Recommending Environmental Big Data Using Semantically Guided Machine Learning - Large Scale and Big Data: Processing and Management - page 475

As the learning rate decreases in competitive learning, large changes are possible
at the beginning of the process. v ( u s , t ) indicates the neighborhood function of the neu-
ron u s . Later with decreasing radius r of the r ( t )-neighborhood of the winner neuron,
only neurons close to the winner are affected. Weights vectors of a SOM can gradually
develop an approximately regular grid after being subjected to input patterns uniformly
distributed over the unit square. SOMs are suitable for solving free learning problems,
but it can also be advantageous to use it to divide the input domain of a fixed learning
problem, for example, counter propagation networks. First, the input domain is parti-
tioned, and then the mean value of the output given by the learning problem for each
individual set of the partition is determined. Finally, the counter propagation provides
for all inputs classified by a neuron of the competition network, and the mean value
over this set as output. This kind of network can only learn piecewise constant function
correctly; linear associates can be used to extend the number of applications to linear
function with the help of the delta rule. The g-SOM clustering on the integrated selected
preprocessed data was quite useful as this technique provided a 2D visual map repre-
sentation of the whole database and natural grouping of the data attributes.
15.3.6 rDF C onversion anD t riPle s tore i imPlementation l ayer
This layer was constructed based on RDF, uniform resource identifier (URI), and
triple store technologies. The aim of this layer was to present the integrated complex
knowledge and associated dynamic recommendations in a more meaningful, trans-
parent, and highly accessible way. W3C introduced the RDF format, which is now
a standard model for machine readable data presentation [7,18,27,55,62]. It decom-
poses data into the pieces (subject, object, and predicate) and gives a URI for each
resource or object. In computing, a URI is a string of characters used to identify a
name or a resource. Such identification enables interaction with representations of
the resource over a network (typically the WWW) using specific protocols. Schemes
specifying a concrete syntax and associated protocols define each URI. Through
the URIs, it is possible to read the information about the particular resource on the
web using the HTTP access. A unified knowledge integration and representation
model was developed using RDF format. Unified knowledge RDFs were created
for all the data sources based on preprocessed data, extracted semantic features,
available metadata, and original provenance information. This made the integrated
environmental feature-based knowledge ready for flexible web integration. The RDF
format provided semantic features sets a unique capability to facilitate data integra-
tion even if the underlying schema differed and it specially supported the evaluation
of schemas over time without requiring the entire data consumption to be changed.
A triple store is a framework used for storing and querying RDF data. It provides a
mechanism for persistent storage and access of RDF graphs. Recently, there has been
a major development initiative in query processing, access protocols and triple store
technologies. The knowledge integration framework was developed using a triple
called “Sesame triple store.” Sesame (Figure 15.4) is an open-source framework for
storage inference and querying of RDF data. Sesame matches the features of Jena
with the availability of a connection API, inference support for multiple back ends
like MySQL and Postgres [7,18,30,32,55,62].