LCS Graph Kernel Based on Wasserstein Distance in Longest Common Subsequence Metric Space
This work aims to improve graph representation learning for graph learning tasks by developing a more robust and efficient graph kernel, addressing issues of over-smoothing in message-passing and information loss in fixed-length path comparisons.
The authors propose a new Graph Kernel based on Longest Common Subsequence (LCS) similarity to address limitations of message-passing and fixed-length path comparison methods. They introduce a novel metric space using LCS similarity and compute a Wasserstein-based graph distance within this space, which prioritizes comparisons between similar paths. An adjacent point merging operation is also proposed to reduce computational cost.
For graph learning tasks, many existing methods utilize a message-passing mechanism where vertex features are updated iteratively by aggregation of neighbor information. This strategy provides an efficient means for graph features extraction, but obtained features after many iterations might contain too much information from other vertices, and tend to be similar to each other. This makes their representations less expressive. Learning graphs using paths, on the other hand, can be less adversely affected by this problem because it does not involve all vertex neighbors. However, most of them can only compare paths with the same length, which might engender information loss. To resolve this difficulty, we propose a new Graph Kernel based on a Longest Common Subsequence (LCS) similarity. Moreover, we found that the widely-used R-convolution framework is unsuitable for path-based Graph Kernel because a huge number of comparisons between dissimilar paths might deteriorate graph distances calculation. Therefore, we propose a novel metric space by exploiting the proposed LCS-based similarity, and compute a new Wasserstein-based graph distance in this metric space, which emphasizes more the comparison between similar paths. Furthermore, to reduce the computational cost, we propose an adjacent point merging operation to sparsify point clouds in the metric space.