Similarity measure based on hierarchical pair-wise sequence

Quan Sun*, Nengqiang He, Lei Xu, Yipeng Li, Yong Ren

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Collaborative filtering systems have achieved great success in both research and business applications. One of the key technologies in collaborative filtering is similarity measure. Cosine-based and Pearson correlation-based methods are popular ways for similarity measure, but have low accuracy. In this paper, we propose a novel method for similarity measure, referred as hierarchical pair-wise sequence (HPWS). In HPWS, we take into account both the sequence property of user behaviors and the hierarchical property of item categories. We design a collaborative filtering recommendation system to evaluate the performance of HPWS based on the empirical data collected from a real P2P application, i.e. byrBT in CERNET. Experiment results show that HPWS outperforms traditional Cosine similarity and Pearson similarity measures under all scenarios.

Original languageEnglish
Title of host publicationProceedings - 2012 International Conference on Computer Science and Electronics Engineering, ICCSEE 2012
Pages512-516
Number of pages5
DOIs
Publication statusPublished - 2012
Externally publishedYes
Event2012 International Conference on Computer Science and Electronics Engineering, ICCSEE 2012 - Hangzhou, Zhejiang, China
Duration: 23 Mar 201225 Mar 2012

Publication series

NameProceedings - 2012 International Conference on Computer Science and Electronics Engineering, ICCSEE 2012
Volume3

Conference

Conference2012 International Conference on Computer Science and Electronics Engineering, ICCSEE 2012
Country/TerritoryChina
CityHangzhou, Zhejiang
Period23/03/1225/03/12

Keywords

  • Collaborative Filtering
  • Hierarchical Graph
  • Sequence Matching
  • Similarity Measure

Fingerprint

Dive into the research topics of 'Similarity measure based on hierarchical pair-wise sequence'. Together they form a unique fingerprint.

Cite this