Conclusion Introduction Recommendation System { "asin": - PowerPoint PPT Presentation

1 Introduction 2 Method 3 Experiment 4 Conclusion

Introduction

Recommendation System { "asin": "0000031852", "title": "Girls Ballet Tutu Hot Pink", "price": 3.17, "related": { "also_bought":["B00JHONN1S", ... ], "also_viewed":["B002BZX8Z6", ... ], "bought_together": ["B002BZX8Z6"] }, "salesRank": {"Toys & Games": 211836}, "brand": "Coxlures", "categories": [["Sports & Outdoors"]] } Predict behavior

Motivation A B usefulness

Motivation C useful

Method

Framework User reviews Item reviews Great Product. I love this … … Product and my children will too. I cant wait til Christmas to Give them their present !

Hierarchical Attention based Neural Network

Intra-review Interaction vector 𝜗ℝ 𝐿 , Attention Mechanism ∗ = 𝑋 𝑈 𝑆𝑓𝑀𝑉 𝑋 𝑏 𝑘 ℎ ℎ 𝑘 + 𝑋 𝑣 𝑤 𝑣,𝑗 + 𝑐 1 + 𝑐 2 𝑏 Great Product. I love this Product and my children will too. I cant wait til Christmas to Give them their present ! Pre-train word embedding

Inter-review External Memory Attention for inter-review

Prediction Layer Fully-Connected Objective function

Experiment

Dataset Amazon Product Data • May 1996 – July 2014 • Each users and items has at least 5 reviews

Baseline Methods CNN Based

Comparison with baseline methods Evaluation Method

Explanation Analysis of HANN

Explanation Analysis of HANN

Conclusion

Conclusion • We design a hierarchical attention framework to learn the interaction between users and items from reviews to construct an explainable recommendation system. • The well-designed hierarchical attention mechanism helps the model capture user profiles and item profiles, making them more explainable and reasonable, and ultimately leads to improvements in rating prediction.