Tag Recommendation in Social Bookmarking sites like Deli.cio.us
Varun Ahuja (201206628)Vinay Singri (201305592)Tanuj Sharma ( 201101138 )
IntroductionAutomated process of suggesting
relevant keywords given a dataset
Given link L, description D, and user U, a set of personalized tags CT(L) are suggested with help from given dataset.
First Approach – STaR ( Social Tag Recommender System )Divided in 3 major steps – Pre-processing,
Indexing and Recommendation
Pre-processing – Remove useless tags, Case Folding, Spam Removal
Indexing – Index existing tags against users.
Recommendation – Combine outputs of Title to Tag, Resource Profile, User Profile Recommender.
Problems in First Approach
Not all tags from the dataset appeared.
Low Precision and Low Recall
Without crawling the given link, this approach gives low accuracy
Final Approach – Supervised Learning Model
Modelled as a ranking problem of candidate tags of a given URL
Consists of 3 stages –
◦Candidates Tag Extraction
◦SVM Features Construction
◦Ranking Process
Ranking SVM is used for ranking candidate tags.
Candidates Tag Extraction
Extracted from –
◦Description field of link L
◦Tags assigned by the same user U previously
◦Tags to assigned to the same link L by other users
Given link L, user U, candidate tags
CT{L} = { description(L) union Tags(U) union Tags(L) }
SVM Features Construction
5 features used for each Candidate Tag ( CT ) –
Candidate Tag's Term Frequency (TF) in link's description terms
Candidate Tag's Term Frequency (TF) in link's URL terms
Candidate Tag’s Term Frequency (TF) in T{Rj} (tags assigned to the same URL in the training data).
Candidate Tag’s Term Frequency (TF) in T{Ui} (tags assigned previously by user in the training data.)
Times of candidate tag being assigned as a tag in the training data.
RankingFor any link in test dataset, Candidate
Tags are extracted
Features stored for each candidate tag.
SVM ranking model ranks the candidate tags from top to bottom
Top K tags selected
Tools Used
Future Work
Extension to various datasets
Giving more enriched recommendation for the seed URL
Candidate Tags can be expanded using content similarity based KNN model.
ReferencesSTaR: a Social Tag Recommender System Cataldo
Musto, Fedelucio Narducci, Marco de Gemmis, Pasquale Lops, and Giovanni Semeraro
Department of Computer Science, University of Bari, Italy
• Social Tag Prediction Base on Supervised Ranking Model
Hao Cao, Maoqiang Xie, Lian Xue, Chunhua Liu, Fei Teng and Yalou Huang
College of Software, Nankai University, Tianjin, P.R.China