Graph Neural Networks for Social Recommendation

最新推荐文章于 2022-12-14 18:03:09 发布

原创最新推荐文章于 2022-12-14 18:03:09 发布 · 6.8k 阅读

0 ·

CC 4.0 BY-SA版权

文章标签：

#图神经网络 #推荐系统

深度学习同时被 3 个专栏收录

5 篇文章

订阅专栏

推荐系统

5 篇文章

订阅专栏

图神经网络

2 篇文章

订阅专栏

本文探讨了图神经网络在社交推荐系统中的应用，面对用户-项目图的复杂性和社交关系的异质性，提出了一个新颖的模型来解决这些挑战。模型通过迭代聚合特征信息，将用户和项目的潜在因素整合到两个不同的图中，即用户-用户社交图和用户-项目图，以实现更精准的推荐。

Graph Neural Networks for Social Recommendation

1. 摘要

构建基于图神经网络的推荐系统的三大挑战
- the user-item graph encodes both interactions and their associated opinions
- social relations have heterogeneous strengths
- users involve in two graphs (e.g., the user-user social graph and the user-item graph)

2. 介绍

难点
- Their main idea is how to iteratively aggregate feature information from local graph neighborhoods using neural networks. Meanwhile, node information can be propagated through a graph after transformation and aggregation.
GNN 的作用
- Hence, GNNs naturally integrate the node information as well as the topological structure and have been demonstrated to be powerful in representation learning [ 5 , 7 , 15 ]. On the other hand, data in social recommendation can be represented as graph data with two graphs.

3. 本文模型

model

3.1 用户模型

3.1.1 Item Aggregation

The purpose of item aggregation is to learn item-space user latent factor $h_{i}^{I}$ by considering items a user $u_{i}$ has interacted with and users’ opinions on these items.

$hiI=σ(W⋅Aggreitems(xia,∀a∈C(i))+b)h^{I}_{i} = σ(W · Aggre_{items} ({x_{ia} ,∀a ∈ C(i)}) + b)$

$hiIh^{I}_{i}$ : item-space user latent factor
$C (i)$ : item-space user latent factor
$x_{ia}$ : a representation vector to denote opinion-aware interaction between $u_{i}$ and an item $v_{a}$

The output of MLP is the opinion-aware representation of the interaction between $u_{i}$ and $v_{a}$ , $x_{ia}$ , as follows:

$x_{ia} = g_{v}([q_{a}⊕e_{r}])$

3.1.2 Social Aggregation

与 Item Aggregation 做法类似

3.2 项目模型

3.2.1 User Aggregation

与 Item Aggregation 做法类似

3.3 预测评分

With the latent factors of users and items (i.e., $h_{i}$ and $z_{j}$ ), we can first concatenate them $h_{i} ⊕ z_{j}]$ and then feed it into MLP for rating prediction as:

$g_{1} = [h_{i} ⊕ z_{j}]$

$g_{2} = σ(W_{2} · g_{1} + b_{2})$

$g_{l-1} = σ(W_{l} · g_{l-1} + b_{l})$

$r^{′}_{ij} = w^{T} · g_{l−1}$

where l is the index of a hidden layer, and $r^{′}_{ij}$ is the predicted rating from $u_{i}$ to $v_{j}$ .

3.4 模型训练

Loss function as follows:

$\frac{1}{2|O|} \sum_{i,j∈O} (r^{′}_{ij} − r_{ij})^{2}$

where $∣ O ∣$ is the number of observed ratings , and $r_{ij}$ is the ground truth rating assigned by the user i on the item j.
Optimizer: RMSprop
Overfitting problem: Dropout