Recommending Environmental Big Data Using Semantically Guided Machine Learning - Large Scale and Big Data: Processing and Management - page 486

FIGURE 15.13 Integrated metadata set.

15.4.5.3 SOM-Based Big Knowledge Recommendation
SOM may be considered a nonlinear generalization of the PCA. After multiple train-
ing SOMs learn to classify input vectors according to how they are grouped in the
input space. They differ from competitive layers in that neighboring neurons in the
SOM learning to recognize neighboring sections of the input space. Thus, SOMs
learn both the distribution (as do competitive layers), and topology of the input vectors
they are trained on. There are two ways to interpret a SOM. As in the training, phase
weights of the whole neighborhood are moved in the same direction, and similar
items tend to excite adjacent neurons. Therefore, SOM forms a semantic map where
similar samples are mapped close together and dissimilar ones mapped apart. The
SOM network was applied on the selected least correlated semantic attributes from
PCA to estimate and visualize the natural grouping of the data. The most significant
PCs covering 99(%) of the data variance were used to design, initialize, and guide
the self-organizing map network. Using 3D scatter plot-based on the first three most
significant PC found that two clear clusters could be formed for the currently avail-
able integrated data matrix. This information was used to design the initial SOM.
Two clear clusters were chosen based on inter-cluster distance and individual loca-
tions of the data points. The dynamically established number of the most significant
PCs (in this case p = 5) and number of clear clusters (in this case c = 2) were used to
define a SOM network of [p × c] dimension. This SOM was trained using the whole