BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Mapping Inventions in the Space of Ideas\, 1836–2022: Representa
 tion\, Measurement\, and Validation
DTSTART:20240301T114500
DTEND:20240301T130000
DTSTAMP:20260427T213103Z
UID:98d2599f34f21d2643d88506a11b467572c5bb21d5f144d9dbb7f231
CATEGORIES:Conferences - Seminars
DESCRIPTION:Vitaly Meursault - Federal Reserve Bank of Philadelphia\nIna G
 anguli\, Jeffrey Lin\, Vitaly Meursault\, Nicholas Reynolds\n\nHow well ca
 n different methods meaningfully represent inventions in the “space of i
 deas?” We evaluate four leading natural language processing (NLP) models
 \, each of which produces a different numerical representation of patent t
 ext. We design three novel\, domain-specific validation tasks to select be
 tween these representations. Sentence-BERT (S-BERT) significantly outperfo
 rms other widely-used NLP models\, creating metrics better aligned with bo
 th expert and non-expert human judgment about patent similarity. The choic
 e of representation matters significantly for economic measurement. Accord
 ing to S-BERT\, contemporaneous patents have declined in similarity over m
 ore than a century\, as inventions have “spread out” on an expanding k
 nowledge frontier. Other representations report ambiguous or diverging pat
 terns. We reproduce the S-BERT result using newly-digitized records of his
 torical interferences\, which show secular declines in the rate of multipl
 e invention. Our results highlight the importance of validation and model 
 selection as an essential step in constructing and using measures derived 
 from patent text.\n\n 
LOCATION:UniL Campus\, Room Extra 126
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR