Research And Implementation Of The Approach To Discover Hypernym-Hyponym Relations Between Social Tags

With the popularity of Web 2.0,Folksonomy has been applied to a growing number of web systems.Folksonomy is an uncanonical and open classification system that allows users to use any tags to annotate resources,and users can share the tags with others.However,Folksonomy does not have hierarchy structure.The lack of hierarchical structure leads to the lack of relevance between social tags,which will lead to low accuracy in resource navigation and information retrieval.The research is generally used to map Folksonomy to ontologies,and can get the hierarchical structure of the Folksonomy with the help of the ontologies.However,due to the dynamic updating characteristics of Folksonomy,there will appear many new words which do not exist in ontologies.This makes it impossible to build the hierarchical structure of a given Folksonomy based on existing ontologies.To solve the above problems,this thesis no longer relies on ontologies.The hierarchical structure of a Folksonomy is constructed from its tags,and the basis of the tag hierarchy is the hypernym-hyponym relation between tags,so this thesis is devoted to mining the hypernym-hyponym relations between social tags.The main work are as follows:(1)Design a tag embedding model for training vectors of social tags.The proposed model does not need to rely on external knowledge sources,and only uses social tags to obtain vectors of social tags.(2)Propose a hypernym-hyponym relation mining model based on Embedding,By combining the tag vector with the SVM or random forest model to identify the hypernym-hyponym relations between the tags.(3)The experimental results on two datasets verify that the proposed method in this paper significantly outperforms several state-of-the-art methods.(4)Design and implement a visualization system for mining the hypernym-hyponym relation between social based on the model Dynamic Tag Similarity Neural Network(DTSNN)...