challenge:
1) how to efficiently generate DAGs
2) how to effectly learn DAG embedding for proximity search
propose : D2AGE (distance-aware DAG Embedding)
source code : http://github.com/shuaiOKshuai
=====================================================================================
Proximity search on hetegeneous graphs aims to measure the proximity between two nodes
previous work : measure the proximity by paths connecting two nodes
weakness : linear sequences have limited expressiveness for the complex connections
propose a model : D2AGE
learning a representation for the DAGs to encode the proximity between two nodes
Semantic proximity search : given a node a, rank other nodes according to some semantic relations
Application : link prediction circle suggestion ...
previous work : measure the semantic proximity by paths connecting two nodes
this paper : looking for a richer structure than paths → DGs(directed graphs)
================================================================================================
DAGs for proximity search:
1) distance awareness in DAGs : RNNs
2) the efficiency in online generating DAGs
- BFS low efficiency
We propose D2AGE :
1) solve distance-aware : assign different weights to different predecessors
2) DAG generation
-1 sample paths as an approximation for the whole graph
-2 feed the samples to a distance-aware DAG generation mechanism to output DAGs and node distance
-3 develop D2AG-LSTM to recursively apply the distance discounts from multiple predecessors
-4 given the training supervision devise an end-to-end solution
====================================================================================
D2AGE
training object : learn
1) the parameter 2) a proximity embedding model to embed