Researches On Decision Tree Algorithm And Its Application To Discovery Of Query Interface

The variety bias problem, which means the attributes with more values are usually preferred to be selected during the process of selection of expanded attributes, is an important problem in the decision tree algorithm. Two methods are proposed to solve the variety bias problem and optimize the traditional decision tree algorithm from two different points of view. Meanwhile, on the basis of theories mentioned above and the Java technique in Eclipse, a demonstration software of decision tree algorithm is developed as an experiment platform to validate the corresponding methods. Then the improved decision tree theories are applied to the discovery of query interface in Deep Web to improve the performance of the discovery and classification.The main research results are concluded as follows:i. A new merging branches algorithm based on equal predictability in decision tree is proposed on the analysis of the variety bias problem in traditional decision tree algorithm. The algorithm uses the pre-pruning strategy, and merges the non-leaf branches which have the equal predictability.ii. In order to solve the variety bias problem in ID3 algorithm which selected the information gain as the a standard of expanding attributes, a kind of AED algorithm based on average Euclidean distance in decision tree is proposed in this paper. The algorithm uses the average Euclidean distance as heuristic information. The experiment results show that the improved AED algorithm can avoid the variety bias of ID3 algorithm.iii. Apply both of the methods to the decision tree demonstration software as an experiment platform and the results have been analyzed to judge the efficiency of these algorithms. All the performances with different parameters are compared to advance the possible improvement in experiment results.iv. On the basis of Deep Web, the corresponding decision tree algorithms are applied to the discovery of the query interface, which solved many problems in Deep Web such as discovery efficiency, classification precision and time cost.