2019 WWW Graph Neural Networks for Social Recommendation

[ZahraTaherikhonakdar]

https://github.com/wenqifan03/GraphRec-WWW19**Main Problem:** This paper proposed a Graph model using neural networks. They represent users' social information as a graph: users' friends, users' ratings, and opinions on the items to improve recommendation systems in social networks. They believe that their proposed neural network graph improved the social recommendation as it is effective to obtain information from graph data. (In my research I need to use the graph to get the information from 1-the user and his friends, 2- the user and interested items or topics and user's rating on them ) moreover, they said that friends who are directly connected to the user have an effect on them (in terms of rating items and etc.). They call it heterogeneous strengths of social relations. (which I should consider in my research too).

Input-Output:

phase: Input: graph of--> users' friends + users' rating and opinion on items Output: recommend items based on rating prediction

Previous Works and their Gaps: They investigated previous models which are developed using deep neural networks. Previous models gain success in social recommendation but none of them used Graph Neural Networks (GNNs) to improve social recommendation.

Result: They categorized the previous model as a baseline in three categories namely: traditional recommender systems, traditional social recommender systems, and deep neural network-based recommender systems. Their model (GraphRec) outperforms previous models.

Data Set: They collected datasets from Ciao and Epinion social networking websites

Gap of this work: To improve the GNN technically I should investigate more. But, regarding graph data, they could include the user's search log.

Code: https://github.com/wenqifan03/GraphRec-WWW19

[hosseinfani]

the code is available in the first author's github