Initializing matrix factorization methods on implicit feedback databases

Balázs Hidasi, D. Tikk

Research output: Contribution to journalArticle

8 Citations (Scopus)

Abstract

The implicit feedback based recommendation problem-when only the user history is available but there are no ratings-is a much harder task than the explicit feedback based recommendation problem, due to the inherent uncertainty of the interpretation of such user feedbacks. Recently, implicit feedback problem is being received more attention, as application oriented research gets more attractive within the field. This paper focuses on a common matrix factorization method for the implicit problem and investigates if recommendation performance can be improved by appropriate initialization of the feature vectors before training. We present a general initialization framework that preserves the similarity between entities (users/items) when creating the initial feature vectors, where similarity is defined using e.g. context or metadata information. We demonstrate how the proposed initialization framework can be coupled with MF algorithms. We experiment with various similarity functions, different context and metadata based similarity concepts. The evaluation is performed on two implicit variants of the MovieLens 10M dataset and four real life implicit databases. We show that the initialization significantly improves the performance of the MF algorithms by most ranking measures.

Original languageEnglish
Pages (from-to)1834-1853
Number of pages20
JournalJournal of Universal Computer Science
Volume19
Issue number12
Publication statusPublished - 2013

Fingerprint

Factorization Method
Matrix Factorization
Matrix Method
Factorization
Initialization
Feedback
Recommendations
Metadata
Feature Vector
Ranking
Uncertainty
Similarity
Evaluation
Experiments
Demonstrate
Experiment

Keywords

  • Contextual information
  • Implicit feedback
  • Initialization
  • Recommender systems
  • Similarity

ASJC Scopus subject areas

  • Computer Science(all)
  • Theoretical Computer Science

Cite this

Initializing matrix factorization methods on implicit feedback databases. / Hidasi, Balázs; Tikk, D.

In: Journal of Universal Computer Science, Vol. 19, No. 12, 2013, p. 1834-1853.

Research output: Contribution to journalArticle

@article{d304a3c425db4847bf85a3c59b9ebda5,
title = "Initializing matrix factorization methods on implicit feedback databases",
abstract = "The implicit feedback based recommendation problem-when only the user history is available but there are no ratings-is a much harder task than the explicit feedback based recommendation problem, due to the inherent uncertainty of the interpretation of such user feedbacks. Recently, implicit feedback problem is being received more attention, as application oriented research gets more attractive within the field. This paper focuses on a common matrix factorization method for the implicit problem and investigates if recommendation performance can be improved by appropriate initialization of the feature vectors before training. We present a general initialization framework that preserves the similarity between entities (users/items) when creating the initial feature vectors, where similarity is defined using e.g. context or metadata information. We demonstrate how the proposed initialization framework can be coupled with MF algorithms. We experiment with various similarity functions, different context and metadata based similarity concepts. The evaluation is performed on two implicit variants of the MovieLens 10M dataset and four real life implicit databases. We show that the initialization significantly improves the performance of the MF algorithms by most ranking measures.",
keywords = "Contextual information, Implicit feedback, Initialization, Recommender systems, Similarity",
author = "Bal{\'a}zs Hidasi and D. Tikk",
year = "2013",
language = "English",
volume = "19",
pages = "1834--1853",
journal = "Journal of Universal Computer Science",
issn = "0948-6968",
publisher = "Springer Verlag",
number = "12",

}

TY - JOUR

T1 - Initializing matrix factorization methods on implicit feedback databases

AU - Hidasi, Balázs

AU - Tikk, D.

PY - 2013

Y1 - 2013

N2 - The implicit feedback based recommendation problem-when only the user history is available but there are no ratings-is a much harder task than the explicit feedback based recommendation problem, due to the inherent uncertainty of the interpretation of such user feedbacks. Recently, implicit feedback problem is being received more attention, as application oriented research gets more attractive within the field. This paper focuses on a common matrix factorization method for the implicit problem and investigates if recommendation performance can be improved by appropriate initialization of the feature vectors before training. We present a general initialization framework that preserves the similarity between entities (users/items) when creating the initial feature vectors, where similarity is defined using e.g. context or metadata information. We demonstrate how the proposed initialization framework can be coupled with MF algorithms. We experiment with various similarity functions, different context and metadata based similarity concepts. The evaluation is performed on two implicit variants of the MovieLens 10M dataset and four real life implicit databases. We show that the initialization significantly improves the performance of the MF algorithms by most ranking measures.

AB - The implicit feedback based recommendation problem-when only the user history is available but there are no ratings-is a much harder task than the explicit feedback based recommendation problem, due to the inherent uncertainty of the interpretation of such user feedbacks. Recently, implicit feedback problem is being received more attention, as application oriented research gets more attractive within the field. This paper focuses on a common matrix factorization method for the implicit problem and investigates if recommendation performance can be improved by appropriate initialization of the feature vectors before training. We present a general initialization framework that preserves the similarity between entities (users/items) when creating the initial feature vectors, where similarity is defined using e.g. context or metadata information. We demonstrate how the proposed initialization framework can be coupled with MF algorithms. We experiment with various similarity functions, different context and metadata based similarity concepts. The evaluation is performed on two implicit variants of the MovieLens 10M dataset and four real life implicit databases. We show that the initialization significantly improves the performance of the MF algorithms by most ranking measures.

KW - Contextual information

KW - Implicit feedback

KW - Initialization

KW - Recommender systems

KW - Similarity

UR - http://www.scopus.com/inward/record.url?scp=84885704132&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84885704132&partnerID=8YFLogxK

M3 - Article

AN - SCOPUS:84885704132

VL - 19

SP - 1834

EP - 1853

JO - Journal of Universal Computer Science

JF - Journal of Universal Computer Science

SN - 0948-6968

IS - 12

ER -