Cross-Domain Data Fusion

In big data, there are datasets with di erent representations, distributions, and scales in di erent domains. How can we unlock the power of knowledge from multiple disparate, but potentially connected, datasets? Addressing this challenge is of paramount importance in big data research. Integrating heterogeneous datasets is essentially what distinguishes big data from traditional data manipulation and analytic tasks. In “Methodologies for CrossDomain Data Fusion: An Overview,” Yu Zheng summarizes the crossdomain data-fusion methodologies for big data analytic tasks (IEEE Trans. Big Data, vol. 1, no. 1, 2015, pp. 16–34). His survey paper categorizes these tasks as stage-based, feature level–based, and semantic meaning–based data-fusion methods. The third category is further divided into four groups: multiview learning–based, similarity-based, probabilistic dependency–based, and transfer learning–based methods. Di erent from traditional data fusion studied in the database community, these cross-domain data-fusion methods focus on knowledge fusion rather than schema mapping and data merging (see Figure 1).