A serendipity-biased Deepwalk for collaborators recommendation.

Area of Science:

• Bibliometrics and Scientometrics
• Recommender Systems
• Network Science

Background:

• Scientific collaboration is prevalent, with existing recommender systems often suggesting familiar collaborators.
• This familiarity can limit researchers' exposure to novel ideas and interdisciplinary connections.
• Serendipity, the act of making fortunate discoveries, is increasingly recognized as valuable in scientific advancement.

Purpose of the Study:

• To design a novel recommender system for identifying serendipitous scientific collaborators.
• To define and operationalize serendipity in collaborator recommendations using relevance, unexpectedness, and value.
• To develop a method that moves beyond conventional collaborator suggestions to foster broader scientific exploration.

Main Methods:

• Introduced a definition of serendipitous collaborators based on relevance, unexpectedness, and value.
• Proposed Seren2vec, a modified DeepWalk algorithm incorporating a serendipity-biased random walk.
• Calculated edge weights in the co-author network based on the three serendipity components.
• Generated collaborator recommendations using cosine similarity of learned scholar vector representations.

Main Results:

• Seren2vec demonstrated superior performance in generating serendipitous collaborator recommendations compared to baseline methods.
• The proposed method effectively balances serendipity with recommendation accuracy.
• Extensive experiments on the DBLP dataset validated the effectiveness of the Seren2vec approach.

Conclusions:

• Seren2vec offers a promising approach to enhance scientific discovery through serendipitous collaborator recommendations.
• The system provides valuable unexpected yet relevant collaborators, broadening researchers' networks.
• This work contributes to the advancement of recommender systems in academic contexts by incorporating the crucial element of serendipity.