Overlapping Spaces for Compact Graph Representations

Authors: Kirill Shevkunov, Liudmila Prokhorenkova

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments confirm that overlapping spaces outperform the competitors in graph embedding tasks with different evaluation metrics. We also perform an empirical analysis in a realistic information retrieval setup, where we compare all spaces by incorporating them into DSSM.
Researcher Affiliation Collaboration Kirill Shevkunov Yandex, MIPT Moscow, Russia shevkunov.ks@phystech.edu Liudmila Prokhorenkova Yandex Research, MIPT, HSE University Moscow, Russia ostroumova-la@yandex.ru
Pseudocode No The paper includes Figure 1 which illustrates the overlapping space computation, but it is a diagram, not structured pseudocode or an algorithm block.
Open Source Code Yes The code of our experiments is available.6 https://github.com/shevkunov/overlapping-spaces-for-compact-graph-representations
Open Datasets Yes We use the following graph datasets: the USCA312 dataset of distances between North American cities [4] (weighted complete graph), a graph of computer science Ph.D. advisoradvisee relationships [1], a power grid distribution network with backbone structure [28], a dense social network from Facebook [17], and Eu Core dataset generated using email data from a large European research institution [16]. We also collected a new dataset by launching the breadth-first search on the Wikipedia category graph, starting from the Linear Algebra category with search depth limited to 6. Further, we refer to this dataset as WLA6; more details are given in Appendix A.2.
Dataset Splits Yes All queries are divided into train, validation, and test sets, and for each signature, the optimal iteration was selected on the validation set.
Hardware Specification No The paper states: 'Technical requirements for our implementation are provided with code, small experiments can be reproduced on a regular computer.' This statement is too general and does not provide specific hardware details (e.g., CPU/GPU models, memory, specific cluster names) used for the experiments.
Software Dependencies No The paper states: 'Technical requirements for our implementation are provided with code'. However, the paper text itself does not list specific software dependencies with their version numbers.
Experiment Setup Yes The training details are given in Appendix A.