\authors (c) Artem Lutov artem@exascale.info
\license AGPL 3.0 (irrespective on the license statements in the file headers) with possibility to license a derivative work that dynamically links this one under EUPL; optional commercial support and relicensing is provided by the request
\organizations eXascale Infolab, Lumais
\keywords parameter-free graph embedding, unsupervised
learning of network representation, automatic feature extraction,
interpretable compact embeddings, scalable graph embedding
The paper:
@inproceedings{Daor19,
author={Artem Lutov and Dingqi Yang and Philippe Cudr{\'e}-Mauroux},
title={Bridging the Gap between Community and Node Representations: Graph Embedding via Community Detection},
year={2019},
keywords={parameter-free graph embedding, unsupervised
learning of network representation, automatic feature extraction,
interpretable compact embeddings, scalable graph embedding},
}
The source code is being prepared for the publication and cross-platform deployment, and will be fully uploaded soon...
Meanwhile, please write me to get the sources.
The DAOR binaries built on Linux Ubuntu 16.04+ x64 can be found in the Releases.
The execution script to produce embeddings with the recommended number of dimensions is ./daor.sh. The required number of dimensions (128 used in the paper) is fetched during the evaluation process when executing the batch evaluation script of the GraphEmbEval as follows:
./run.sh -m jaccard -a 'daoc-gr=1' -e 128 --force-dims
- GraphEmbEval - Graph (Network) Embeddings Evaluation Framework via classification, which also provides gram martix construction for links prediction.
- DAOC - Deterministic and Agglomerative Overlapping Clustering algorithm for the stable clustering of large networks (totally redesigned former HiReCS High Resolution Hierarchical Clustering with Stable State).
- Clubmark - a parallel isolation framework for benchmarking and profiling clustering (community detection) algorithms considering overlaps (covers), includes a dozen of clustering algorithms for large networks.
- PyExPool - multiprocess execution pool and load balancer, which provides [external] applications scheduling for the in-RAM execution on NUMA architecture with capabilities of the affinity control, CPU cache vs parallelization maximization, memory consumption and execution time constrains specification for the whole execution pool and per each executor process (called worker, executes a job).
- NodeSketch - Highly-Efficient Graph Embeddings via Recursive Sketching
- HARP - Hierarchical Representation Learning for Networks
- NetHash - Efficient Attributed Network Embedding via Recursive Randomized Hashing
- Deepwalk - Online Deep Learning of Social Representations on Graphs
Note: Please, star this project if you use it.