.