aiSelections: Computational Techniques for Matching Faculty Research Profiles to Library Acquisitions

aiSelections: Computational Techniques for Matching Faculty Research Profiles to Library Acquisitions

Peter M. Broadwell – CLIR Postdoctoral Fellow, UCLA LibraryTimothy R. Tangherlini – Professor, Scandinavian Section and Department of Asian Languages, UCLA

Faculty research interests

Publication records also are very useful, if parsed well

Advanced faculty profiles: The Opus project at UCLA (envisioned)

Faculty CV analysis in RapidMiner

TF-IDF vectors of faculty interestsstudy,0.833333333history,0.48984375work,0.259709821center,0.142113095culture,0.218452381artwork,0.322321429southeast,0.012946429architecture,0.12797619Asian,0.066145833bronze,0.0375research,0.196875translation,0.055133929south,0.071428571design,0.071130952image,0.078683036museum,0.145982143place,0.071316964

ancient,0.03813244Chinese,0.036830357national,0.094791667light,0.028869048landscape,0.033854167project,0.065848214field,0.058928571state,0.052380952material,0.050892857review,0.069568452east,0.060639881Indian,0.023214286make,0.080729167life,0.034375discipline,0.113095238change,0.050558036Asia,0.02641369

society,0.047209821form,0.04077381cultural,0.070833333salt,0.010044643institute,0.07421875house,0.009821429china,0.023809524survey,0.085044643period,0.01875report,0.083035714space,0.032291667serve,0.040848214learn,0.081026786world,0.044270833gallery,0.025446429analyze,0.084077381build,0.016666667

Monograph records from WorldCat$worldCatQuery = “(srw.lc+all+N1*+or+srw.lc+all+N2*+or+srw.lc+all+N3*+or+srw.lc+all+N4*+or+srw.lc+all+N5*+or+srw.lc+all+N6*+or+srw.lc+all+N7*+or+srw.lc+all+N8*+or+srw.lc+all+N9*+or+srw.lc+all+NB*+or+srw.lc+all+NC*+or+srw.lc+all+ND*+or+srw.lc+all+NE*+or+srw.lc+all+NK*+or+srw.lc+all+NX*+or+srw.lc+all+TR*)+and+srw.yr>2005+and+srw.yr<2014+and+srw.mt+any+bks+and+(srw.la+all+eng+or+srw.la+all+fre+or+srw.la+all+ger+or+srw.la+all+ita+or+srw.la+all+spa+or+srw.la+all+dut+or+srw.la+all+por)+not+srw.mt+any+juvenile+not+srw.mt+all+ebk+not+srw.mt+all+elc”;

Process all ~160,000 results…

Monograph records from WorldCat

Cosine similarity between faculty profiles and book records

French Impressionism

Early Qing painting

Cosine similarity between faculty profiles and book records

French Impressionism

Early Qing painting

UCLA faculty

Cosine similarity between faculty profiles and book records

French Impressionism

Early Qing painting

UCLA facultyBook A

Book B

Cosine similarity between faculty profiles and book records

French Impressionism

Early Qing painting

UCLA facultyBook A

Book B

Cosine similarity between faculty profiles and book records

French Impressionism

Early Qing painting

UCLA facultyBook A

Book B

Evaluation data setsActual selections, Jan 2007 – Feb 2013

◦ 10,471 books in targeted subject areas, published after 2005 (subset of WorldCat data set, described below)

◦ 3,573 firm orders, 6,989 approval plan ordersCirculation records, Jan 2008 – Feb 2013

◦ 4,118 new, unique titles acquired after Jan 2007 circulated between Jan 2008 and Feb 2013

◦ This is 39.3% of acquisitions since 2007◦ Firm orders were 10% more likely to circulate◦ 606 books published since 2006 were borrowed via

interlibrary loan, many at no cost (intra-UC)All potential selections, published 2006-

2012◦ 130,042 unique titles (duplicates resolved) published

from Jan 2006, as returned by WorldCat query

Circulation of actual selections vs. simulated algorithmic selections

Circulation of actual selections vs. simulated algorithmic selections

Faculty profile matching:Applications and considerationsAppend a “faculty match” score

to vendor approval list entries◦Helps to target selections for the

short and medium term◦Not as useful for long-term, large-

scale collection developmentRefine subscriptions to online

periodicals and other resources◦Requires that online subscriptions

can be done a la carte, rather than via bulk packages

Faculty profile matching:Future directionsEnhance faculty profiles

◦Promising, due to growth in publication bibliometrics, faculty network analysis tools like Vivo and Profiles

Enhance resource profiles by obtaining more data◦For pre-publication monographs: unlikely◦Might be possible with online publications

Incorporate graduate student, undergraduate research interests

Combine circulation-based selection recommendations with faculty interest data

aiSelections: Computational Techniques for Matching Faculty Research Profiles to Library Acquisitions

Peter M. Broadwell – CLIR Postdoctoral Fellow, UCLA LibraryTimothy R. Tangherlini – Professor, Scandinavian Section and Department of Asian Languages, UCLA