摘要
针对Deep Web数据源主题分类问题,首先研究了不同位置的特征项对Deep Web接口领域分类的影响,提出一种基于分级权重的特征选择方法RankFW;然后提出一种依赖领域知识的量子自组织特征映射神经网络模型DR-QSOFM及其分类算法,该模型在训练的不同阶段对特征向量和目标向量产生不同程度的依赖,使竞争层中获胜神经元的分布更为集中,簇的区域划分更为明显;最后,在扩展后的TEL-8数据集上进行的实验验证了RankFW和DR-QSOFM的有效性。
In order to solve the problem of Deep Web data sources classification,this paper firstly researched how features in different position could effect the domain of Deep Web interfaces,and proposed a feature selection method RankFW which is based on Ranked weights.Then,a quantum self-organization feature mapping network model was proposed with a classification algorithm.This model relies on the feature vectors and target vectors incoordinately in different phases of training,making a more centralized distribution of winner neurons in competition layer and more ob-vious boundaries among clusters.Finally,some experiments were designed and carried out on the expanded TEL-8 dataset to test the validity of RankFW and DR-QSOFM.
出处
《计算机科学》
CSCD
北大核心
2011年第6期205-210,共6页
Computer Science
基金
军队国防科技项目资助
关键词
DEEP
WEB接口
特征选择
主题分类
分级权重
领域依赖
量子自组织特征映射
Deep Web interface
Feature selection
Topic classification
Ranked weight
Domain relied
Quantum self-organization feature mapping
