Semantic Data Mining of Financial News Articles

Subgroup discovery aims at constructing symbolic rules that describe statistically interesting subsets of instances with a chosen property of interest. Semantic subgroup discovery extends standard subgroup discovery approaches by exploiting ontological concepts in rule construction. Compared to previously developed semantic data mining systems SDM-SEGS and SDM-Aleph, this paper presents a general purpose semantic subgroup discovery system Hedwig that takes as input the training examples encoded in RDF, and constructs relational rules by effective top-down search of ontologies, also encoded as RDF triples. The effectiveness of the system is demonstrated through an application in a financial domain with the goal to analyze financial news in search for interesting vocabulary patterns that reflect credit default swap (CDS) trend reversal for financially troubled countries. The approach is showcased by analyzing over 8 million news articles collected in the period of eighteen months. The paper exemplifies the results by showing rules reflecting interesting news topics characterizing Portugal CDS trend reversal in terms of conjunctions of terms describing concepts at different levels of the concept hierarchy.