Google Books: The Metadata Mess

8/8/2019 Google Books: The Metadata Mess

1/29


2/29


3/29


4/29


5/29


6/29

Three ways of using GBS!"Batch processing": data mining and ""electronic philology" !

"It's only reporters and computational linguists whocare if [hit-count estimation] is really precise." PeterNorvig, Google !

Text databases and the "new philologies": !The importance of language to social, intellectual, andpolitical history & literary study !

Coincides emergence of large-scale historical textdatabases !

When did happiness replace felicity in 17 th c?!

Plotting the rise & fall of propaganda !

How did liberalismspread in the early nineteenth-centuryEuropean context?. "6!


7/29

Good enough for scholarship? !Will GBS be an adequate resource for scholarlyneeds now and in the future? !

Depends on: !Quality of imaging !Reliability and robustness of search tools !

Quality and reliability of metadata !e.g., date, edition history, author, subject classication,

etc.!

7!


8/29

Good enough for scholarship? !Will GBS be an adequate resource for scholarlyneeds now and in the future? !

Depends on:!

Quality of imaging !

Reliability and robustness of search tools !

Quality and reliability of metadata !e.g., date, edition history, author, subject classication,

etc.!

But GBS metadata are awful. !

8!


9/29

Quality Issues : !

Botched Scans, OCR, &c.!

9!


10/29

10!

Metadata Issues: !

1899, annus mirabilis !


11/29

Random Dates!

11!

1905!

1900!

1848!

1888!


12/29

The pervasiveness of

misdatings !527 hits returned for"Internet" before 1950 !

12!

1899 !

1905 !

1878 !

1905 !

1946 !

1905 !

1905 !

1905 !

1939 !


13/29

Famous before their lifetime !182 hits reported for "CharlesDickens" before birthdate(1812) !

Cf Jimi Hendrix, 81; LedZeppelin, 59 etc. !

13!

1878 !

1905 !

1946 !

1905 !

1905 !


14/29

Ego-surng,Edgar Cayce

Style!

14!

"Our reputationprecedes us" !


15/29

The frequency of misdatings !

15!

Search on "candy bar" < 1920yields 66 hits, 46 of themmisdated (70%) !


16/29

Classication Errors!

16!


17/29

Classication Errors!

17!


18/29

The Pervasiveness of

Misclassication!

18!

family and relationships (4)

fiction (4)

biography and autobiography (1)

Unlabeled (1)(others classified as "music,""history," "literary collections")

Classications of rst 10 hits for !Tristram Shandy !


19/29

The Pervasiveness of

Misclassication!

19!

First 10 hits for Leaves of Grass classify it as: "

Juvenile Nonction"Poetry !Fiction!

Literary Criticism!Biography & Autobiography,!

Counterfeits and Counterfeiting !


20/29

More bad metadata !

20!


21/29

More bad metadata !

21!

Reader, Imarketed him .


22/29

Other metadata issues !Books ascribed to authors of introductions, orgiven no author at all. !

22!


23/29

Other metadata issues !Titles linked to unrelated works. !

23!


24/29

Other metadata issues !Strange bedfellows !

24!


25/29

Who is to blame and what is

to be done? !"We got the metadata from the libraries": !

yes, sometimes but libraries didn't classify Hamlet as"antiques and collectibles" or Speculum as "Health & Fitness" !

Libraries don't use BISAC headings like "Antiques andCollectibles" and "Health & Fitness" in the rst place !

And publishers didn't assign BISAC codes to bookspublished before the 1980's !

25!


26/29

The world according to BISAC!Making space for Bambi & Bullwinkle !

and Schiller, Petrarch & Verlaine !

26!


27/29

The world according to BISAC!Making shelf space for Bambi & Bullwinkle !

and scrunching together Schiller, Petrarch & Verlaine !

27!Squeezing the universal library into a sububan bookstore !


28/29

Correcting the Problem!Google: "We're on it (but it isn't a rst priority)" !

Correcting errors as noticed (like bad scans)? !

Crowd Sourcing? !

But errors/bad metadata affect 000,000's of records !

"Error correction" doesn't address poor & missingmetadata, inconsistent/confusing/inappropriateclassication schemes !

Why should the metadata decisions be left to Googleengineers? !

28!


29/29

Correcting the Problem!HathiTrust to the rescue? !

But HathiTrust makes available only out-of-copyrightworks, has (relatively) limited computational resources !

Why should Google have no obligations to doGBS right? !

Google Book Search is "a tremendous public good forstudents, for teachers, for scholars, for everyone."Derek Slater, Google !

But a public good implies a public trust !

29!

Documents

Google Books: The Metadata Mess