Effective attributed network embedding with information behavior extraction | JoVE Visualize

Area of Science:

Network Science
Machine Learning
Data Mining

Background:

Network embedding methods commonly utilize topological and attribute features for node representation.
Existing methods often overlook implicit information behavior features (inquiry, interaction, sharing), potentially limiting downstream task performance.
Effective node embeddings are crucial for tasks like link prediction, node classification, and community detection.

Purpose of the Study:

To propose a novel network embedding framework, Information Behavior Extraction (IBE), that integrates topological, attribute, and information behavior features.
To enhance the accuracy and effectiveness of network embedding by capturing implicit user behaviors.
To improve performance in downstream network analysis tasks, particularly link prediction.

Main Methods:

Developed the Information Behavior Extraction (IBE) framework, a joint embedding approach.
Utilized existing embedding methods (e.g., SDNE, CANE, CENE) to extract basic node vectors from topological and attribute features.
Introduced a Topic-Sensitive Network Embedding (TNE) model with an Importance Score Rating (ISR) algorithm to capture information behavior features, considering topic-based communities and node interactions.
Concatenated basic vectors with information behavior feature vectors to generate final joint embeddings.

Main Results:

The proposed IBE framework achieved significant and consistent improvements in link prediction performance.
The integration of information behavior features demonstrably enhanced the quality of node embeddings.
Experimental results showed superior performance compared to several state-of-the-art network embedding methods.

Conclusions:

The Information Behavior Extraction (IBE) framework effectively captures implicit information behavior features, leading to superior network representations.
Incorporating information behavior features is crucial for advancing the effectiveness of network embedding techniques.
The proposed method offers a promising direction for improving various network analysis tasks.