Data fusion algorithm

Data science experts of our company have implemented a method of classification of textual and visual information, followed by making a general decision by means of voting.

Image classification was performed on the base of visual descriptors, resistant to scaling and other alternations, with their subsequent grouping by k-nearest neighbors algorithm. Similar group descriptors have also been established for textual information.

SVM classifier was used to make decisions based on early and late fusion of visual and textual information.

The resulting system was tested on the basis of their tweets and their respective image files.