Big Data - Characterizing an Emerging Research Field Using Topic Models

Abstract

Big Data is one of the latest emerging topics in the field of business information systems, and is marketed as being the key for companies future success. Many analytic solutions are offered by IT companies to help other businesses with the flood of data that is generated within and outside of a company. Despite the extensive use of the notion Big Data for marketing purposes, there is no common understanding of how to characterize the elements of the Big Data concept. The authors contribute to the clarification of this concept with a methodologically enriched literature review by deriving characteristic dimensions from existing definitions of Big Data. These dimensions are validated and enriched with a two-step approach by applying topic models on 248 publications relevant to Big Data. The authors propose that the concept of Big Data can be described by the dimensions of data, IT infrastructure, applied methods, and an applications perspective. The assignment of the results to a generic data analysis process reveals that recent publications focus on data analysis and processing, and less attention is given to the initial data selection or the visualization and utilization of the analysis results.

Understanding big data themes from scientific biomedical literature through topic modeling

“…The meaning of each word was assessed, finding the important concepts for each of the themes and where research focus lies in the corpus. Our work differs from [16] in three ways. First, their analysis was based on only three big data themes, whereas we used multiple definitions which led to twelve themes.…”

“…The largest challenges were encountered during model selection. Two model evaluation methods (i.e., perplexity and harmonic mean) are often used in TM literature [16,19,46,47]. The harmonic mean method calculates an approximation of the marginal likelihood of a fitted model, while perplexity measures how well a fitted model can predict unseen data.…”

“…Some research has been done in objectifying the analysis of TM results [27,30,49,50]. However, so far, the results of TM cannot be quantitatively evaluated [16,48]. For the purpose of this study, a group of seven observers was deemed enough for the topic analysis.…”

“…Most similar to our approach is a study by Hansmann and Niemeyer [16], which applied TM to a big data corpus to discover its characteristics. Their research identified three themes, namely IT infrastructure, methods, and data, and applied TM in two stages.…”

“…Note that all the cases mentioned above are based on qualitative literature studies. Hansmann and Niemeyer [16], however, used text mining to understand the themes included in big data literature. They combined automatic and manual approaches to identify three themes: IT infrastructure, methods, and data.…”