Classification of text documents and extraction of semantically related words using hierarchical Latent Dirichlet Allocation
Imane Chatri · 2015
The amount of available data in our world has been exploding lately. Effectively managing large and growing collections of information is of utmost importance because of criticality and importance of these data to different entities and companies (government, security, education, tourism, health, insurance, finance, etc.). In the field of security, many cyber criminals and victims alike share their experiences via forums, social media and other cyber platforms. These data can in fact provide significant information to people operating in the security field. That is why more and more computer scientists turned to study data classification and topic models. However, processing and analyzing all these data is a difficult task. In this thesis, we have developed an efficient machine learning approach based on hierarchical extension of the Latent Dirichlet Allocation model to classify textual documents and to extract semantically related words. A variational approach is developed to infer and learn the different parameters of the hierarchical model to represent and classify our data. The data we are dealing with in the scope of this thesis is textual data for which many frameworks have been developed and will be looked at in this thesis. Our model is able to classify textual documents into distinct categories and to extract semantically related words in a collection of textual documents. We also show that our proposed model improves the efficiency of the previously proposed models. This work is part of a large cyber-crime forensics system whose goal is to analyze and discover all kind of information and data as well as the correlation between them in order to help security agencies in their investigations and help with the gathering of critical data.