| G. Drakopoulos, L. Theodorakopoulos, S. Sioutas, and Ph. Mylonas |
| Functional Programming Meets Pinecone: Recommending Graph Structured Documents |
| IEEE International Conference on Big Data (IEEE BigData 2025), December 8-11, 2025, Macau, China |
|
ABSTRACT
|
| Vector databases such as Faiss, Pinecone, Chroma, and Milvus facilitate efficient similarity evaluation between embeddings drawn from high dimensional attribute spaces. Recommendation schemes for massive document collections where each such document is internally structured as a large graph are a prime application since the plethora of connectivity patterns associated with said structure lead to long embeddings which can lead to superior performance in terms of established evaluation metrics. Moreover, documents abound with recursive patterns and immutable strings representing key-value pairs or formatted data, paving thus the way for functional implementations. Additionally, vector databases exploit the geometry inherent in data points and queries alike as graph databases rely heavily on topology, allowing flexible query design as well as embedding engineering in order to maintain efficiency against strict efficiency constraints. As a concrete case, the effect of alternative embedding schemes on precision and recall is explored in a collection consisting of documents with synthetic graph structure stored in Pinecone and processed with a Python application following the functional programming paradigm. Furthermore, an extensive overview of the aforementioned four vector databases is given.
|
| 08 December , 2025 |
| G. Drakopoulos, L. Theodorakopoulos, S. Sioutas, and Ph. Mylonas, "Functional Programming Meets Pinecone: Recommending Graph Structured Documents", IEEE International Conference on Big Data (IEEE BigData 2025), December 8-11, 2025, Macau, China |
[
BibTex] [
Print] [
 Back] |
