Same Same

Sometimes a claim is supported by multiple citations. One general form of this kind of a cited claim is: "Here is a claim (article_1, article_2, article_n)."

Multiple citations for the same claim mean that the cited research is making the same point.

This insight can be used to:

• Build similarity matrices between the cited articles
• Test whether the articles cited together are actually saying the same thing.
• Use 'autosum' for one cited article to learn about the other cited article (see the first point)

Final Outputs:

1. Process a directory of PDFs to create a CSV that includes 3 columns:

   cited_by_article_id, cited_article_id, claim

   If a claim is supported by multiple citations, then add a row for each cited_article. For instance, we can article with article ID '345X123' which includes a claim: "Earth is roughly a sphere (article_ID_445B333, article_ID_2243S52)." We expand it into 2 rows:

   345X123, 445B333, "Earth is roughly a sphere" 345X123, 2243S52, "Earth is roughly a sphere"

   But same pair of articles can be cited jointly for multiple claims. So later on article ID '345X123' may also have "Sun is hot (article_ID_445B333, article_ID_2243S52, article_ID_5D43S53)." And we will expand it as:

   345X123, 445B333, "Sun is hot" 345X123, 2243S52, "Sun is hot" 345X123, 5D43S53, "Sun is hot"

2. Use output from 1 to create n by n article similarity matrix (similar to cites/cited by matrices) where value of each entry = number of times articles are jointly cited as making the same claim (we lose information about what claim --- overlap on 'earth is roughly a sphere' and 'sun is hot' is counted together.)

            article_1, article_2, ...., article_n
   

   article_1, 1, 2, 0, .... article_2, 2, 1, 1, ....

   This will be a very sparse matrix. But even small overlaps are likely to be very informative.