Automatic Construction Of Topic Hierarchy Of Folksonomy

Folksonomy provides a new perspective for understanding ontology. This paper proposes the automatic construction of topic hierarchy of folksonomy to realize mutual advantages for common development of both folksonomy and ontology. We aim to overcome the defects of folksonomy and alleviate the bottleneck of knowledge acquisition from ontology. It can also help to solve common problems on construction of both topic tag hierarchy and ontology concept hierarchy, which is an effective way to promote the development of folksonomy and ontology. Automatic construction of topic tag hierarchy is a core issue of common concern in the field of natural language processing, knowledge management, semantic web, digital library, etc. It can also promote the development of related research, such as information retrieval, machine translation, question answering, intelligent navigation, and recommendation system. This paper consists of four parts.First of all, we propose a method of topic key tag extraction based on edge weight, which forms the foundation for the following automatic construction of topic tag hierarchy. In view of semantic association characteristics of folksonomy, we design the edge weight which combines the local co-occurrence in a specific topic with the global semantic similarity over all the topic dimensions in the whole collection considered. The new edge weight can decompose a traditional random walk into multiple topic-sensitive random walks, and each of these walks run on a tag graph specific to the corresponding topic. And then, each of these walks outputs a list of tags ordered on the basis of importance score. Then, the top-ranking tags are extracted as the topic key tags for each topic. Experiments show that the proposed method outperforms state-of-the-art methods, which can not only effectively identify topic key tags, but can also well connect the most related tags under a specific topic.Then, we propose a method of topic key tag extraction combined with preference to further boost the performance of topic key tag extraction. We propose a new preference which biased the tag most relative to the given topic. Then, we identify the topic key tag by the joint effect of edge weight and preference. Meanwhile, we explore multiple folksonomy data sources to overcome bias of any single data source. Experiments on both single and mixed dataset show that the proposed method can advance the performance of topic key tag extraction by improving the importance of abstract tags. The combination of method improvement and resource exploration can achieve better performance.And then, we propose a method of topic tag hierarchy construction based on heterogeneous evidence from multiple sources. This method can provide a hierarchy organization of topic tags and relations. We leverage the characteristics of data sources of domain-dependent folksonomy and domain-independent ontology for heterogeneous evidence extraction. Furthermore, we present a novel scheme to use the heterogeneous evidence extracted from multiple sources separately in both initial topic tag hierarchy construction and topic tag hierarchy improvement by distinguishing them into undirected and directed evidence. Results of comprehensive experiments indicate that the proposed method performs well in automatic construction of topic tag hierarchy and improves more than 20% on recall compared with state-of-the-art methods.At last, we propose a method of tag recommendation based on topic hierarchy, which extends the evaluation and application of topic tag hierarchy. The topic tag hierarchy can be applied in tag recommendation with proposed novel tag recommendation strategy. For one thing, we can improve the performance of tag recommendation through topic tag hierarchy, and then overcome the defects of folksonomy. For another, the quality of topic tag hier archy can be evaluated indirectly through the performance of tag recommendation. Comparison with state-of-the-art methods and data sources on tag recommendation indicates that the proposed method can promote the development of tag recommendation and topic tag hierarchy evaluation effectively.In conclusion, focusing on automatic topic hierarchy construction from folksonomy, this paper conducts a thorough study on topic key tag extraction, topic relation identification, topic tag hierarchy construction, topi c tag hierarchy evaluation, and tag recommendation. We report the results of this study mean to support for more related research and application.