Graph contextualized self-attention network for software service sequential recommendation

With the broad application of software services, an increasing number of developers are turning to social coding sites for constructing their applications or conducting further research. These software services generate spatiotemporal data with numerous unique features. GitHub, being the world's largest code hosting platform, is essential to efficiently provide recommendation services for its users. In order to make accurate recommendations and establish effective user-item and item-item rela-tionships, we propose a Graph Contextualized Self-attention Network for Software Service Sequential Recommendation (GCSAN). This model captures global repository-to-repository relationships based on contextual information and recommends suitable repositories to users. Specifically, we leverage the relationships between repositories in all behavior sequences and graph embedding technique to alleviate the data sparsity problem. Moreover, we employ a self attention mechanism to capture user's repository preferences at different time points, assigning varying weights accordingly. Finally, the experimental results on real-world datasets demonstrate the superior performance of our proposed model compared to benchmark recommendation methods.(c) 2023 Elsevier B.V. All rights reserved.