# SWeG: Lossless and Lossy Summarization of Web-Scale Graphs

##### Citations

276 citations

94 citations

### Cites background from "SWeG: Lossless and Lossy Summarizat..."

...sting methods aim to extract smaller subgraphs from the given graphs to preserve pre-defined properties or randomly remove/sample edges during the training process to prevent GNNs from over-smoothing [17, 38, 41, 46]. However, within unsupervised settings, subgraphs sampled from these approaches may be suboptimal for downstream tasks and also lack persuasive rationales to explain the outcomes of the model for the...

[...]

^{1}

37 citations

### Cites background or methods from "SWeG: Lossless and Lossy Summarizat..."

...We focus on SWeG, a recent scheme [124] that constructs supervertices with a generalized Jaccard similarity....

[...]

...In the SWeG lossy summarization [124], ε controls the approximation ratio while I is the number of iterations (originally set to 80 [124])....

[...]

...4) Lossy summarization with Jaccard similarity (SWeG [124]) m ± 2εm‡ O(mI)‡ , ∗ O(m + n) Count of common neighbors Past schemes for lossy graph compression (some might be integrated with Slim Graph in future versions): (§ 4....

[...]

...Finally, for completeness, we also express and implement a recent variant of lossy graph summarization [124]....

[...]

...We analyze its feasibility for practical usage and we express and implement representative schemes as Slim Graph compression kernels, covering spanners [105], spectral sparsifiers [130], graph summarization [124], and others [94]....

[...]

29 citations

### Cites background or methods from "SWeG: Lossless and Lossy Summarizat..."

...5 billion web pages with 128 billion hyperlinks [25], (b) professional networks with more than 20 billion connections [33], and (c) social networks with hundreds of billions of connections [8]....

[...]

...As outputs, [13, 17, 26, 33] yield an unweighted summary graph and edge corrections (i....

[...]

...Inspired by simulated annealing [14] and SWeG [33], we let the threshold decrease over iterations as follows: θ (t) := { (1 + t)−1 if t T 0 if t = T , (21) where t denotes the current iteration number....

[...]

...Inspired by simulated annealing [14] and SWeG [33], we let the threshold decrease over iterations as follows:...

[...]

...A number of algorithms were developed for variants of the graph summarization problem [13, 17, 18, 26, 33, 35]....

[...]

^{1}

26 citations

##### References

4,337 citations

3,266 citations

2,414 citations

2,361 citations

1,518 citations