Dimensionality reduction in data summarization approach to learning relational data
ACIIDS'13 Proceedings of the 5th Asian conference on Intelligent Information and Database Systems - Volume Part I
| Hi-index | 0.00 |
The structure in the multiple tables is so complex that we should not only improve the efficiency, but also insure the accuracy of classification when we classify the data. Some existing classification algorithms have good results in terms of the efficiency and the accuracy, for example: an efficient multi-relational Bayesian classifier based on the semantic relationship graph. But how to get the information from each table can affect the final accuracy while traversing the semantic relationship graph and the existing algorithms always query all the tables. It not only cost much time, but also has less improvement on the accuracy of the classification. This paper firstly defines the contribution of the single relation based on singular value decomposition, and measures the effect of the tables on the classification according to their contribution. Then we can reduce some tables which have a little effect on the classification and query the tables according to their contribution, and we can find some tables which can make the greatest accuracy of the classification. When we do like this, we can improve the efficiency of the classification and ensure the accuracy at the same time. The experiment proves the method is right and efficient.