Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DivNet: Diversity-Aware Self-Correcting Sequential Recommendation Networks

Shuai Xiao, Zaifan Jiang

TL;DR

A self-aware self-correcting sequential recommendation net works (DivNet) that is able to estimate utility by capturing the complex interactions among sequential items and diversify recommendations simultaneously is proposed.

Abstract

As the last stage of a typical \textit{recommendation system}, \textit{collective recommendation} aims to give the final touches to the recommended items and their layout so as to optimize overall objectives such as diversity and whole-page relevance. In practice, however, the interaction dynamics among the recommended items, their visual appearances and meta-data such as specifications are often too complex to be captured by experts' heuristics or simple models. To address this issue, we propose a \textit{\underline{div}ersity-aware self-correcting sequential recommendation \underline{net}works} (\textit{DivNet}) that is able to estimate utility by capturing the complex interactions among sequential items and diversify recommendations simultaneously. Experiments on both offline and online settings demonstrate that \textit{DivNet} can achieve better results compared to baselines with or without collective recommendations.

DivNet: Diversity-Aware Self-Correcting Sequential Recommendation Networks

TL;DR

A self-aware self-correcting sequential recommendation net works (DivNet) that is able to estimate utility by capturing the complex interactions among sequential items and diversify recommendations simultaneously is proposed.

Abstract

As the last stage of a typical \textit{recommendation system}, \textit{collective recommendation} aims to give the final touches to the recommended items and their layout so as to optimize overall objectives such as diversity and whole-page relevance. In practice, however, the interaction dynamics among the recommended items, their visual appearances and meta-data such as specifications are often too complex to be captured by experts' heuristics or simple models. To address this issue, we propose a \textit{\underline{div}ersity-aware self-correcting sequential recommendation \underline{net}works} (\textit{DivNet}) that is able to estimate utility by capturing the complex interactions among sequential items and diversify recommendations simultaneously. Experiments on both offline and online settings demonstrate that \textit{DivNet} can achieve better results compared to baselines with or without collective recommendations.

Paper Structure

This paper contains 14 sections, 13 equations, 3 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Recommendation System
  4. Collective Recommendation
  5. Self-correcting Process
  6. Collective Recommendation via DivNet
  7. Model Architecture
  8. Learning Algorithms
  9. Experiments
  10. Datasets
  11. Experimental Setup
  12. Offline Experiments
  13. Online Deployment Results
  14. Conclusion

Figures (3)

  • Figure 1: Typical architecture of industrial-level recommendation systems. It includes feature layer, matching, ranking. The last layer in the yellow box is collective recommendation which puts the final touches to the recommended items and their layout so as to optimize objectives such as click-through-rate or dwell time.
  • Figure 2: The network structures of the proposed self-correcting sequential recommendation network $\textit{DivNet}$. It first projects items into a hidden space with awareness of the existence of contextual items. Then DivNet sequentially selects items by considering influences of previously-selected items and passes their influence to subsequent items. Note that DivNet considers the sequential self-correcting of item interaction explicitly as shown in the decoding network when recommendation items are sequentially viewed by users.
  • Figure 3: Typical self-exciting pattern between items.