On compositionality in data embedding
Published in International Symposium on Intelligent Data Analysis, Springer, 2023
Recommended citation: Z. Xu, Z. Guo, and N. Cristianini, "On compositionality in data embedding," in International Symposium on Intelligent Data Analysis, Springer, 2023, pp. 484–496. https://doi.org/10.1007/978-3-031-30047-9_38
This paper investigates the principles of compositionality in data embedding approaches, exploring how complex representations can be constructed from simpler components.
Abstract: We analyze compositionality in various data embedding methods, examining how embeddings capture both semantic and structural information across different data types including text, graphs, and knowledge bases.
Key Contributions:
- Theoretical analysis of compositionality in embedding methods
- Empirical evaluation across multiple embedding types
- Framework for understanding compositional properties
- Applications to knowledge graph and text embeddings
Results: The analysis revealed fundamental principles governing how embeddings compose to form more complex representations, with implications for both interpretability and performance.
Recommended citation: Z. Xu, Z. Guo, and N. Cristianini, “On compositionality in data embedding,” in International Symposium on Intelligent Data Analysis, Springer, 2023, pp. 484–496.