A PSO-Neural Network-Based Feature Matching Approach in Data Integration

Abstract

This chapter presents a feature matching approach based on a particle swarm optimization neural network (PSONN) in data integration to identify the corresponding features in different datasets. Unlike previous probability-based feature matching using a weighted average of multiple measures calculating matching probability, the proposed approach utilizes PSONN, obtaining similarity rules of feature matching to find matched features in different datasets. The feature matching strategy utilizing bidirectional matching, two-stage matching, and feature combination is also provided for solving all types of feature matching, including 1:0, 0:1, 1:1, 1:n, m:n, and m:1. The proposed approach is implemented for matching features from different datasets and is compared with a probability-based feature matching method. The experiments show that the weights of the same measures may vary for different data contexts. In addition, the results demonstrate the availability and advantages of the proposed approach in feature matching.