cite2vec: Citation-Driven Document Exploration via Word Embeddings

Area of Science:

• Information Visualization
• Computer Science
• Digital Libraries

Background:

• Traditional document visualization focuses on content (words).
• Effective exploration of large document collections remains a challenge.
• User search behavior in research offers an alternative perspective for visualization.

Purpose of the Study:

• To develop a new visualization scheme, cite2vec, for exploring document collections based on document usage.
• To enable dynamic exploration and comparison of documents through their citation contexts.
• To leverage user-driven semantic concept exploration for document discovery.

Main Methods:

• Developed cite2vec, a usage-oriented visualization scheme.
• Utilized a joint high-dimensional embedding space for words and documents.
• Enabled user interaction through semantic concept prescription (phrase/document compositions and analogies).
• Captured document usage via citation contexts.

Main Results:

• Demonstrated dynamic exploration and browsing of documents based on usage.
• Showcased the ability to steer document projections using semantic concepts.
• Validated the approach through case studies on a computer vision research paper corpus.

Conclusions:

• Usage-based document visualization offers a powerful alternative to content-based methods.
• cite2vec facilitates intuitive exploration and comparison of documents by their use.
• This approach enhances document discovery and understanding within large collections.