UniSKGRep: A unified representation learning framework of social network and knowledge graph

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Mining

Background:

• User modeling for human-oriented applications faces challenges integrating social network (SN) and knowledge graph (KG) data.
• Existing graph representation learning methods often fail to jointly analyze SN and KG, limiting comprehensive feature consideration and performance in downstream tasks.

Purpose of the Study:

• To introduce a Unified Social Knowledge Graph Representation learning framework (UniSKGRep) for improved user modeling.
• To leverage multi-view information from SN and KG by creating a unified Social Knowledge Graph (SKG).

Main Methods:

• UniSKGRep employs an Intra-graph model with separate encoders for social and knowledge views, creating distinct embedding spaces.
• An Inter-graph model bridges these spaces by learning associations between overlapping node pairs.
• An overlapping node enhancement module aligns spaces, considering a limited number of overlapping nodes, through iterative joint training.

Main Results:

• Experiments on two real-world SKG datasets demonstrated UniSKGRep's effectiveness.
• The framework achieved substantial performance improvements compared to strong baseline methods in various downstream tasks.

Conclusions:

• UniSKGRep is the first unified representation learning framework for SN and KG.
• The proposed method successfully integrates diverse features from both graph types, enhancing user modeling capabilities.