Upgrade to Pro — share decks privately, control downloads, hide ads and more …

X-Cite

Avatar for lannajin lannajin
October 18, 2015

 X-Cite

Avatar for lannajin

lannajin

October 18, 2015

More Decks by lannajin

Other Decks in Programming

Transcript

  1. Asserting Statements with Relevant Citations is Frustrating and Time-consuming. Writing

    is an essential form of communication. It however can be an inefficient process when more time is spent finding relevant citations than writing itself.
  2. Common Problem • Academia • Law • Patents Industry •

    Big Pharma • Medical • Education • Science • Industry This inefficiency in the writing process is a common problem found across many fields.
  3. The difficulty of finding relevant citations to support sentences is

    not due to a lack of available resources, but conversely because there are too many resources out there.
  4. Papers P1 Curabitur convallis sollicitudin tellus, fringilla tincidunt turpis aliquam

    vitae C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P2 Donec cursus, erat eget faucibus iaculis, libero massa sollicitudin nulla, eu vulputate ligula massa vitae odio C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P3 Lorem ipsum dolor sit amet, consectetur adipiscing elit C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 . Curabitur convallis sollicitudin tellus, fringilla tincidunt turpis aliquam vitae C4 Algorithm Within the Python framework, I scraped over 130,000 articles from open access journals
  5. Papers C1 C2 C3 C4 P1 Curabitur convallis sollicitudin tellus,

    fringilla tincidunt turpis aliquam vitae C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P2 Donec cursus, erat eget faucibus iaculis, libero massa sollicitudin nulla, eu vulputate ligula massa vitae odio C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P3 Lorem ipsum dolor sit amet, consectetur adipiscing elit C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 . Curabitur convallis sollicitudin tellus, fringilla tincidunt turpis aliquam vitae C4 Citations Algorithm Extracting all citations within articles and their corresponding sentences.
  6. Papers C1 C2 C3 C4 P1 Curabitur convallis sollicitudin tellus,

    fringilla tincidunt turpis aliquam vitae C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P2 Donec cursus, erat eget faucibus iaculis, libero massa sollicitudin nulla, eu vulputate ligula massa vitae odio C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P3 Lorem ipsum dolor sit amet, consectetur adipiscing elit C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 . Curabitur convallis sollicitudin tellus, fringilla tincidunt turpis aliquam vitae C4 Citations Algorithm For a given citation, we can contextually abstract the importance of the citation from the paper that cites it without having to read the citation source.
  7. Papers C1 C2 C3 C4 P1 Curabitur convallis sollicitudin tellus,

    fringilla tincidunt turpis aliquam vitae C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P2 Donec cursus, erat eget faucibus iaculis, libero massa sollicitudin nulla, eu vulputate ligula massa vitae odio C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P3 Lorem ipsum dolor sit amet, consectetur adipiscing elit C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 . Curabitur convallis sollicitudin tellus, fringilla tincidunt turpis aliquam vitae C4 Citations Algorithm “After the elucidation of the double helical structure of DNA in 1953, …” “The right-handed double helical structure of B-form DNA (B-DNA) has been known since 1953” “The bases are found to be ∼3.5 Å apart, which is characteristic of DNA double-helical structures” “After the description of DNA structure by Watson and Crick, …” “Watson and Crick published their article on the double helix structure of DNA in 1953” Watson & Crick (1953) For example, we can infer that Watson and Crick’s work was relevant to the double helical structure of DNA from a sample of articles that all cite Watson and Crick’s famous paper.
  8. Papers C1 C2 C3 C4 P1 Curabitur convallis sollicitudin tellus,

    fringilla tincidunt turpis aliquam vitae C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P2 Donec cursus, erat eget faucibus iaculis, libero massa sollicitudin nulla, eu vulputate ligula massa vitae odio C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P3 Lorem ipsum dolor sit amet, consectetur adipiscing elit C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 . Curabitur convallis sollicitudin tellus, fringilla tincidunt turpis aliquam vitae C4 Citations Algorithm For each citation and its corresponding sentences…
  9. Papers C1 C2 C3 C4 P1 Curabitur convallis sollicitudin tellus,

    fringilla tincidunt turpis aliquam vitae C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P2 Donec cursus, erat eget faucibus iaculis, libero massa sollicitudin nulla, eu vulputate ligula massa vitae odio C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P3 Lorem ipsum dolor sit amet, consectetur adipiscing elit C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 . Curabitur convallis sollicitudin tellus, fringilla tincidunt turpis aliquam vitae C4 Citations Algorithm x y z C1 C2 C3 C4 Doc2Vec Doc2Vec vectorizes each sentence and corresponding citation into multi-dimensional space.
  10. Papers C1 C2 C3 C4 P1 Curabitur convallis sollicitudin tellus,

    fringilla tincidunt turpis aliquam vitae C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P2 Donec cursus, erat eget faucibus iaculis, libero massa sollicitudin nulla, eu vulputate ligula massa vitae odio C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P3 Lorem ipsum dolor sit amet, consectetur adipiscing elit C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 . Curabitur convallis sollicitudin tellus, fringilla tincidunt turpis aliquam vitae C4 Citations Algorithm x y z C1 C2 C3 C4 Doc2Vec When a user inputs a sentence within the X-Cite app…
  11. Papers C1 C2 C3 C4 P1 Curabitur convallis sollicitudin tellus,

    fringilla tincidunt turpis aliquam vitae C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P2 Donec cursus, erat eget faucibus iaculis, libero massa sollicitudin nulla, eu vulputate ligula massa vitae odio C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P3 Lorem ipsum dolor sit amet, consectetur adipiscing elit C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 . Curabitur convallis sollicitudin tellus, fringilla tincidunt turpis aliquam vitae C4 Citations Algorithm x y z C1 C2 C3 C4 Doc2Vec The sentence is vectorized and mapped into the pre-existing multi-dimensional space.
  12. Papers C1 C2 C3 C4 P1 Curabitur convallis sollicitudin tellus,

    fringilla tincidunt turpis aliquam vitae C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P2 Donec cursus, erat eget faucibus iaculis, libero massa sollicitudin nulla, eu vulputate ligula massa vitae odio C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 P3 Lorem ipsum dolor sit amet, consectetur adipiscing elit C1 . Pellentesque velit ligula, viverra sit amet varius a, cursus eget libero C2 . Mauris rutrum cursus interdum C3 . Curabitur convallis sollicitudin tellus, fringilla tincidunt turpis aliquam vitae C4 Citations Algorithm x y z C1 C2 C3 C4 Doc2Vec X-Cite then returns the citations closest to the mapped vector as recommended citations.
  13. + Most-cited Citations Most Accurate Validation Because citations vary in

    how often they are cited, I was curious how citation frequency affects prediction accuracy. This figure shows the result of how citations were ranked when I validated a test set consisting of a thousand citations for each level citation frequency, where the red horizontal line indicates random. True Citation Rank Citation Frequency
  14. + Most-cited Citations Most Accurate Validation True Citation Rank Citation

    Frequency For example, the first box plot shows for the 1,000 citations that were input into my model once, that the true citations were ranked rather poorly, but better than random; while the last box plot shows for the 1,000 citations that were input into my model 9+ times, that the true citations were ranked a lot better than compared to random.
  15. + Most-cited Citations Most Accurate Validation True Citation Rank Citation

    Frequency In other words, citations frequently cited performed better in my model than those infrequently cited.