Building user profiles based on sequences for content and collaborative filtering

作者:

Highlights:

• A generic method to transform users to sequences using collaborative and content-based information.

• We define a user similarity metric based on LCS algorithm (extensible to any string-based comparison algorithm) than can produce competitive recommendations.

• Definition of various parameters (confidence, preference, normalizations, and threshold) to produce better recommendations in the LCS-based algorithm.

• Normalization functions help to improve accuracy without hurting diversity or novelty.

• Preference filtering does not decrement the performance but reduces its computational cost.

摘要

•A generic method to transform users to sequences using collaborative and content-based information.•We define a user similarity metric based on LCS algorithm (extensible to any string-based comparison algorithm) than can produce competitive recommendations.•Definition of various parameters (confidence, preference, normalizations, and threshold) to produce better recommendations in the LCS-based algorithm.•Normalization functions help to improve accuracy without hurting diversity or novelty.•Preference filtering does not decrement the performance but reduces its computational cost.

论文关键词:Hybrid recommender systems,Preference filtering,Content-based filtering,Collaborative filtering,Longest Common Subsequence

论文评审过程:Received 6 April 2018, Revised 26 September 2018, Accepted 8 October 2018, Available online 18 October 2018, Version of Record 18 October 2018.

论文官网地址:https://doi.org/10.1016/j.ipm.2018.10.003