Modelling Local and Global Dependencies for Next-Item Recommendations
- Springer International Publishing
- Publication Type:
- Conference Proceeding
- Web Information Systems Engineering – WISE 2020, 2020, 12343, pp. 285-300
- Issue Date:
Copyright Clearance Process
- Recently Added
- In Progress
- Closed Access
This item is closed access and not available.
Session-based recommender systems (SBRSs) aim at predicting the next item by modelling the complex dependencies within and across sessions. Most of the existing SBRSs make recommendations only based on local dependencies (i.e., the dependencies between items within a session), while ignoring global dependencies (i.e., the dependencies across multiple sessions), leading to information loss and thus reducing the recommendation accuracy. Moreover, they are usually not able to recommend cold-start items effectively due to their limited session information. To alleviate these shortcomings of SBRSs, we propose a novel heterogeneous mixed graph learning (HMGL) framework to effectively learn both local and global dependencies for next-item recommendations. The HMGL framework mainly contains a heterogeneous mixed graph (HMG) construction module and an HMG learning module. The HMG construction module map both the session information and the item attribute information into a unified graph to connect items within and across sessions. The HMG learning module learns a unified representation for each item by simultaneously modelling the local and global dependencies over the HMG. The learned representation is then used for next-item recommendations. Results of extensive experiments on real-world datasets show the superiority of HMGL framework over the start-of-the-art methods in terms of recommendation accuracy.
Please use this identifier to cite or link to this item: