CiteSee: Augmenting Citations in Scientific Papers with Persistent and Personalized Historical Context

Title of the Paper

CiteSee: Augmenting Citations in Scientific Papers with Persistent and Personalized Historical Context

Paper Information

• Domain: Human-Computer Interaction (HCI), scientific literature reading, and personalized recommendation
• Keywords: reading interface, scientific literature, personalized recommendation, literature review, citation enhancement, academic tools

Research Background and Problem

• Problem or Challenge:

  • Scientific research relies on integrating and expanding existing knowledge, with embedded citations in literature serving as a key method for discovering relevant studies.
  • During the reading process, researchers often face a large number of citations, making it difficult to determine which ones deserve attention, potentially overlooking critical historical studies.
  • Current mainstream literature tools lack support for personalized user interests and reading history, failing to effectively help users identify and prioritize relevant citations.

• Importance:

  • Ignoring important literature can lead to redundant research or missing critical background information, significantly impacting the quality and efficiency of academic work.
  • During literature reviews, scholars need to efficiently filter and understand the relevance of cited works to save research time.

• Research Motivation:

  • Current academic tools (e.g., recommendation systems and citation analysis tools) either adopt non-personalized approaches or are disconnected from the literature reading experience, unable to assist users in perceiving citation context and importance in real-time reading scenarios.
  • Users desire tools that can track cross-references among multiple citations and identify key literature frequently cited across papers in a field.

Solution

• Approach:

  • Introduced a Chrome extension named CiteSee to enhance the scientific literature reading experience.
  • Provides visualized and personalized citation enhancement based on users' reading history, literature library, and publication records.

• Innovations:

  • Automatically highlights citations related to the user's reading history and provides personalized historical context for these citations within the text.
  • Employs intuitive visual enhancement methods (e.g., color differentiation, icon markers) to help users instantly perceive the familiarity and importance of cited works.
  • Incorporates dynamic interaction features, allowing users to click on citations to view detailed contextual information (e.g., cited article titles, abstracts, and relevant paragraphs).

• Implementation Steps and Key Technologies:

  1. Citation Enhancement: Dynamically enhances citation display based on whether the citation is "reencountered" (mentioned in the reading history) or "known" (saved, accessed, or cited previously).
  2. Contextual Information: Provides "literature cards" for each citation, detailing its context across multiple papers recently read by the user.
  3. Personalized Prioritization: Assigns weights to citations based on user behaviors such as reading, saving, and citing, dynamically adjusting display priority (e.g., using varying color intensity).
  4. Technical Dependencies:
     • Utilizes Semantic Scholar API to retrieve paper metadata.
     • Employs Grobid for automatic citation and reference parsing.
     • Uses a backend database (PostgreSQL) to store user reading history and behavior logs.

Research Outcomes

• Specific Results:

  • Developed the CiteSee tool and validated its effectiveness through two rounds of user studies.
  • In experiments, CiteSee's citation prioritization significantly outperformed traditional citation ranking methods (e.g., global citation counts or semantic embedding-based rankings).
  • In practical application tests, users using CiteSee achieved nearly 2.7 times higher efficiency in literature discovery compared to traditional methods.

• Comparison with Existing Solutions:

  • Compared to existing literature reading tools, CiteSee offers deeper personalization, dynamically providing familiar citations within the reading context.
  • Compared to recommendation systems, CiteSee delivers immediate and uninterrupted support for literature discovery within reading scenarios.

• Experiment and Evaluation Results:

  • Experiment 1 (Controlled Study):
    • A study involving 10 users demonstrated that the Reencountered Citations Strategy based on user reading history was more effective (p<0.001) in literature discovery tasks compared to three other methods.
  • Experiment 2 (Field Study):
    • Six participants used CiteSee in real-world literature review scenarios, reporting that CiteSee improved the systematicity and efficiency of their reviews.
    • On average, users discovered 57% of relevant literature through CiteSee's embedded citation discovery, significantly higher than the 21% reported in previous studies.

• Limitations and Future Directions:

  • Suffers from a cold-start problem, requiring users to accumulate a reading history to fully leverage CiteSee's advantages.
  • Current analysis is limited to academic papers in PDF format.
  • Future improvements could include supporting multi-task literature reading, such as automatically identifying reading history segments relevant to user tasks.
  • Beyond citation enhancement, further exploration could focus on supporting persistent context recording and interaction for scientific concepts.