Kwan Yi

Linking folksonomy to Library of Congress subject headings: an exploratory study

Purpose – The purpose of this paper is to investigate the linking of a folksonomy (user vocabulary) and LCSH (controlled vocabulary) on the basis of word matching, for the potential use of LCSH in bringing order to folksonomies.Design/methodology/approach – A selected sample of a folksonomy from a popular collaborative tagging system, Delicious, was word‐matched with LCSH. LCSH was transformed into a tree structure called an LCSH tree for the matching. A close examination was conducted on the characteristics of folksonomies, the overlap of folksonomies with LCSH, and the distribution of folksonomies over the LCSH tree.Findings – The experimental results showed that the total proportion of tags being matched with LC subject headings constituted approximately two‐thirds of all tags involved, with an additional 10 percent of the remaining tags having potential matches. A number of barriers for the linking as well as two areas in need of improving the matching are identified and described. Three important tag...

A hidden Markov model-based text classification of medical documents

The purpose of the study is to test the application of the hidden Markov model (HMM) using prior knowledge in medical text classification (TC). HMM has been applied to a wide range of applications in information processing, but not so much in TC applications. The Medical Subject Heading (MeSH) is utilized for prior knowledge in the model. A prototype for an HMM-based TC model is designed, and an experimental model based on the prototype is implemented so as to categorize medical documents into MeSH. A subset of OHSUMED is used for the experiments. Our results show that the performance of our model is comparable to those reported in the literature.

Harnessing collective intelligence in social tagging using Delicious

A new collaborative approach in information organization and sharing has recently arisen, known as collaborative tagging or social indexing. A key element of collaborative tagging is the concept of collective intelligence (CI), which is a shared intelligence among all participants. This research investigates the phenomenon of social tagging in the context of CI with the aim to serve as a stepping-stone towards the mining of truly valuable social tags for web resources. This study focuses on assessing and evaluating the degree of CI embedded in social tagging over time in terms of two-parameter values, number of participants, and top frequency ranking window. Five different metrics were adopted and utilized for assessing the similarity between ranking lists: overlapList, overlapRank, Footrule, Fagins measure, and the Inverse Rank measure. The result of this study demonstrates that a substantial degree of CI is most likely to be achieved when somewhere between the first 200 and 400 people have participated in tagging, and that a target degree of CI can be projected by controlling the two factors along with the selection of a similarity metric. The study also tests some experimental conditions for detecting social tags with high CI degree. The results of this study can be applicable to the study of filtering social tags based on CI; filtered social tags may be utilized for the metadata creation of tagged resources and possibly for the retrieval of tagged resources.

An empirical study on the automatic resolution of semantic ambiguity in social tags

Due to the popularity of collaborative tagging services and systems, the role of social tags in information organization and retrieval has become increasingly critical within the tagging system. Consequently, the resolution of semantic ambiguity of tag sense (i.e., meaning) plays an important role in the enhancement of information organization and retrieval in collaborative tagging applications. Our approach to tackle the task of automatic resolution of semantic ambiguity is based on the hypothesis that given a target tag, some of the co-occurring social tags that were selectively assigned by the same or by different people to the same resource can serve as a useful dataset. Four different methods (i.e., two LSA-based, LIN-based, and Co-occurrence method) for tag sense disambiguation are proposed and their results are compared, aiming to automatically rank the senses associated with a given target tag. The experimental results with a Delicious dataset indicate that LSA_w and LIN-based methods produce more stable performances, and suggest the same two methods for the Tag Sense Disambiguation task, particularly in the situation that the Boolean effect is more likely to arise.

A content analysis of Twitter hyperlinks and their application in web resource indexing

Twitter has emerged as a popular source of sharing and delivering news information. In tweet messages, URLs to web resources and hashtags are often included. This study investigates the potential of the hyperlinks and hashtags as topical clues and indicators to tweet messages. For this study, we crawled and analyzed about 1.5 million tweets for a 3‐month period covering any topic or subject. The findings of this study revealed a power law relationship for the ranking and frequency of (a) the host names of URLs, and (b) a pair of hashtags and URLs that appeared in the tweet messages. This study also discovered that the most popular URLs used in tweets come from news and media websites, and a majority of the hyperlinked resources are news web pages. One implication of this study is that Twitter users are becoming more active in sharing already published information than producing new information. Finally, our investigation on hashtags for web resource indexing reveals that hashtags have the potential to be used as indexing terms for co‐occurring URLs in the same tweet. We also discuss the implications of this study for web resource recommendation.