Big Data mining and machine learning techniques applied to real world scenarios

Data mining techniques allow the extraction of valuable information from heterogeneous and possibly very large data sources, which can be either structured or unstructured. Unstructured data, such as text files, social media, mobile data, are much more than structured data, and grow at a higher rate. Their high volume and the inherent ambiguity of natural language make unstructured data very hard to process and analyze. Appropriate text representations are therefore required in order to capture word semantics as well as to preserve statistical information, e.g. word counts. In Big Data scenarios, scalability is also a primary requirement. Data mining and machine learning approaches should take advantage of large-scale data, exploiting abundant information and avoiding the curse of dimensionality. The goal of this thesis is to enhance text understanding in the analysis of big data sets, introducing novel techniques that can be employed for the solution of real world problems. The presented Markov methods temporarily achieved the state-of-the-art on well-known Amazon reviews corpora for cross-domain sentiment analysis, before being outperformed by deep approaches in the analysis of large data sets. A noise detection method for the identification of relevant tweets leads to 88.9% accuracy in the Dow Jones Industrial Average daily prediction, which is the best result in literature based on social networks. Dimensionality reduction approaches are used in combination with LinkedIn users' skills to perform job recommendation. A framework based on deep learning and Markov Decision Process is designed with the purpose of modeling job transitions and recommending pathways towards a given career goal. Finally, parallel primitives for vendor-agnostic implementation of Big Data mining algorithms are introduced to foster multi-platform deployment, code reuse and optimization.