On Interpretability of Linear Autoencoders
Open Access
- 8 October 2024
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 975-980
Abstract
We derive a novel graph-based interpretation of linear autoencoder models easer, slim, and their approximate variants. Contrary to popular belief, we reveal that the weights of these models should not be interpreted as dichotomic item similarity but merely as its magnitude. Consequently, we propose a simple modification that considerably improves retrieval ability in sparse domains and yields interpretable inference with negative inputs, as demonstrated by both offline and online experiments. Experiment codes and extended results are available at https://osf.io/bjmuv/.Keywords
Funding Information
- Czech Science Foundation (GA?R) (22-21696S)
This publication has 25 references indexed in Scilit:
- Embarrassingly shallow auto-encoders for dynamic collaborative filteringUser Modeling and User-Adapted Interaction, 2022
- Variational Autoencoders for Collaborative Filtering2018
- Wide & Deep Learning for Recommender Systems2016
- Deep Neural Networks for YouTube Recommendations2016
- Collaborative Denoising Auto-Encoders for Top-N Recommender Systems2016
- The MovieLens DatasetsACM Transactions on Interactive Intelligent Systems, 2015
- HOSLIM: Higher-Order Sparse LInear Method for Top-N Recommender SystemsLecture Notes in Computer Science, 2014
- Learning Attitudes and Attributes from Multi-aspect Reviews2012
- SLIM: Sparse Linear Methods for Top-N Recommender Systems2011
- Amazon.com recommendations: item-to-item collaborative filteringIEEE Internet Computing, 2003