본문

서브메뉴

Platform Operations: From Models to Methods
Platform Operations: From Models to Methods
Platform Operations: From Models to Methods

상세정보

자료유형  
 학위논문 서양
최종처리일시  
20260202104643
ISBN  
9798286498277
DDC  
005
저자명  
Kumar, Akshit.
서명/저자  
Platform Operations: From Models to Methods
발행사항  
[Sl] : Columbia University, 2025
발행사항  
Ann Arbor : ProQuest Dissertations & Theses, 2025
형태사항  
255 p
주기사항  
Source: Dissertations Abstracts International, Volume: 87-01, Section: B.
주기사항  
Advisor: Besbes, Omar;Kanoria, Yash.
학위논문주기  
Thesis (Ph.D.)--Columbia University, 2025.
초록/해제  
요약Online platforms - from on-demand home-services and e-commerce fulfillment networks to content streamers - thrive on their ability to match demand with supply. Doing so well requires algorithms that cope with uncertainty, high-dimensional type spaces, mis-aligned and multiple objectives, and informational frictions. This dissertation develops models and methods that illuminate these challenges and propose simple, and often near-optimal solutions. The dissertation is organized into five self-contained but thematically linked chapters - the first three chapters are related to problems in online matching, dynamic resource allocation and multi-objective optimization in order fulfillment problems, while, the last two chapters deal with design and optimization of recommendation systems.First, we tackle algorithm design questions that arise in online matching and dynamic resource allocation problems. How should a centralized matching platform dynamically match heterogeneous service providers with heterogeneous customers? What are the fundamental drivers of algorithmic performance in these dynamic resource allocation platforms? Is there a unifying algorithmic principle which can solve large class of dynamic resource allocation problems? How to make resource allocation decisions with multiple and often competing objectives? In Chapter 1, we study dynamic two-sided matching with heterogeneous demand and supply modeled as highly dimensional weight and feature vectors respectively. We show that simple myopic policies such as Greedy which are practically prevalent can impact be highly sub-optimal. We develop a forward-looking supply aware policy dubbed Simulate-Optimize-Assign-Repeat (SOAR). We prove that SOAR achieves the optimal regret scaling under different assumptions on the demand and supply distributions. In Chapter 2, we broaden the scope to general online resource allocation problems. We identify a novel driver of algorithmic performance - the spatial distribution of demand types. We develop a unifying algorithm dubbed Repeatedly Act using Multiple Simulations (RAMS) which is a generalization of SOAR studied in Chapter 1. In Chapter 3, we turn to multi-objective optimization in the context of order fulfillment problems. We develop a principled framework for weight generation to enable the weighted objective approach.Next, we take the viewpoint of a system designer and tackle platform design questions in the context of recommendation systems. By optimizing for measurable proxies, are recommendation systems at risk of significantly under-delivering on utility? If so, how can one improve utility which is seldom measured? Different information provisioning tools exists, such as public rankings and personalized recommendations, but when do these tools work and when do they not? What is the role of the market setting in the driving the efficacy of these different information provisioning tools? In Chapter 4, we study a stylized model of repeated user consumption. We demonstrate that optimizing for measurable proxies like engagement can lead to significant utility losses. Instead, we propose a utility-aware policy that initially recommends diverse set of options. As the platform becomes more forward-looking, our utility-aware policy achieves the best of both worlds: near- optimal utility and near-optimal engagement simultaneously. Our study elucidates an important feature of recommendation systems; given the ability to suggest multiple items, one can perform significant exploration without incurring significant reductions in engagement. By recommending high-risk, high-reward items alongside popular items, systems can enhance discovery of high utility items without significantly affecting engagement. In Chapter 5, we ask when personalized recommendations are worth their added complexity relative to public rankings. In unconstrained supply settings, both public rankings and personalized recommendations improve welfare, with their relative value determined by the degree of preference heterogeneity. In contrast, in supply-constrained settings, revealing just the common term of the utility, as done by public rankings, provides limited benefit since the total common value available is limited by capacity constraints, whereas personalized recommendations, by revealing both common and idiosyncratic terms, significantly enhance welfare by enabling agents to match with items they idiosyncratically value highly.
일반주제명  
Web studies
일반주제명  
Information technology
키워드  
Online platforms
키워드  
Algorithmic principle
키워드  
Platform design
키워드  
Idiosyncratic terms
키워드  
Simulate-Optimize-Assign-Repeat
기타저자  
Columbia University Business
기본자료저록  
Dissertations Abstracts International. 87-01B.
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008260126s2025        us                              c    eng  d
■001000017358319
■00520260202104643
■006m          o    d                
■007cr#unu||||||||
■020    ▼a9798286498277
■035    ▼a(MiAaPQ)AAI32114391
■040    ▼aMiAaPQ▼cMiAaPQ
■0820  ▼a005
■1001  ▼aKumar,  Akshit.
■24510▼aPlatform  Operations:  From  Models  to  Methods
■260    ▼a[Sl]▼bColumbia  University▼c2025
■260  1▼aAnn  Arbor▼bProQuest  Dissertations  &  Theses▼c2025
■300    ▼a255  p
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  87-01,  Section:  B.
■500    ▼aAdvisor:  Besbes,  Omar;Kanoria,  Yash.
■5021  ▼aThesis  (Ph.D.)--Columbia  University,  2025.
■520    ▼aOnline  platforms  -  from  on-demand  home-services  and  e-commerce  fulfillment  networks  to  content  streamers  -  thrive  on  their  ability  to  match  demand  with  supply.  Doing  so  well  requires  algorithms  that  cope  with  uncertainty,  high-dimensional  type  spaces,  mis-aligned  and  multiple  objectives,  and  informational  frictions.  This  dissertation  develops  models  and  methods  that  illuminate  these  challenges  and  propose  simple,  and  often  near-optimal  solutions.  The  dissertation  is  organized  into  five  self-contained  but  thematically  linked  chapters  -  the  first  three  chapters  are  related  to  problems  in  online  matching,  dynamic  resource  allocation  and  multi-objective  optimization  in  order  fulfillment  problems,  while,  the  last  two  chapters  deal  with  design  and  optimization  of  recommendation  systems.First,  we  tackle  algorithm  design  questions  that  arise  in  online  matching  and  dynamic  resource  allocation  problems.  How  should  a  centralized  matching  platform  dynamically  match  heterogeneous  service  providers  with  heterogeneous  customers?  What  are  the  fundamental  drivers  of  algorithmic  performance  in  these  dynamic  resource  allocation  platforms?  Is  there  a  unifying  algorithmic  principle  which  can  solve  large  class  of  dynamic  resource  allocation  problems?  How  to  make  resource  allocation  decisions  with  multiple  and  often  competing  objectives?  In  Chapter  1,  we  study  dynamic  two-sided  matching  with  heterogeneous  demand  and  supply  modeled  as  highly  dimensional  weight  and  feature  vectors  respectively.  We  show  that  simple  myopic  policies  such  as  Greedy  which  are  practically  prevalent  can  impact  be  highly  sub-optimal.  We  develop  a  forward-looking  supply  aware  policy  dubbed  Simulate-Optimize-Assign-Repeat  (SOAR).  We  prove  that  SOAR  achieves  the  optimal  regret  scaling  under  different  assumptions  on  the  demand  and  supply  distributions.  In  Chapter  2,  we  broaden  the  scope  to  general  online  resource  allocation  problems.  We  identify  a  novel  driver  of  algorithmic  performance  -  the  spatial  distribution  of  demand  types.  We  develop  a  unifying  algorithm  dubbed  Repeatedly  Act  using  Multiple  Simulations  (RAMS)  which  is  a  generalization  of  SOAR  studied  in  Chapter  1.  In  Chapter  3,  we  turn  to  multi-objective  optimization  in  the  context  of  order  fulfillment  problems.  We  develop  a  principled  framework  for  weight  generation  to  enable  the  weighted  objective  approach.Next,  we  take  the  viewpoint  of  a  system  designer  and  tackle  platform  design  questions  in  the  context  of  recommendation  systems.  By  optimizing  for  measurable  proxies,  are  recommendation  systems  at  risk  of  significantly  under-delivering  on  utility?  If  so,  how  can  one  improve  utility  which  is  seldom  measured?  Different  information  provisioning  tools  exists,  such  as  public  rankings  and  personalized  recommendations,  but  when  do  these  tools  work  and  when  do  they  not?  What  is  the  role  of  the  market  setting  in  the  driving  the  efficacy  of  these  different  information  provisioning  tools?  In  Chapter  4,  we  study  a  stylized  model  of  repeated  user  consumption.  We  demonstrate  that  optimizing  for  measurable  proxies  like  engagement  can  lead  to  significant  utility  losses.  Instead,  we  propose  a  utility-aware  policy  that  initially  recommends  diverse  set  of  options.  As  the  platform  becomes  more  forward-looking,  our  utility-aware  policy  achieves  the  best  of  both  worlds:  near-  optimal  utility  and  near-optimal  engagement  simultaneously.  Our  study  elucidates  an  important  feature  of  recommendation  systems;  given  the  ability  to  suggest  multiple  items,  one  can  perform  significant  exploration  without  incurring  significant  reductions  in  engagement.  By  recommending  high-risk,  high-reward  items  alongside  popular  items,  systems  can  enhance  discovery  of  high  utility  items  without  significantly  affecting  engagement.  In  Chapter  5,  we  ask  when  personalized  recommendations  are  worth  their  added  complexity  relative  to  public  rankings.  In  unconstrained  supply  settings,  both  public  rankings  and  personalized  recommendations  improve  welfare,  with  their  relative  value  determined  by  the  degree  of  preference  heterogeneity.  In  contrast,  in  supply-constrained  settings,  revealing  just  the  common  term  of  the  utility,  as  done  by  public  rankings,  provides  limited  benefit  since  the  total  common  value  available  is  limited  by  capacity  constraints,  whereas  personalized  recommendations,  by  revealing  both  common  and  idiosyncratic  terms,  significantly  enhance  welfare  by  enabling  agents  to  match  with  items  they  idiosyncratically  value  highly.
■590    ▼aSchool  code:  0054.
■650  4▼aWeb  studies
■650  4▼aInformation  technology
■653    ▼aOnline  platforms
■653    ▼aAlgorithmic  principle
■653    ▼aPlatform  design
■653    ▼aIdiosyncratic  terms
■653    ▼aSimulate-Optimize-Assign-Repeat
■690    ▼a0796
■690    ▼a0454
■690    ▼a0489
■690    ▼a0646
■71020▼aColumbia  University▼bBusiness.
■7730  ▼tDissertations  Abstracts  International▼g87-01B.
■790    ▼a0054
■791    ▼aPh.D.
■792    ▼a2025
■793    ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17358319▼nKERIS▼z이  자료의  원문은  한국교육학술정보원에서  제공합니다.

미리보기

내보내기

chatGPT토론

Ai 추천 관련 도서


    신착도서 더보기
    최근 3년간 통계입니다.

    소장정보

    • 예약
    • 소재불명신고
    • 나의폴더
    • 우선정리요청
    • 비도서대출신청
    • 야간 도서대출신청
    소장자료
    등록번호 청구기호 소장처 대출가능여부 대출정보
    TF17880 전자도서 대출가능 마이폴더 부재도서신고 비도서대출신청 야간 도서대출신청

    * 대출중인 자료에 한하여 예약이 가능합니다. 예약을 원하시면 예약버튼을 클릭하십시오.

    해당 도서를 다른 이용자가 함께 대출한 도서

    관련 인기도서

    로그인 후 이용 가능합니다.