Abstract: Classification algorithms need labeled examples to train their model. However there are not enough labeled examples in some domains. There is an approach that training a classifier in one domain and use it to classify examples on different domains. This method is not always successful and this is called domain transfer problem. Spectral Feature Alignment is proposed as a solution to this problem [3]. In this study I investigate this algorithm with a demonstrative example and I do classification experiments on randomly created datasets. Support vector machines are used as a classification tool and the aim of the experiments is to find the impact of the spectral feature alignment on the classification accuracy. Based on the results of these experiments, I will discuss the possible research opportunities on this area.

• Publication Date: 11-Dec-2015
• DOI: 10.15224/978-1-63248-077-4-101
• Views: 0
• Downloads: 0

Abstract: Businesses face the problem of processing extremely large amount of data every day. Finding and analyzing relationships between enormous set of connected data will be the key to successful business. Thus, our work discusses graph databases which are designed for dealing with densely connected data. The paper is focused on comparing Neo4j graph database and traditional Oracle relational database for querying data warehouses. The first results show that Neo4j graph database better deals with more complex questions when amount of data increases, which is very important for multidimensional analysis. Moreover, the query performance does not depend on graph dimensionality but only on size of subgraph covered by query.

• Publication Date: 11-Dec-2015
• DOI: 10.15224/978-1-63248-077-4-120
• Views: 0
• Downloads: 0