Thursday, 11 May 2023

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

 Source: https://blog.allenai.org/citesee-e0f9e9d46569

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

Left: Titled “contextualize” at the top, followed by a paragraph figure with 4 inline citations. Citation 1 is red, 2 is green, 3 is overlaid with a red quotation mark, and 4 is overlaid with a red heart emoji. Right: Titled “Discover” at the top, followed by a paragraph figure with 3 inline citations. Citation 12 is highlighted in light yellow, and citation 10 is highlighted in a more saturated yellow. Citation 13 is highlighted in a very saturated yellow-orange color.
CiteSee augments inline citations to known papers to help contextualize the current paper. This includes saved(1, red) and visited papers (2, green), papers previously cited by current user (3”), and their own publications (♥). CiteSee also highlights citations to unknown papers (10–12) to help discover important prior work based a user’s engagements on their citing papers.

Inline citations play a crucial role in the scholarly research process, as they allow researchers to contextualize the paper they are reading within the cited work, draw connections among relevant papers, and build up a higher level view of the research fields. A prior work estimated that inline citations account for around 1 in 5 paper discoveries during active research (King et al. 2009). Despite their importance and ubiquity, our preliminary interviews showed that it can be challenging for scholars to prioritize which inline citations to explore, considering the sheer volume of citations they encounter during literature reviews and the varying relevance to a reader’s interests. The most common concern was the fear of overlooking important citations, which could lead to significant research consequences. Additionally, participants expressed difficulties in tracking their progress and retaining context around saved or visited papers.

From left to right, common citation type with 3 examples with different yellow highlight shadings. Visited papers rendered in green, saved papers rendered in red, cited papers with a red quotation mark overlaying the top right corner, and “own papers” with a red heart emoji overlay the reference number. Under the common citations examples, there is an “unexplored papers” label, and under the other 4, there is an “Explored /familiar papers” label.
Overview of different visual augmentation types, with one category for citations to unexplored papers, and four categories for explored / familiar papers.

These findings informed the development of CiteSee, a personalized paper reading tool for in-situ citation sensemaking. CiteSee visually augments citations within scientific papers based on their connections to a user’s research activities to better reflect their research interests and literature review progress. By leveraging a user’s publications, paper library and recent reading history, CiteSee offers a range of visual citation augmentation types.These augmentation types enable users to both prioritize unexplored inline citations most related to their interests during literature reviews (i.e., re-encountered across papers), as well as keeping track of which inline citations were already explored (e.g., visited or saved). In addition, CiteSee also presents persistent and personalized historical context around citations, allowing users to make sense of how a citation connects to them personally. For this, users can further interact with inline citations by clicking on them, revealing personalized contexts in a Paper Card, such as the last time the paper was opened or the citing sentences from across papers they have recently explored. This allows for a more personalized and context-rich understanding of the citations within a paper. By leveraging these core mechanisms, CiteSee effectively supports users in discovering relevant citations, surfacing familiar papers, and providing personalized context around inline citations to aid in conducting literature reviews.

Two screenshots. Left: A popup card of citation [33] highlighted and yellow. The content of the card is as follows: The main area has the title, authors, and abstract of the cited paper. The bottom half of the card contained a list of other paper titles and citing sentences from the user’s reading history. Right: A similar screenshot of a paper card for citation [56] in red showing that it is previously saved. The bottom of the card contained “Saved from:” a paper title and a citing sentence.
[Left] To help users discover important prior work, unexplored citations are highlighted in different shades of yellow to indicate their potential relevance to the user. [Right] To help users keep track of which citations were already explored and to draw connections between familiar papers to the current paper, inline citations to familiar papers (e.g., saved) are rendered in red. [Both] To see personalized context around inline citations, users can click on a citation to see its Paper Card with personalized context such as citing sentences from recently read papers or the citing sentence where the cited paper was saved.

In a lab study, we validate CiteSee’s core functionality of highlighting relevant citations for paper discovery during literature reviews. Participants read a set of papers and actively examined citations to find important prior work, with results showing that CiteSee’s personalized approach significantly outperformed three baselines. We also conducted a field deployment study to further understand CiteSee’s real-world benefits. We recruited participants who had planned to conduct literature reviews and installed on their computers for one to two weeks for a planned literature review. We found that participants were actively engaged with the system, and that the majority of papers they discovered and saved were via highlighted inline citations. In the post-interviews, participants expressed benefits in using CiteSee, with its visual augmentations helping them discover more relevant prior work, remember papers they have examined or encountered in the past, and make sense of common citations across multiple papers, ultimately proving to be a valuable tool in supporting literature review tasks and enhancing understanding of inline citations.

In conclusion, CiteSee is a promising scientific paper reading tool that enhances the literature review process by personalizing the user’s experience and providing contextualized inline citations. By tracking and exploiting the user’s past research activities, CiteSee allows researchers to prioritize highly relevant inline citations during literature reviews and explore additional personalized context to make better sense of them. Our studies demonstrate the advantages of CiteSee over baseline strategies for paper discovery, and the positive impact it has on real-world literature review tasks. As scientific research publications continue to grow rapidly, intelligent reading tools like CiteSee will play a crucial role in helping researchers navigate the vast landscape of existing literature, ensuring they can identify and understand the most relevant and impactful work in their fields and how they relate and build on each other.

CiteSee received the Best Paper Award (1%) in ACM #CHI2023, and we will present this work at the conference in Hamburg, Germany this month. CiteSee is a collaborative effort among researchers from

@ , and including Joseph Chee Chang, Amy X. Zhang, Jonathan Bragg, Andrew Head, Kyle Lo, Doug Downey, and Daniel S. Weld.

Follow @allen_ai and @semanticscholar on Twitter, and subscribe to the AI2 Newsletter to stay current on news and research coming out of AI2.

Joseph Chee Chang
AI2 Blog

😉💻🔄 Research Scientist @ AI2/Semantic Scholar | prev @ Carnegie Mellon

No comments:

Post a Comment