Ade Idowu - Hands on Intro to Developing Explainability for Recommendation Systems

[andynice]

This is the video: https://www.youtube.com/watch?v=FS8pBksh_aI

00:00 Introduction
00:23 Who am I?
00:53 Agenda
03:28 Taxonomy of RecSys
04:33 Content-based Vs Collaborative filtering
06:09 Types of Recommendations
08:05 Intro to Matrix Factorization (MF) for RecSys
10:20 Techniques for solving Latent Factor models
11:26 Computing MF with stochastic gradient descent
13:05 XAI for RecSys
14:19 Why should a RecSys be explainable?
15:16 ML Accuracy Vs Explainability Trade-off
16:22 Types of RecSys Explanation Styles
18:24 Examples of explanation styles
19:46 Example sentence style explanation
20:03 Example of Visual style explanation
20:13 Example social explanation
20:35 Exploration of the Movielens workshop data
22:46 Types of Explainability approaches
25:31 Model-based RecSys explainers
26:40 Model-based (Ante-hoc) RecSys Explanation
27:39 ALS Explainer
30:27 ALS Explainer Example
30:47 EMF Explainer
33:20 EMF Explainer Example
35:11 Post-hoc RecSys Explanation
36:56 Post Hoc RecSys Explainers
39:23 AR Explainer
40:54 AR Explainer workflow
42:26 AR Explainer Example
42:57 Post-hoc k Explainer
43:30 kNN Explainer Example
43:38 FM - LIME Explainer
44:58 FM - LIME Explainer approach
47:07 FM - LIME Explainer Example
47:13 Performance Metrics
49:27 Future Work
53:17 Demos
53:44 Key python packages
01:12:44 Q & A