Threddy: An Interactive System for Personalized Thread-based Exploration and Organization of Scientific Literature

Title of the Paper

Threddy: An Interactive System for Personalized Thread-based Exploration and Organization of Scientific Literature

Paper Information

• Domain: Personalized exploration and organization of scientific literature, focusing on enhancing user reading experience.
• Keywords: Scientific literature management, personalized exploration, reading support tools, literature review, thread construction, citation context, scientific recommendation systems, document interaction design, information visualization, academic research workflow support.

Research Background and Problem Statement

• Problems and Challenges:

  • With the exponential growth of scientific literature, researchers struggle to track and comprehend numerous research "threads" relevant to their studies.
  • Conducting literature reviews requires reading introductions and related work sections, inferring research frameworks from context, and tracking complex citation relationships, often leading to significant context-switching issues.
  • Researchers' focus on multiple sub-research threads results in an overwhelming amount of information to manage, which traditional tools fail to effectively support in terms of integrating and expanding research threads.
  • Existing tools primarily emphasize literature retrieval and recommendation, with few supporting dynamic construction and expansion of personalized research threads during natural reading processes.

• Significance:

  • Understanding relevant research threads is critical for innovation, positioning research contributions, and building domain knowledge.
  • Simplifying literature management processes and enhancing user reading flow can effectively reduce cognitive load and improve research efficiency.

• Motivation and Related Work:

  • Previous studies, such as Passages, Apolo, and PaperQuest, have contributed tools for literature recommendation and visualization but have not deeply supported users in constructing and expanding research threads during reading.
  • Current literature workflow support tools (e.g., Google Docs, Zotero) do not focus on simplifying immediate context-switching and thread construction processes.

Solution

• Proposed Approach:

  • Develop the system "Threddy," seamlessly integrated into users' PDF reading environments, providing context-based research thread extraction, organization, and discovery functionalities.
  • The system collects compressed information blocks (e.g., citation contexts) from user-read literature and automatically extracts relevant citations, enabling users to save and expand research threads.

• Innovations:

  1. Automated Citation Extraction and Linking: Automatically bind citations to context and link metadata, reducing the workload of switching between documents.
  2. Thread Navigation and Organization Tools: Support thread creation, nesting, renaming, and structural modification via drag-and-drop.
  3. Recommendation System Integration: Recommend additional literature aligned with research themes based on user-organized thread data to dynamically expand threads.
  4. Seamless Integration with User Behavior: Replace browser PDF readers, eliminating the extra burden of switching to new platforms for task execution.

• Implementation Steps and Key Technologies:

  • Introduce PDF full-text decoding technology based on GROBID parsing to automatically locate citation information from user-selected text.
  • Combine high-dimensional vector embeddings (MiniLM) to calculate semantic similarity between threads and cited literature.
  • Construct a hierarchical structure based on user thread behavior, integrating recommendation methods (e.g., citation coverage and semantic similarity calculations) to dynamically discover related literature.

Research Outcomes

• Specific Results:

  • Threddy demonstrated significant utility in functional validation experiments, including enhancing user reading flow, reducing context-switching, and accelerating research thread expansion.
  • The system achieved a 2.6x increase in the number of referenced literature collected and a 2x increase in the number of text fragments gathered, showing significant improvement compared to the baseline condition (Google Docs).

• Advantages Compared to Alternatives:

  • Avoids the time-consuming process of manually cross-checking citations, saving substantial time.
  • Provides persistent thread context display, helping users maintain cognitive continuity when reading different documents.
  • Compared to Google Docs workflows, Threddy significantly reduces users' cognitive and operational burden during information organization.

• Experimental or Evaluation Results:

  • In experiments, users organized an average of 9.9 text fragments and 20.4 referenced papers per research thread (significantly higher than Google Docs' 4.9 fragments and 7.9 papers).
  • Users reported experiencing higher levels of "reading flow" while using Threddy (mean score of 51.0 compared to 42.9 on a 7-point scale, p = 0.02).
  • Threddy's recommendation feature was perceived as capable of "guiding users to discover the latest trends in the field."

• Limitations and Future Directions:

  • Limitations:

    • Current support for organizing user behavior does not include a global view of threads or efficient reorganization tools.
    • Citation recommendation functionality primarily relies on citation coverage, potentially limiting the discovery of interdisciplinary literature.

  • Future Directions:

    1. Optimize for long-term usage, such as large-scale thread management and scalable thread merging and splitting functionalities.
    2. Introduce multi-dimensional recommendation strategies (e.g., cross-disciplinary literature discovery) to break out of current citation chains.
    3. Explore flexible graphical or network models to replace hierarchical thread structures, supporting collaboration.

This study demonstrates that Threddy significantly enhances users' exploration and management efficiency in scientific literature through seamless integration into task workflows and organized data collection. It holds substantial potential for academic and tool dissemination.