Use this URL to cite or link to this record in EThOS:
Title: Content-awareness and graph-based ranking for tag recommendation in folksonomies
Author: Landia, Nikolas
Awarding Body: University of Warwick
Current Institution: University of Warwick
Date of Award: 2013
Availability of Full Text:
Access from EThOS:
Access from Institution:
Tag recommendation algorithms aid the social tagging process in many userdriven document indexing applications, such as social bookmarking and publication sharing websites. This thesis gives an overview of existing tag recommendation methods and proposes novel approaches that address the new document problem and the task of ranking tags. The focus is on graph-based methods such as Folk- Rank that apply weight spreading algorithms to a graph representation of the folksonomy. In order to suggest tags for previously untagged documents, extensions are presented that introduce content into the recommendation process as an additional information source. To address the problem of ranking tags, an in-depth analysis of graph models as well as ranking algorithms is conducted. Implicit assumptions made by the widely-used graph model of the folksonomy are highlighted and an improved model is proposed that captures the characteristics of the social tagging data more accurately. Additionally, issues in the tag rank computation of FolkRank are analysed and an adapted weight spreading approach for social tagging data is presented. Moreover, the applicability of conventional weight spreading methods to data from the social tagging domain is examined in detail. Finally, indications of implicit negative feedback in the data structure of folksonomies are analysed and novel approaches of identifying negative relationships are presented. By exploiting the three-dimensional characteristics of social tagging data the proposed metrics are based on stronger evidence and provide reliable measures of negative feedback. Including content into the tag recommendation process leads to a significant increase in recommendation accuracy on real-world datasets. The proposed adaptations to graph models and ranking algorithms result in more accurate and computationally less expensive recommenders. Moreover, new insights into the fundamental characteristics of social tagging data are revealed and a novel data interpretation that takes negative feedback into account is proposed.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: QA Mathematics ; ZA Information resources