Data & Text Mining

Alongside physics, a good part of my career has gone into building machine-learning systems that make sense of unstructured text. The instinct is the same as in physics: find the structure hidden inside noisy data.

Quantum Text®

I was the lead designer and developer of Quantum Text® (Hilbert Technology, 2006–2008) — an advanced text-clustering and categorization engine. The suite is in production use by many US companies and by major US government agencies.

At its heart is a vector model I developed for clustering and categorizing documents: it groups and labels large text collections from only a handful of exemplar documents per category, without hand-written rules.

Reference: “A vector model for text clustering and categorization,” A. A. Maarouf.

Document categorization by machine learning

I’ve carried the thread into academic work — a document-categorization engine built on machine-learning techniques (presented at ICCIS 2019, with collaborators) — and I’ve taught it directly. In a 2018 summer program I led six computer-science students in building a working document categorizer: an analysis engine that classifies input text against categories defined by a few exemplar documents each, plus a graphical interface.

Adjacent data work

• IRAM — a virus-capsid database and analysis resource (Database, 2019): structuring and querying biological sequence and structure data.
• First-principles materials simulation at scale — my physics research runs on national supercomputers (KAUST Shaheen, IBM Blue Gene, Bibliotheca Alexandrina), generating and reducing very large simulation datasets.

Tools

C · C++ · FORTRAN · Mathematica · Hilbert script · Forth