A Link Analysis Approach to Recommendation under Sparse Data

Zan Huang, Hsinchun Chen, Daniel Zeng

Research output: Contribution to conferencePaperpeer-review

14 Scopus citations


Collaborative filtering is one most successful approach to recommendation reported in the literature. It automates the “Word of Mouth” recommendation by suggesting products liked by other consumers who showed similar preference patterns as the target consumer. A serious limitation of the collaborative filtering approach is the sparsity problem, referring to the situation where insufficient historical transactions are available for inferring reliable consumer similarities. In this paper, we represent the consumer transactional data as a graph consisting nodes representing consumers and products and links representing transactions. Under this consumer-product graph, we propose to explore the global graph structure to facilitate collaborative filtering under sparse data. We developed a link analysis recommendation algorithm based on the similar ideas implemented in Web graph analysis algorithms. We tested the proposed algorithm using an online bookstore dataset against standard collaborative filtering algorithms that do not consider transitive associations. The experimental results showed that our proposed algorithm outperformed the benchmark algorithms, especially when insufficient amount of transactional data is available.

Original languageEnglish (US)
Number of pages9
StatePublished - 2004
Event10th Americas Conference on Information Systems, AMCIS 2004 - New York, United States
Duration: Aug 6 2004Aug 8 2004


Conference10th Americas Conference on Information Systems, AMCIS 2004
Country/TerritoryUnited States
CityNew York


  • collaborative filtering
  • link analysis
  • Recommender systems
  • sparsity problem

ASJC Scopus subject areas

  • Library and Information Sciences
  • Information Systems
  • Computer Science Applications
  • Computer Networks and Communications


Dive into the research topics of 'A Link Analysis Approach to Recommendation under Sparse Data'. Together they form a unique fingerprint.

Cite this