【论文】 modeling dual period-varing preferences

Introduction

外卖推荐和传统推荐的区别：

1. 外卖推荐是用户、店家、食物三元的推荐。比如：
2. 时间偏好不同
   
   所以，提出了DPVP模型，
   · 双重交互模型（dual interaction-aware module）, aiming to capture users’ preferences
   based on the ternary interaction data.
   · propose a time-based decomposition module to capture users’ dual-period-specific preferences in the period hierarchy and simultaneously learn users’ period-shared preferences.
   · a time-aware gate network is utilized to fuse the user’s dual period-specific and period-shared preferences to obtain the user’s dual period-varying preferences.
   · Eventually, the model make a prediction based on the balance between the users’ dual period-varying preferences for stores and foods.

建模

· 构建三元交互数据的图网络 G
U： user
S： store
O: food
M: time ( |M| = 4）
node: U、S、O
edge：US、UO、OS的两两交互

1. dual interaction aware
   把full interaction graph 拆成 food-level和 store-level
   

2. time-based decomposition
   再分别拆成四个时间段
   

3. embedding
   · capture the user’s preferences for two completely different types of entities（food and store）
   · 不同时间段的embedding是共享的

4. mean-pooling： the aggregated neighbor embedding for the central food/store node in the food-level/store-level graph
   

5. user- aware gate: 用户的个人偏好（同一个store, 有的用户偏向包子，有的偏向面条）
   

6. time-aware gate: 不同时间段的偏好
   

预测

评估

offline

online