Semantically Document Clustering Using Contextual Similarities

摘要

Efficient Document clustering can be performed based on the term level, sentence level and concept level techniques in the high dimensional document space. Most of the existing techniques have problems such as two-variable problem, high computational time and low similarity relatedness which reduces the clustering efficiency. To overcome the existing drawbacks, a hybrid clustering algorithm called Semantically Document Clustering algorithm is proposed in this paper. The Semantically Document Clustering algorithm is developed by combining the features of Directed Ridge Regression (DRR), Fuzzy relational Hierarchical clustering (FHC) and Conceptual clustering methods presented in our previous researches. The proposed Semantically Document Clustering algorithm utilizes the semantic weight of terms related to the concepts from Wikipedia and Word Net to categorize the texts in the documents. Then the similarity between the sentences is calculated by using the Jiang and Conrath measure which considers the concept weight and the similarity measure for effective clustering. The direct ridge regression is applied to build a Laplacian matrix and the diagonal elements of the normalized Laplacian matrix are varied to solve the two-variable problem. Then the fuzzy hierarchical rules are employed to classify the rows of the normalized Laplacian matrix into classes for calculating the membership for the observations and the center vectors. Thus the term relatedness, sentence relatedness and concept relatedness can be calculated and the documents can be clustered efficiently. Experiment results also show that the proposed hybrid approach Semantically Document Clustering method provides more accurate document clustering than the state-of-the-art clustering methods.