基于潜在语义索引和自组织映射网的检索结果聚类方法

Search Result Clustering Method Based on SOM and LSI

随着互联网的不断发展和数据量的不断增加,搜索引擎的作用日益明显,用户更多地依靠搜索引擎来查找需要的信息.利用潜在语义索引(LSI)理论和自组织映射神经网络(SOM)理论,提出了一种文本聚类的新方法——LSOM.该方法应用SOM网络来实现检索结果文本聚类,不必预先给定类别个数,具有聚类灵活和精度高等特点;同时,该方法应用LSI理论来建立向量空间模型,在词条的权重中引入了语义关系,对于高维的文本特征向量,消减原词条矩阵中包含的噪声,提高聚类速度.LSOM使用一种新的类别标签提取方法,并将提取的标签用于解决SOM基本类划分问题,算法在类别标签和聚类效果评价指标上都比已有的算法有所提高.

Along with the constant development of the Internet and the ever-increasing amount of data, the role of search engines has become increasingly evident. More users rely on search engines to find the information needed. In order to cluster the search results more effectively, thus facilitating the positioning of information among the original unstructured results, the authors propose a text clustering algorithm--the LSOM algorithm, which is based on the self-organizing map （SOM） and the latent semantic index （LSI） theory. It requires no predefined number of clusters and has the advantages of flexibility and preciseness. For high-dimensional texts feature space, LSI is performed to discover a new low-dimensional semantic space, in which the semantic relationship between features is strengthened while the noisy features in the original space are weakened or eliminated. In addition, the clustering process is more efficient due to the effective dimension reduction. In LSOM, a cluster label extraction method is also developed. The extracted labels are further used in resolving the cluster boundary detection problem, which is non-trivial when SOM is applied in text clustering. Experimental results show that the LSOM algorithm performs better than those existing counterparts in evaluation measures of both cluster label and F-measure.