video demo End-User Web Scraping: Google Scholar Edition Sarah - - PowerPoint PPT Presentation

▶
video demo end user web scraping google scholar edition
SMART_READER_LITE
LIVE PREVIEW

video demo End-User Web Scraping: Google Scholar Edition Sarah - - PowerPoint PPT Presentation

Aug 23, 2023 198 likes •838 views

video demo End-User Web Scraping: Google Scholar Edition Sarah Chasins data scraping tool input demonstration of how to collect the first row of a relational dataset F r o m h i g h l y s t r u c t u r e d w e b p a

slide-1
SLIDE 1

video demo

slide-2
SLIDE 2

End-User Web Scraping: Google Scholar Edition

Sarah Chasins

slide-3
SLIDE 3

F r

  • m

h i g h l y s t r u c t u r e d w e b p a g e s

data scraping tool input

demonstration of how to collect the first row of a relational dataset

  • utput

a script that collects the rest of the dataset

slide-4
SLIDE 4

case study: Google Scholar data

current author title year citations authors venue vapnik Statistical Learning Theory 1998 54228 VN Vapnik Wiley-Interscience vapnik The Nature of Statistical Learning Theory 1995 53976 V Vapnik Data mining and knowledge discovery vapnik Support-vector networks 1995 15513 C Cortes, V Vapnik Machine learning 20 (3), 273-297 vapnik A training algorithm for

  • ptimal margin classifiers

1992 6095 BE Boser, IM Guyon, VN Vapnik Proceedings of the fifth annual workshop

  • n Computational learning theory ...

vapnik An introduction to variable and feature selection 2003 6059 I Guyon, A Elisseeff The Journal of Machine Learning Research 3, 1157-1182 vapnik Gene selection for cancer classification using support vector machines 2002 4058 I Guyon, J Weston, S Barnhill, V Vapnik Machine learning 46 (1-3), 389-422 ... ... ... ... ... ...

slide-5
SLIDE 5

current author title year citations authors venue vapnik Statistical Learning Theory 1998 54228 VN Vapnik Wiley-Interscience vapnik The Nature of Statistical Learning Theory 1995 53976 V Vapnik Data mining and knowledge discovery vapnik Support-vector networks 1995 15513 C Cortes, V Vapnik Machine learning 20 (3), 273-297 vapnik A training algorithm for

  • ptimal margin classifiers

1992 6095 BE Boser, IM Guyon, VN Vapnik Proceedings of the fifth annual workshop

  • n Computational learning theory ...

vapnik An introduction to variable and feature selection 2003 6059 I Guyon, A Elisseeff The Journal of Machine Learning Research 3, 1157-1182 vapnik Gene selection for cancer classification using support vector machines 2002 4058 I Guyon, J Weston, S Barnhill, V Vapnik Machine learning 46 (1-3), 389-422 ... ... ... ... ... ...

case study: Google Scholar data

slide-6
SLIDE 6

scale authors limit

2000

papers per author limit

500

limits placed by user at demo time

slide-7
SLIDE 7

two central questions

did the tool generate a good script? at what age do researchers peak?

slide-8
SLIDE 8

did the tool generate a good script?

slide-9
SLIDE 9

should we trust this data at all?

vapnik Statistical Learning Theory 1998 54228 VN Vapnik Wiley-Interscience vapnik The Nature of Statistical Learning Theory 1995 53976 V Vapnik Data mining and knowledge discovery vapnik Support-vector networks 1995 15513 C Cortes, V Vapnik Machine learning 20 (3), 273-297 vapnik A training algorithm for

  • ptimal margin classifiers

1992 6095 BE Boser, IM Guyon, VN Vapnik Proceedings of the fifth annual workshop

  • n Computational learning theory ...

vapnik An introduction to variable and feature selection 2003 6059 I Guyon, A Elisseeff The Journal of Machine Learning Research 3, 1157-1182 vapnik Gene selection for cancer classification using support vector machines 2002 4058 I Guyon, J Weston, S Barnhill, V Vapnik Machine learning 46 (1-3), 389-422

S

  • c

h e c k i n g u p

  • n

t h e d a t a a f t e r w a r d s i s h a r d . . .

slide-10
SLIDE 10

what do we expect?

2000 authors up to 500 papers per author

slide-11
SLIDE 11

what did we actually get?

rows: 157,159

slide-12
SLIDE 12

what did we actually get?

rows: 157,159 unique authors: 1993

slide-13
SLIDE 13

what did we actually get?

rows: 157,159 unique authors: 1993

  • h

n

  • !

t

  • l

m e s s e d u p a n d I

  • n

l y h a v e a w e e k t

  • f

i x i t ?

slide-14
SLIDE 14

what did we actually get?

rows: 157,159 unique authors: 1993

  • h

n

  • !

t

  • l

m e s s e d u p a n d I

  • n

l y h a v e a w e e k t

  • f

i x i t ?

possible explanations: 1. tool doesn’t work as well as I thought :( (my problem) 2. data updates during scraping (problem inherent in long scraping tasks) 3. Scholar lists some authors twice (Scholar problem) 4. some authors share names (not a problem!)

maybe not!

slide-15
SLIDE 15

what did we actually get?

rows: 157,159 unique authors: 1993

more thorough author analysis: author names that appear separated by other author names:

Yves Deville : listed as author 183 and 191 Giovanni Pau : listed as author 355 and 1736 Henry Lin : listed as author 1024 and 1403 Fabrizio Messina : listed as author 1391 and 1396

authors whose citation counts jump in the middle of their runs:

Marco Ronchetti : listed as author 225 and 226 Joefon Jann : listed as author 810 and 811 Marcin Kubica : listed as author 1069 and 1070

remember papers were listed in order

  • f decreasing

citation count

slide-16
SLIDE 16

what did we actually get?

rows: 157,159 unique authors: 1993

more thorough author analysis: author names that appear separated by other author names:

Yves Deville : listed as author 183 and 191 Giovanni Pau : listed as author 355 and 1736 Henry Lin : listed as author 1024 and 1403 Fabrizio Messina : listed as author 1391 and 1396

authors whose citation counts jump in the middle of their runs:

Marco Ronchetti : listed as author 225 and 226 Joefon Jann : listed as author 810 and 811 Marcin Kubica : listed as author 1069 and 1070

remember papers were listed in order

  • f decreasing

citation count

Marco Ronchetti Defects in Amorphous Solids: a Possible Approach 1984آ M Ronchetti Computer Simulation in Physical Metallurgy, 129-143 Marco Ronchetti Dynamical Properties of Classical Liquids and Liquid Mixtures 1984آ G Jacucci, M Ronchetti, W Schirmacher Condensed Matter Research Using Neutrons, 139-161 Marco Ronchetti Didattica per competenze: che supporto dalla tecnologia?آ S Giaffredo, M Ronchetti, A Valerio Marco Ronchetti Insegnare l'informatica a non-informatici: emergenza annunciataآ S Giaffredo, L Mich, M Ronchetti Marco Ronchetti Some considerations from ontological standpoint of modeling processes in the social domainآ A Ghosh, M Ronchetti, R Ferrario Marco Ronchetti LEZIONI SUL TELEFONINO: PORTING IN AMBIENTE SYMBIANآ M Ronchetti, J Stevovic Marco Ronchetti Costruzione di un'interfaccia-utente per Lavagne Interattive Multimediali nel caso di simulazioni bidimensionali di fisicaآ M Ronchetti, N Dorigatti Marco Ronchetti A Service-Oriented Architecture for the NEEDLE (Next gEneration sEarch engine for Digital LibrariEs) Multimodal Search Engineآ M Ronchetti, MJN Krishnan, M Jarke Marco Ronchetti Predizione contestuale di termini per fornire supporto a studenti con varie forme di disabilitأ .آ A Zanella, M Ronchetti Marco Ronchetti Spacetime: A Two Dimensions Search and Visualisation Engine Based on Linked Dataآ M RONCHETTI, F VALSECCHI Marco Ronchetti Dipartimento di Informatica e Telecomunicazioni Universitأ degli Studi di Trento, 38050 Povo (Trento) Italyآ M Ronchetti Marco Ronchetti Dipartirnento di InfoImatica e Studi Aziendali Universitli di Trento via F. Zeni 8, 1-38068 Rovereto (TN) ITALYآ G Kovacs, G Succi, F Baruchelli, M Ronchetti Marco Ronchetti Lﻷ°ﻗuso di video su Internet nella didattica universitaria.آ M Ronchetti Marco Ronchetti Bond-orientational order in liquids and glasses 1983 1608 PJ Steinhardt, DR Nelson, M Ronchetti Physical Review B 28 (2), 784 Marco Ronchetti Icosahedral bond orientational order in supercooled liquids 1981 261 PJ Steinhardt, DR Nelson, M Ronchetti Physical Review Letters 47 (18), 1297

slide-17
SLIDE 17

what did we actually get?

rows: 157,159 unique authors: 1,993 unique author runs: 2,000

splitting into runs based on new author or jump in citation count

slide-18
SLIDE 18

what did we actually get? what if the runs weren’t the first 2,000?

Scholar page at end of run confirms they really were the first 2,000

slide-19
SLIDE 19

what did we actually get? what if the runs weren’t the first 2,000?

Scholar page at end of run confirms they really were the first 2,000

1. tool doesn’t work as well as I thought :( (my problem) 2. data updates during scraping (problem inherent in long scraping tasks) 3. Scholar lists some authors twice (Scholar problem) 4. some authors share names (not a problem!)

slide-20
SLIDE 20

what did we actually get? can we eliminate explanation 2 also?

1. tool doesn’t work as well as I thought :( (my problem) 2. data updates during scraping (problem inherent in long scraping tasks) 3. Scholar lists some authors twice (Scholar problem) 4. some authors share names (not a problem!)

slide-21
SLIDE 21

what did we actually get? what did we actually get?

slide-22
SLIDE 22

what did we actually get? what did we actually get?

slide-23
SLIDE 23

what did we actually get? can we eliminate explanation 2 also?

1. tool doesn’t work as well as I thought :( (my problem) 2. data updates during scraping (problem inherent in long scraping tasks) 3. Scholar lists some authors twice (Scholar problem) 4. some authors share names (not a problem!)

I s u s p e c t 3 i s t r u e c a u s e f

  • r

a l l s e v e n , b u t c a n ’ t b e p

  • s

i t i v e .

slide-24
SLIDE 24

what did we actually get?

slide-25
SLIDE 25

papers per author what we expect to see

many authors with few papers a few authors with many papers spike around 500, from truncation

what we don’t want to see

spikes around multiples of 20

slide-26
SLIDE 26

papers per author

slide-27
SLIDE 27

papers per author

  • ne paper authors?

turns out, yes

slide-28
SLIDE 28

at what age do researchers peak?

slide-29
SLIDE 29

citations by year

slide-30
SLIDE 30

citations by year

no future dates, though...

slide-31
SLIDE 31

citations by year

papers removed for having no year information

14,115 (9.0%)

papers removed for being more than 50 years from author mean

169 (0.1%)

papers remaining

142,875 (90.9%)

slide-32
SLIDE 32

citations by year

slide-33
SLIDE 33

citations by author-year

slide-34
SLIDE 34

citations by author-year

but this allows a few authors with high citation counts to skew results

slide-35
SLIDE 35

citations by author-year

David S. Johnson Computers and intractability 51,032 Peter E. Hart Pattern classification 46,535 vapnik The Nature of Statistical Learning Theory 53,976 vapnik Statistical Learning Theory 54,228

slide-36
SLIDE 36

citations by author-year

but this allows a few authors with high citation counts to skew results alternatives

authors’ percent citations by year authors’ highest cited paper years

slide-37
SLIDE 37

citations by author-year

each dot is one paper

slide-38
SLIDE 38

citations by author-year

slide-39
SLIDE 39

citations by author-year

across all authors, average percentage of citations that come in a given author-year

The average author receives about 9% of his or her total citations on papers from year 0 of his or her publishing career.

slide-40
SLIDE 40

citations by author-year

but this puts extra weight on early papers because some authors have short careers

for authors with 1 paper, 100% of citations in year 0...

slide-41
SLIDE 41

citations by author-year

1,340 authors with 10 years or more publishing

slide-42
SLIDE 42

citations by author-year

647 authors with 20 years or more publishing

slide-43
SLIDE 43

citations by author-year

285 authors with 30 years or more publishing

slide-44
SLIDE 44

citations by author-year

110 authors with 40 years or more publishing

slide-45
SLIDE 45

citations by author-year

10+ 20+ 40+ 30+

slide-46
SLIDE 46

citations by author-year

751 authors with 0-10 years publishing

slide-47
SLIDE 47

citations by author-year

732 authors with 10-20 years publishing

slide-48
SLIDE 48

citations by author-year

391 authors with 20-30 years publishing

slide-49
SLIDE 49

citations by author-year

187 authors with 30-40 years publishing

slide-50
SLIDE 50

citations by author-year

0-10 10-20 20-30 30-40

slide-51
SLIDE 51

citations by author-year

each dot is a paper 4 papers with very high citation counts not included

slide-52
SLIDE 52

most-cited papers

slide-53
SLIDE 53

most-cited papers

but still the problem with career length skewing results...

slide-54
SLIDE 54

most-cited papers

e a c h d

  • t

i s

  • n

e a u t h

  • r
slide-55
SLIDE 55

most-cited papers

slide-56
SLIDE 56

all papers

slide-57
SLIDE 57

all papers

slide-58
SLIDE 58

all papers

slide-59
SLIDE 59

truncation

recent papers may not have had time to accumulate citations authors still working may not have reached true peak yet

slide-60
SLIDE 60

truncation

recent papers may not have had time to accumulate citations authors still working may not have reached true peak yet

c

  • n

t r

  • l

l i n g f

  • r

c a r e e r l e n g t h h e l p s h e r e b i g c

  • n

c e r n , b u t r e m

  • v

i n g a u t h

  • r

s w h

  • ’

v e w r i t t e n i n l a s t 5 y e a r s l e a v e s

  • n

l y 6 8

slide-61
SLIDE 61

future work

remove the papers per author limit

good for analyzing my tool, not the author peak question

slide-62
SLIDE 62

future work

not all computer science authors tagged with “computer science” label

plans to search CS string and label, scrape common tags, then scrape larger set of authors

above approach -> larger data set

should allow better analysis of effects of truncation

slide-63
SLIDE 63

future work

collect data on conference committees (DBLP)?

aligning data with citation count data may reveal correlation

  • ther suggestions?