Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Greedy SLIM: A SLIM-Based Approach For Preference Elicitation

Claudius Proissl, Amel Vatic, Helmut Waldschmidt

TL;DR

The paper tackles the cold-start problem in recommender systems by introducing Greedy SLIM, a SLIM-based approach for preference elicitation. It trains the SLIM matrix greedily, building a static questionnaire of $k$ items that maximize information gain through the SLIM loss $l_{SLIM}$; the resulting questionnaire guides onboarding and enables top-$N$ recommendations. Offline experiments on Movielens-25M and Netflix, complemented by a user study, show that Greedy SLIM can outperform a popular latent-factor–based baseline when the questionnaire is sufficiently long, with especially strong gains on long-tail items. The work demonstrates a practical SLIM-based alternative for onboarding new users, while acknowledging preprocessing costs and outlining directions for dynamic adaptations and broader applicability.

Abstract

Preference elicitation is an active learning approach to tackle the cold-start problem of recommender systems. Roughly speaking, new users are asked to rate some carefully selected items in order to compute appropriate recommendations for them. To the best of our knowledge, we are the first to propose a method for preference elicitation that is based on SLIM , a state-of-the-art technique for top-N recommendation. Our approach mainly consists of a new training technique for SLIM, which we call Greedy SLIM. This technique iteratively selects items for the training in order to minimize the SLIM loss greedily. We conduct offline experiments as well as a user study to assess the performance of this new method. The results are remarkable, especially with respect to the user study. We conclude that Greedy SLIM seems to be more suitable for preference elicitation than widely used methods based on latent factor models.

Greedy SLIM: A SLIM-Based Approach For Preference Elicitation

TL;DR

The paper tackles the cold-start problem in recommender systems by introducing Greedy SLIM, a SLIM-based approach for preference elicitation. It trains the SLIM matrix greedily, building a static questionnaire of items that maximize information gain through the SLIM loss ; the resulting questionnaire guides onboarding and enables top- recommendations. Offline experiments on Movielens-25M and Netflix, complemented by a user study, show that Greedy SLIM can outperform a popular latent-factor–based baseline when the questionnaire is sufficiently long, with especially strong gains on long-tail items. The work demonstrates a practical SLIM-based alternative for onboarding new users, while acknowledging preprocessing costs and outlining directions for dynamic adaptations and broader applicability.

Abstract

Preference elicitation is an active learning approach to tackle the cold-start problem of recommender systems. Roughly speaking, new users are asked to rate some carefully selected items in order to compute appropriate recommendations for them. To the best of our knowledge, we are the first to propose a method for preference elicitation that is based on SLIM , a state-of-the-art technique for top-N recommendation. Our approach mainly consists of a new training technique for SLIM, which we call Greedy SLIM. This technique iteratively selects items for the training in order to minimize the SLIM loss greedily. We conduct offline experiments as well as a user study to assess the performance of this new method. The results are remarkable, especially with respect to the user study. We conclude that Greedy SLIM seems to be more suitable for preference elicitation than widely used methods based on latent factor models.
Paper Structure (22 sections, 1 theorem, 13 equations, 3 figures, 5 tables)

This paper contains 22 sections, 1 theorem, 13 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. Questionnaire
  5. SLIM
  6. Evaluation Method
  7. Evaluation Metrics
  8. Offline Experiments
  9. Reproducibility
  10. Bandit Questionnaire
  11. A First Approach
  12. Greedy SLIM
  13. Offline Experiments
  14. Datasets
  15. Baselines
  16. ...and 7 more sections

Key Result

Lemma 1

Given a SLIM matrix $W$ with empty rows $I_W\subseteq\mathcal{I}$, let $W'$ be another SLIM matrix that is equal to $W$ except for one row ${\textbf{w}'}_i^T$ with $i\in I_W$. We then have

Figures (3)

  • Figure 1: Performance of different SLIM questionnaires and $R_{\text{Gain}}$ for ML-25. Left: all items, right: long-tail items. While for long-tail items the NDCG increases with the number of questions, this is not the case for all items.
  • Figure 2: Comparison of our approach $Q_{\text{GSLIM}}$ with the baselines $Q_{\text{Bandit}}$, $R_{\text{Gain}}$ and $Q_{\text{Greedy}}$. Top: ML-25, bottom: Netflix, left: all items, right: long-tail items.
  • Figure 3: This figure shows the layout of a question. It contained the title, a poster and an abstract of the movie. The layout of a recommendation looked similar.

Theorems & Definitions (2)

  • Lemma 1
  • proof