ComLittee: Literature Discovery with Personal Elected Author Committees

Document Title

ComLittee: Literature Discovery with Personal Elected Author Committees

Document Information

• Subject Area: Human-Computer Interaction and Intelligent Literature Discovery System Design
• Keywords: Literature Discovery System, Author Recommendation, Explainable Recommendation, Interactive Machine Learning, Author-Centric Discovery

Research Background and Problem

• Identified Problems or Challenges:

  • Traditional literature discovery systems primarily rely on document content for interaction, with limited utilization of research network signals from authors.
  • Current systems fail to effectively capture users' evolving understanding of documents and authors, lacking direct pathways for author exploration.
  • Users may miss critical academic research leads because existing systems fail to recommend documents that cover relevant author groups.

• Importance:

  • Authors often publish multiple related papers around a specific research agenda. Exploring author relationship networks can significantly enhance the depth and breadth of academic discovery.
  • Integrating users' accumulated knowledge and interests into the system can improve the efficiency and quality of literature and author recommendations, fostering innovation and cross-disciplinary discoveries.

• Research Motivation and Related Work:

  • Author-enhanced systems in literature discovery represent an emerging design field, where author relationship networks can improve user efficiency in academic discovery processes.
  • Researchers have proposed the concept of "author-enhanced literature discovery," emphasizing the integration of document recommendation techniques with author graph signals.
  • This study draws inspiration from prior work on recommendation algorithms and explainable system design, such as FeedLens and Citation Networks.

Solution

• Proposed Method or Solution:

  • ComLittee is an author-enhanced literature discovery system that facilitates the discovery of authors and documents through user-selected "author committees."
  • The system employs author-centric interaction, allowing users to manually select relevant authors and build committees to guide subsequent system recommendations.
  • ComLittee integrates document recommendation scores, user behavior history, and co-citation and co-author networks to provide richer and more personalized recommendation signals.

• Innovative Features:

  • Unlike traditional literature discovery systems that focus on document-centric interaction, ComLittee encourages direct interaction with authors.
  • The system expands author recommendations using triadic closure mechanisms in citation networks, helping users connect familiar authors with unfamiliar ones.
  • It provides interactive explainable filters, enabling users to quickly identify key authors and documents within specific research topics.

• Implementation Steps and Key Techniques:

  1. Users select a topic and save seed documents.
  2. The system analyzes user feedback and seed documents to recommend an initial list of authors.
  3. Users save relevant authors to their personal committee, updating system signals to improve recommendations.
  4. Author network signals (e.g., citation relationships and co-author networks) are used to expand the recommendation scope.
  5. The system presents recommendation rationales through interactive explainable filters, enhancing user understanding and exploration efficiency.

Research Outcomes

• Specific Outcomes:

  • Users discovered more relevant authors and documents through ComLittee, with significantly improved efficiency compared to baseline systems.
  • ComLittee enabled users to identify more novel and engaging authors and documents within a short time frame.
  • The system effectively mitigated the "cold start" problem by constructing exploration paths based on authors familiar to the user.

• Advantages Over Existing Solutions:

  • In author discovery, the system helped users identify more unknown yet relevant authors, with superior user experience.
  • In literature discovery, users saved more documents on average and rated their interest in the documents significantly higher.

• Experimental or Evaluation Results:

  • Experiments showed a significant increase in the number of documents saved through the author recommendation feature, with higher post-experiment ratings for the relevance and novelty of discovered documents and authors.
  • Embedded experiments comparing author- and document-centric discovery approaches demonstrated the advantages of author-centric interaction.
  • User feedback indicated high compatibility of the new system with their academic discovery processes, with a significantly higher willingness to adopt it compared to baseline systems.

• Limitations and Future Directions:

  • Limitations:
    • Difficulty in breaking out of specific research groups or "schools of thought," potentially leading to overly concentrated recommendations.
    • Current recommendation mechanisms rely heavily on citation networks, limiting discovery potential in under-cited research areas.
  • Future Directions:
    • Develop cross-disciplinary recommendation algorithms to enhance the system's ability to identify unconnected research across different fields.
    • Support dynamic adjustment of recommendations over time, such as allowing users to revisit or redefine the impact of early feedback.
    • Explore grouping methods based on academic history and author backgrounds to help users better understand relationships between research communities.