REPOSITORY > RESULTS

Abstract

The thesis addresses the development of an innovative data mining platform ClowdFlows
and novel knowledge discovery scenarios implemented therein as executable data mining
workflows. The ClowdFlows platform is implemented as a cloud-based web application
with a graphical user interface which supports the construction, execution and sharing of
data mining workflows. Big data analytics is supported by several algorithms, including
novel ensemble techniques implemented using the MapReduce programming model, and
a special stream mining module for real-time analysis using continuous parallel workflow
execution.
The adaptability of ClowdFlows is demonstrated through descriptions of two platforms
that expand the usage of the web-based workflow environment to fields other than data
mining. The ConCreTeFlows platform is a novel platform for computational creativity,
while TextFlows is focused mainly on text mining and natural language processing. The
ConCreTeFlows platform is demonstrated with a conceptual blending use case, while features
of TextFlows are demonstrated on three use cases: comparison of document classifiers
and different part-of-speech taggers on a text categorization problem, and outlier detection
in document corpora.
We present novel use cases in Inductive Logic Programming (ILP) and Relational Data
Mining (RDM). The main novelty is a propositionalization technique called wordification
which can be seen as a transformation of a relational database into a corpus of text documents.
The wordification methodology and the evaluation procedure are implemented as
executable workflows in ClowdFlows. The implemented workflows include several other
ILP and RDM algorithms, as well as the utility components that enable access to these
techniques to a wider research audience.
The real-time analysis and data stream mining capabilites are demonstrated on a novel
active learning scenario for dynamic adaptive sentiment analysis, which is able to handle
changes in data streams and adapt its behavior over time. Established stream mining
techniques are transferred to the visual programming paradigm and demonstrated through
the sentiment analysis use case, using active learning of microblogging data streams with
a linear Support Vector Machine.
The thesis contributes to open-source scientific software. The ClowdFlows platform, its
adaptations ConCreTeFlows and TextFlows, and all the described use cases are publicly
available. This enables experiment reproducibility, as well as workflow adaptations and
enhancements.