Clustering Algorithms for Tags

Clustering Algorithms for Tags

Yu Zong (West Anhui University, China & University of Science and Technology of China, China) and Guandong Xu (University of Technology Sydney, Australia)
Copyright: © 2013 |Pages: 15
DOI: 10.4018/978-1-4666-2806-9.ch003

Abstract

With the development and application of social media, more and more user-generated contents are created. Tag data, a kind of typical user generated content, has attracted lots of interests of researchers. In general, tags are the freely chosen textual descriptions by users to label digital data sources in social tagging systems. Poor retrieval performance remains a major problem of most social tagging systems resulting from the severe difficulty of ambiguity, redundancy, and less semantic nature of tags. Clustering method is a useful tool to increase the ability of information retrieval in the aforementioned systems. In this chapter, the authors (1) review the background of state-of-the-art tagging clustering and the tag data description, (2) present five kinds of tag similarity measurements proposed by researchers, and (3) finally propose a new clustering algorithm for tags based on local information that is derived from Kernel function. This chapter aims to benefit both academic and industry communities who are interested in the techniques and applications of tagging clustering.
Chapter Preview
Top

Introduction

With the proliferation of social media, lots of User Generated Content (UGC) have been brought and UGC becomes one of the main prevailing Web trends (Baeza-Yates, 2009). Various types of data, e.g., text, photo, music, and video, are generated and viewed.

As a typical type of UGC, social tag (also known as collaborative tag or social annotation) has obtained significant development and it also has become popular for their revolutionary ways of organizing online resources. Tags are simple, uncontrolled and ad-hoc labels that are assigned by users to describe or annotate any kind of resource. Since the distribution or types of contents are diverse and change dynamically, tagging is especially suitable for online copra.

Many social tagging sites have been established, such as Del.icio.us (http://itwwt.com/tag), etc., have been established in China for users to annotate their topics.

The low technical barrier of tag based recommender system and easy usage of tagging have attracted a large amount of research interest. The user-contributed tags are not only an effective way to facilitate personal organization but also provide a possibility for users to search for information or discover new things. However, the ambiguity, redundancy and less semantic nature are the major problems suffering all tagging systems. For example, for one same resource, different users will use their own textual description to annotate, resulting in the tagging behavior much ambiguous and redundant. In order to deal with these difficulties, recently clustering method has been introduced into tag-based recommender system to find meaningful information conveyed by tags. As the user tagging behaviors can be modeled as data record with triple attributes, i.e. user, resource, and tag, clustering on tag data could be conducted on these three attributes respectively. The effectiveness of clustering of tag data is the ability of aggregating tags into topic domains. In this chapter, we (1) briefly discuss research background and related work on tagging clustering, (2) introduce the form of tag data and various tag similarity measurements, (3) propose a clustering algorithm named Local Information Passing Clustering algorithm (LIPC). Especially, in LIPC, We first estimate a KNN neighbor directed graph G of tags, the Kernel density of each tag in its neighborhood is calculated at the same time; we then use Local coverage and Local Kernel to capture the local information of each tag; thirdly, we define two operators, that is, I and O, to pass the local information on G; then a tag priority is generated when I and O are converged; at last, we use the tag priority values to find out the clusters of tags by using Depth First Search (DFS) on G. Experimental results demonstrate the efficiency and the improved outcome of tag clusters by using the proposed method.

Complete Chapter List

Search this Book:
Reset