A Semantic BI Process for Detecting and Analyzing Mentions of Interest for a Domain in Tweets

• Vilmar César Pereira Júnior UFSC
• Renato Fileto UFSC
• Willian Santos de Souza UFSC
• Matthias Wittwer Univ. of Leipzig
• Olaf Reinhold Univ. of Leipzig
• Rainer Alt Univ. of Leipzig

Resumo

Social media posts and other time-stamped text can carry lots of useful information. However, their proper analysis requires capturing the semantics of what is mentioned in their contents, filtering what is of interest for particular application domains, and structuring the extracted information for analytical purposes. This work proposes an approach to analyze the incidences of mentions of interest for some domain in these texts, by combining Semantic Web and Business Intelligence (BI) technologies. This approach is supported by an automatic ETL process that semantically annotates textual clips with Linked Open Data (LOD), filters LOD resources of interest in the annotations by using bridges between LOD classes and a high-level domain ontology, and adapts existing LOD hierarchies accordingly to serve as analysis dimensions. Experimental results show that our proposal: (i) is able to find a considerable number of mentions to things of interest for business in tweets recently sent from Brazil; (ii) allows the identification of the most mentioned (classes of) things of interest; and (iii) enables new useful queries for information analysis on data cubes with some dimensions derived from existing LOD hierarchies.

Palavras-chave: Semantic Web, linked data, semantic annotations, social media, data warehousing