Automatic Test Case Generation Based on Path Coverage (ATCG-PC) is a black-box problem, due to the one-to-many relationship between test cases and paths, the generated set of test cases always contains redundant test cases that satisfy the path coverage criteria. As the dimensional variables of the test cases change, the progeny individuals will cover different paths. As the dimensional variables of the procedures to be tested become more complex, a single relationship between paths and dimensional variables do not significantly improve the search algorithm. Inspired by this, a path structure matrix is designed that contains three types of structural information: dimensional variables, path nodes, and node towards(Path Structure matrix: DNT matrix). And an automated test case generation based on the path structure matrix and Mainfold-Inspired search is proposed(DNTSA). The method explores the relationship between test cases and path-specific structural information and empower the Mainfold-Inspired search algorithm, in addition, it mines the invisible relationship between the size of the search space and the dimension variables, and uses the information of the use case sample features obtained during the search process as the reference conditions for target path selection. Experiments on four kinds of real-world datasets demonstrate that the DNTSA outperforms previous superior methods.

Session-based recommendation (SBR) predicts the next user interaction by exploiting the anonymous user’s short-term dynamic behaviour. The information available for SBR is limited, and some methods propose to extract the topological information of a session using graph neural networks. The existing problem is that graphs suffer from data sparsity. To this end, we propose Attribute-Enhanced Hypergraph Neural Networks for Session-based Recommendation (A-HGNN). First, the session is modelled as a hypergraph to enhance the integrity of the graph structure from a global view. Then, a hypergraph convolutional network is used for dual information aggregation to obtain item feature representations. On the other hand, item attributes from the local view are fused to accurately capture user preferences. Finally, contrastive learning is used to train the model to supervise and refine the learned session representations from both view. Experiments on two real-world datasets show that the A-HGNN recommendation outperforms previous superior methods.