Research And Application Of XML Classification Based On Extreme Learning Machine With Particle Swarm Optimization

As a typical semi-structured data, XML contains both structural and semantic information. The first important issue of XML documents classification is the construction of XML representation model, based on which the structural and semantic information of XML documents can be represented effectively. In this thesis, an improved model named I-SLVM based on SLVM is proposed. With I-SLVM, feature words with strong classification ability are selected by category coverage criteria. The attributes of the feature words are calculated by improved TFIDF, where weights of XML element are introduced into the feature vectors of documents. Therefore, the structural and semantic information of XML documents can be represented effectively and the problem of high dimensions of features can be solved.To avoid the issues that Extreme Learning Machine may encounter due to the random selection of hidden-layer node parameters, a new evolutionary learning algorithm named CPSO-ELM is proposed in this thesis. CPSO-ELM is designed based on improved PSO algorithm named CPSO chaotic optimization search strategy. CPSO is used to select the input weights and bias of hidden node in singlc-hidden-layer feedforward neural networks. The output weights of hidden node are mathematically calculated by using generalized matrix inversion. Besides, the proposed CPSO-ELM algorithm determines the input weights and bias of hidden nodes according to both the classification accuracy on the validation set and the output weights norm of hidden nodes. Experimental results show that compared with traditional ELM and E-ELM, the CPSO-ELM algorithm achieves better classification performance and less complex network structure than traditional ELM.