The main goal of reading paper is not just understanding it. Try to understand the key concept, but we need to get new ideas and research directions from the paper.
Paper information
- title: Personalizing Graph Neural Networks with Attention Mechanism for Session-based Recommendation
- authors: Wu, S., Zhang, M., Jiang, X., Ke, X., & Wang, L.
- venue: arXiv preprint arXiv:1910.08887, 2019.
- pdf link: link
- github: link
- abstract:
The problem of personalized session-based recommendation aims to predict users’ next click based on their sequential behaviors. Existing session-based recommendation methods only consider all sessions of user as a single sequence, ignoring the relationship of among sessions. Other than that, most of them neglect complex transitions of items and the collaborative relationship between users and items. To this end, we propose a novel method, named Personalizing Graph Neural Networks with Attention Mechanism, A-PGNN for brevity. A-PGNN mainly consists of two components: One is Personalizing Graph Neural Network (PGNN), which is used to capture complex transitions in user session sequence. Compared with the traditional Graph Neural Network (GNN) model, it also considers the role of users in the sequence. The other is Dot-Product Attention mechanism, which draws on the attention mechanism in machine translation to explicitly model the effect of historical sessions on the current session. These two parts make it possible to learn the multi-level transition relationships between items and sessions in user-specific fashion. Extensive experiments conducted on two real-world data sets show that A-PGNN significantly outperforms the state-of-the-art personalizing session-based recommendation methods consistently.
Summary: problems to address, key ideas, quick results
presentation link
Questions about the paper?
I have questions on some formular notations.
How to get staic user embedding eu and item embedding evi?
What do you like?
They convert the session sequence into graph, which has stronger informative power than sequence structure.
It is critical to model the impact of historical sessions on current session.
What you don't like?
In the user behavior graph, the author put the historical session and current session into one graph without distinction, which lose some chronological information.
How to improve?
Perhaps we can get the initial user embedding and item embedding through better representation learning method.
Any new ideas?
In this paper the edge from node i to j represents the user interact j after i in one session. However, if a user interact j after i in both sessions, they think the edges are identical. If we add a timestamp to every edge in user behavior graph, then every edge would be different. In this way, we can transform this problem into the link prediction problem in dynamic graph network. Maybe we can learn from the methods in this field.
Reproducing results (if any)
The main goal of reading paper is not just understanding it. Try to understand the key concept, but we need to get new ideas and research directions from the paper.
Paper information
The problem of personalized session-based recommendation aims to predict users’ next click based on their sequential behaviors. Existing session-based recommendation methods only consider all sessions of user as a single sequence, ignoring the relationship of among sessions. Other than that, most of them neglect complex transitions of items and the collaborative relationship between users and items. To this end, we propose a novel method, named Personalizing Graph Neural Networks with Attention Mechanism, A-PGNN for brevity. A-PGNN mainly consists of two components: One is Personalizing Graph Neural Network (PGNN), which is used to capture complex transitions in user session sequence. Compared with the traditional Graph Neural Network (GNN) model, it also considers the role of users in the sequence. The other is Dot-Product Attention mechanism, which draws on the attention mechanism in machine translation to explicitly model the effect of historical sessions on the current session. These two parts make it possible to learn the multi-level transition relationships between items and sessions in user-specific fashion. Extensive experiments conducted on two real-world data sets show that A-PGNN significantly outperforms the state-of-the-art personalizing session-based recommendation methods consistently.
Summary: problems to address, key ideas, quick results
presentation link
Questions about the paper?
I have questions on some formular notations.
How to get staic user embedding eu and item embedding evi?
What do you like?
They convert the session sequence into graph, which has stronger informative power than sequence structure.
It is critical to model the impact of historical sessions on current session.
What you don't like?
In the user behavior graph, the author put the historical session and current session into one graph without distinction, which lose some chronological information.
How to improve?
Perhaps we can get the initial user embedding and item embedding through better representation learning method.
Any new ideas?
In this paper the edge from node i to j represents the user interact j after i in one session. However, if a user interact j after i in both sessions, they think the edges are identical. If we add a timestamp to every edge in user behavior graph, then every edge would be different. In this way, we can transform this problem into the link prediction problem in dynamic graph network. Maybe we can learn from the methods in this field.
Reproducing results (if any)