Analysis of the recommendation systems based on the tensor factorization techniques, experiments and the proposals

Leginus, Martin ; Zemaitis, Valdas

1. term

Computer Science, Master

2011

2011-01-04

70

Dette speciale undersøger moderne anbefalingssystemer, der bruger tensorfaktorisering, en metode til at nedbryde komplekse, flerdimensionale data for at finde mønstre. Vi sammenligner førende tilgange, beskriver almindelige begrænsninger og faldgruber og bygger en prototypeløsning baseret på HOSVD (en tensorfaktoreringsmetode). Vi gennemfører eksperimenter med vores implementering og rapporterer resultaterne. Eksperimenterne bekræfter de fleste af de problemer, vi identificerede. Vi foreslår desuden mulige løsninger og forbedringer. Til sidst præsenterer vi hovedmålene for næste semester.

This thesis examines modern recommendation systems that use tensor factorization, a way to break down complex, multi-dimensional data to find patterns. We compare leading approaches, outline common limitations and pitfalls, and build a prototype recommender based on HOSVD (a tensor factorization technique). We run experiments with our implementation and report the findings. The experiments confirm most of the issues we identified. We also suggest possible fixes and improvements to address these problems. Finally, we present the main objectives for the next semester.

[This abstract was generated with the help of AI]

recommendation system ; tensor factorization

Download
View record in AAU Student Projects

Leginus, Martin:

• Speeding up tensor based recommenders with clustered tag space and improving quality of recommendations with Non-negative tensor factorization (2011)

Zemaitis, Valdas:

• Speeding up tensor based recommenders with clustered tag space and improving quality of recommendations with Non-negative tensor factorization (2011)