Abstract:
Heterogeneous information networks (HIN) can model complex real-world data because their nodes and edges have different semantic types. Thus, HIN embedding, which learns node representations in a low-dimensional space to preserve HIN structural and semantic information, has garnered attention.
Metagraphs, which model common and recurring patterns on HINs, can capture semantic-rich and often latent relationships. Metagraphs have been used for data mining, but not for HIN embedding. This paper uses metagraphs to self-supervisedly learn relationship-preserving HIN embedding for relationship mining tasks.
Most current approaches underuse metagraphs, which are only used in pre-processing and do not actively guide representation learning. Thus, mg2vec, a novel framework, learns metagraph and node embeddings together. Metagraphs actively learn by mapping themselves to the same embedding space as nodes.
Metagraphs also guide learning through first- and second-order constraints on node embeddings to model latent relationships between nodes and individual preferences. Finally, we extensively test three public datasets. mg2vec outperforms state-of-the-art baselines in relationship mining tasks like prediction, search, and visualization.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
