4 Pages

Lecture 7a

Course: WIN 235, Fall 2009
School: Stanford
Rating:
 
 
 
 
 

Word Count: 976

Document Preview

235, MS&E Internet Commerce Stanford University, Winter 2007-08 Instructor: Prof. Ashish Goel, Notes Scribed by Shrikrishna Shrin Lecture 7a: Long Tails Imagine a market where some class of products is being sold. We plot the ith largest product on the X axis and the market size (Revenue/No of units sold/No of customers) on the Y axis. A long tail typically looks like the graph in the image 1 below. The...

Register Now

Unformatted Document Excerpt

Coursehero >> California >> Stanford >> WIN 235

Course Hero has millions of student submitted documents similar to the one
below including study guides, practice problems, reference materials, practice exams, textbook help and tutor support.

Course Hero has millions of student submitted documents similar to the one below including study guides, practice problems, reference materials, practice exams, textbook help and tutor support.
235, MS&E Internet Commerce Stanford University, Winter 2007-08 Instructor: Prof. Ashish Goel, Notes Scribed by Shrikrishna Shrin Lecture 7a: Long Tails Imagine a market where some class of products is being sold. We plot the ith largest product on the X axis and the market size (Revenue/No of units sold/No of customers) on the Y axis. A long tail typically looks like the graph in the image 1 below. The graph is said to have a long tail if it can be represented by an equation of the form m(i) a(b + i)e where e is called the exponent. Notice that the market size only decreases polynomially with increase in i. For simplicity, let us assume that m = ie . We can plot another curve with market size on the X axis and frequency (number of items with the corresponding market size) on the Y axis. We nd that as market size increases, frequency decreases. di dm It is observed that long tails are created as a consequence of human endeavor or manufactured (ie. not natural) activities of human beings. Natural traits however do not exhibit the long tail phenomenon and tend to decay exponentially (of the form bi , notice that i is in the exponent in this case) instead. A few quotes on how long tails are created: f = Long tails are a signature of human activity. 1 Original image obtained from http://www.entrepreneurs-journey.com/images/long-tail- graph.gif 1 Long tails are a result of the tension between human individualism and social behavior. For example, if customers were to walk into a shoe store, some customers would buy a Nike shoe because it is a popular shoe and many of his friends may have bought that shoe (social behavior). However, some people would prefer to buy a shoe that is not very common and it is this tension between social behavior and individualism that leads to the long tail. If all customers bought a shoe only based on social behavior, then everyone would buy the same shoe. On the other hand, if everyone bought a shoe only based on individualism, everyone would buy a dierent shoe (assuming there are enough choices). Modeling the long tail phenomenon Let us assume that customers arrive one by one at a music website. Every customer does two things. 1. Choose a previously heard song with probability proportional to the number of times it has been heard in the past (Social behavior) 2. Find a new obscure song and listen to it. After t customers arrive, we have 2t hearings and t songs that have been heard. Let mt (1), mt (2), .....mt (t) = the number of times song i has been heard after t steps. Probability that a customer at time t + 1 listens to the rst song is mt (1)/2t. Probability that t + 1th customer song chooses i(i t) = mt (i)/2t. Somewhat informally, mt (i)/2t can be thought of as the expected increase in volume of product i (song i) at time t + 1. Therefore, mt+1 (i) = mt (i) + mt (i)/2t. Rewriting the above equation we get, mt (i) mt+1 (i) mt (i) = t+1t 2t On taking the limit, the above equation becomes: dmt (i) = mt (i)/2t dt On rearranging the above equation we get: dmt (i) dt = mt (i) 2t On integrating with limits i to t ( mt (i) is not dened for t < i) we get: ln(mt (i)) ln(mi (i)) = 1 (ln(t) ln(i)) 2 2 Now, ln(mi (i)) is 0 as at t = i, mi (i) = 1. Therefore, we get: t i This equation is the power law. Suppose customers were to choose a new product with probability p and an old product with probability 1 p then the only change in the power law will be the value of the exponent. Now we can obtain the frequency as: mt (i) = f = 1 di = dm dm di 2 = 3 1 t 3 i2 i Substituting for i we get, 3 2 t ( mt )3 t t m3 As m increases, f of selling the item decreases. However, the volume of sale can still be signicant. Determining the median The median is that point at which half the market lies in the tail. Now, we can derive that market size = 2t as follows: t mi (t) i Integrating from limits i = 1 to i = t we get, t mi (t) di = 1 t t 1 1 di i = 2 t( t 1) 2t mi (t) di = Let M be the total market size and med be a value such that 1 M 2 = t (M = 2t). (Half the market lies below this product and half the market lies after this product (ie. in the tail)). 3 med Therefore, 2 t med t. t Which implies med 4 . The top 25% of the pro...

Find millions of documents on Course Hero - Study Guides, Lecture Notes, Reference Materials, Practice Exams and more. Course Hero has millions of course specific materials providing students with the best way to expand their education.

Below is a small sample set of documents:

Stanford - WIN - 235
MS&amp;E 235, Internet CommerceStanford University, Winter 2007-08Instructor: Prof. Ashish Goel, Notes Scribed by Shrikrishna ShrinLecture 11: Small WorldThe small world conjecture is that we are all connected to each other through a small path of a
Stanford - WIN - 235
MS&amp;E 235, Internet CommerceStanford University, Winter 2007-08Instructor: Prof. Ashish Goel, Notes Scribed by Shrikrishna ShrinLecture 12: Small World and Internet ArchitectureSmall World Model Continued Consider a network structure with nodes A
Stanford - WIN - 235
MS&amp;E 235, Internet CommerceStanford University, Winter 2007-08Instructor: Prof. Ashish Goel, Notes Scribed by Shrikrishna ShrinLecture 14: Internet Architecture Contd.Note:TCP requires the following information: Source IP address, Destination I
Stanford - WIN - 235
MS&amp;E 235, Internet CommerceStanford University, Winter 2007-08Instructor: Prof. Ashish Goel, Notes Scribed by Shrikrishna ShrinLecture 15: Peer To Peer Networks and iTunesPeer to Peer Networks Vs. YoutubeWhile it might appear that P2P networks
Stanford - WIN - 235
C+ CODE FOR PROBLEM 1. #include &lt;stdio.h&gt; #include &lt;stdlib.h&gt; #include &lt;assert.h&gt; int main(int argc, char *argv) { int n,i; int a,b,c,d; assert (argc = 2); n = atoi(argv[1]); assert(n &gt; 0); assert(n &lt; 1000); for (i=0; i&lt;n; +i) { a = random() % 256; b
Stanford - WIN - 235
Gittins Index index theta V beta alpha 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 478 The indifference point (or the Gittins index) for an (alpha,beta) machine is
Berkeley - EE - 290
Course Announcement for the Fall of 2003Special Issues in Semiconductor ManufacturingEE290HProfessors Costas J. Spanos and Kameshwar Poolla Tue &amp; Thu 9:30-11:00Grading based on 2 Exams, Homeworks and Independent Projects (3 Units) This course ad
UConn - MATH - 211
Math 211Solutions to Practice Final Exam1. Use Euler's Method (with t = 0.1) to approximate y(0.3) where y(t) is the solution to the initial value problem: dy = t2 - y, dt Answer: yk+1 = yk + f (tk , yk )t k 0 1 2 3 tk 0 .1 0.2 0.3 yk 1 0.9 0.811
UConn - MATH - 211
UConn - MATH - 211
Sveriges lantbruksuniversitet - MATH - 416
MATH 416 - Problem set #4Due, Friday, November 29, 2002 Problem 1 Dispersion relations. Consider a &quot;normal mode&quot; or &quot;plane wave&quot; solution u(x, t) = ei(t+x) to a linear p.d.e. with constant coefficients that is first order in t. The wave number is r
University of Texas - ME - 244
ME 244L Dynamic Systems and Controls Lab Prof. R.G. LongoriaDepartment of Mechanical Engineering The University of Texas at Austin Spring 2000Experimental Determination of Parameters for a Linear Second Order System For a physical system that is
University of Texas - ME - 244
ME 244L Dynamic Systems and Controls Lab Prof. R.G. LongoriaDepartment of Mechanical Engineering The University of Texas at Austin Spring 2000Note on Accelerometers1IntroductionSensors that detect acceleration are called accelerometers. The
University of Texas - ME - 244
ME 244L Dynamic Systems and Controls Lab Prof. R.G. LongoriaDepartment of Mechanical Engineering The University of Texas at Austin Fall 2000Modeling of a Permanent-Magnet DC Motor1IntroductionThe perpetual growth in the role of dc machines
Wisconsin - CS - 838
enrightNatalie Enright and Dana VantreaseTo Include or Not to Include?C:\Documents and Settings\enrightn\My Documents\cs838.ppt12/12/03/p/course/cs838-david/public/html/presentations/enright.pptmatthewMatthew Allen, Allison HollowayExporing C
UMass (Amherst) - WORD - 3440
1Microbial Physiology MBIO 3440 Lab 1 Review of bacteria dilution and plating technique LAB REPORTMaintain report format and add requested information.Date: Student Name: Group #: 1. a) Record requested information in the following table. Table 1.
Sveriges lantbruksuniversitet - M - 495
Math 495 Special Topics in Applied Mathematics Spring 2007The Mathematical Origins of PatternsSpatial patterns abound in nature. Principles of energy and entropy might suggest that the natural tendency of the universe is a steady decay to boring
Sveriges lantbruksuniversitet - M - 462
Fluid Dynamics (MATH 462)SyllabusSpring 2002A Mathematical View of Fluid Motions Fluid dynamics, the study of the motion of liquids and gases, is one of the classical branches of applied mathematics. Sciences such as aerodynamics, hydrodyna
Temple - CIS - 307
This is my lab report on 11/04/96: 1 Example of using shared memory: #include &lt;stdio.h&gt; #include &lt;sys/types.h&gt; #include &lt;sys/ipc.h&gt; #include &lt;sys/shm.h&gt; #define SHM_MODE (SHM_R|SHM_W) int main(vo
Washington - PCC - 587
ATM 552 Notes: Matrix Methods: EOF, SVD, ETC.D.L.HartmannPage 514. Matrix Methods for Analysis of Structure in Data Sets:Empirical Orthogonal Functions, Principal Component Analysis, Singular Value Decomposition, Canonical Correlation Analysis
Washington - PCC - 587
PERSPECTIVESCalculated molecular conformations of triphenyl phosphite (TPP) (9). The structure on the right, found in the known crystalline phase of TPP, forms intermolecular hydrogen bonds. The structure on the left, thought to be the standard conf
Washington - PCC - 587
brief communicationsContrails reduce daily temperature rangeA brief interval when the skies were clear of jets unmasked an effect on climate.Change in DTR (C)The potential of condensation trails (contrails) from jet aircraft to affect regiona
Washington - PCC - 587
Chapter 3 Atmospheric ThermodynamicsRevised: 26 May 2005NOT FOR DISTRIBUTIONiiContents3 Atmospheric Thermodynamics 3.1 The Gas Laws . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 Virtual Temperature . . . . . . . . . . . . . . . . 3.2 T
Washington - PCC - 587
Contents2 The Earth System 2.1 Components of the Earth System . . . . . . . . . . . . . . 2.1.1 The oceans . . . . . . . . . . . . . . . . . . . . . . 2.1.2 The cryosphere . . . . . . . . . . . . . . . . . . . . 2.1.3 The terrestrial biosphere . . .
Washington - PCC - 587
Contents3 Atmospheric Thermodynamics 3.1 The Gas Laws . . . . . . . . . . . . . . . . . . . . 3.2 The Hydrostatic Equation . . . . . . . . . . . . . . 3.3 The First Law of Thermodynamics . . . . . . . . . 3.4 Adiabatic Processes . . . . . . . . . .
Washington - PCC - 587
ATMS 587: Notes on Empirical Orthogonal Functions and Principal Components D.L. Hartmann November 2002. Suppose we have a variable expressed in terms of space and time y(x,t). In Fourier analysis we can express this data set in terms of a specified s
Washington - PCC - 587
Washington - PCC - 587
VOL. 16, NO. 23JOURNAL OF CLIMATE1 DECEMBER 2003Letters ENSO-Forced Variability of the Pacic Decadal OscillationMATTHEW NEWMAN, GILBERT P. COMPO,ANDMICHAEL A. ALEXANDERNOAACIRES Climate Diagnostics Center, University of Colorado, Boulder,
Temple - CIS - 307
OrbixWeb Programmer's GuideIONA Technologies PLC November 1997IONA Technologies PLC The IONA Building 810 Lr. Pembroke St. Dublin 2 Ireland Phone: . . . . . . . . . . . . . +353-1-662 5255 Fax: . . . . . . . . . . . . . +353-1-662 5244 IONA Techn
Wisconsin - AOS - 452
1Lab #1: Data sources, WEATHER program, &amp; decoding METARsObjectives: Become familiar with available weather data, the weather program, and the decoding of METAR (surface) observations. DATA SOURCES We are fortunate enough to have access to a varie
Wisconsin - AOS - 452
1Lab #2: Decoding Model Data and Forecast Preparation:Objective: Become familiar with other types of data accessible using the weather program MODEL DATA: As was mentioned in the previous lab, the weather program can be used to access a variety of
Wisconsin - AOS - 452
1Lab #4: GEMPAK Part II: Gridded Data ProgramsObjective: Gain experience in using GEMPAK to create plots from gridded data sets GEMPAK: This lab will concentrate on working with observational data sets, so we will be using the GD(gridded data) pro
Wisconsin - AOS - 452
1Lab #10: Calculating the Q vectorObjectives: Compare Q vector calculations done by hand with those created in GEMPAK GEMPAK For this lab, we will be calculating Q vectors using GEMPAK for the same case study we have been using for several lab exe
Wisconsin - AOS - 452
1Lab #11: Isentropic Subjective AnalysesObjectives: Perform subjective analysis on an isentropic surface ISENTROPIC ANALYSES As discussed in lecture, isentropic analyses are useful for a variety of purposes. For this lab exercise, you will need to
Wisconsin - AOS - 452
1Lab #12: Adiabatic Method, Airstreams, and Limiting StreamlineObjectives: 1) Diagnose vertical motion by using the adiabatic method 2) Identify air streams through a cyclone (warm/cold conveyor belts) 3) Identify the limiting streamline Adiabatic
Harvard - EXTENSION - 22821
harvard university extension school star e113 Book Art II (22539) Spring 2009 Tuesdays 7:359:35 pm 51 Brattle Street, Grossman Common Room Dr. Ilavenil Subbiahcourse descriptionThis is an advanced book art course for students who have taken Book
Butler - ED - 245
M&amp;Ms Colors Company percentages 13 Student counts Brown Red count 1 1 count 2 count 3 count 4 count 5 count 6 count 7 count 8 count 9 count 10 count 11 count 12 count 13 count 14 count 15 count 16 count 17 count 18 count 19 count 20 count 21 total Cl
Butler - ED - 245
Is Halloween Candy a Rip Off?An Estimation Exercise and Cost Analysis EstimatesBefore opening your bag of M&amp;Ms, how many do you think are in the bag? (Please do not feel and count, but only estimate fro Estimate -&gt; 1How many M&amp;Ms do your think a
Butler - ED - 245
Title First Name Dr. and Mrs. Matt Mr. and Ms. oe J Mr. JudyLast Name Maurer Smith JonesAddress1 Address2 City 15501 Wildflower Lane Westfield 2 Sparks Lane Apt. 16 Indianapolis 100 1st St. IndianapolisState Zip Child IN 46074 Julie IN 46208 Sa
Butler - ED - 245
AddressBlock GreetingLine We will be having an open house in the classroom on March 16th. I know Child would really be happy to show you around the classroom and fill you in on all the work we have been doing. Please let me know if you will be able t
East Los Angeles College - EMAT - 10920
EMAT10920 Lesson 5 - ExercisesDr Philip J NaylorWorking individually, do the following exercises in H:\EMAT10920\lesson05 1. Write a program that mixes data types and printf() format specifiers. In particular, what happens if you print an int usin
UNC Wilmington - ECN - 324
Title: Series ID: Source: Release: Seasonal Adjustment: Frequency: Units: Date Range: Last Updated: Notes: DATE 1938-07-01 1939-07-01 1940-07-01 1941-07-01 1942-07-01 1943-07-01 1944-07-01 1945-07-01 1946-07-01 1947-07-01 1948-07-01 1949-07-01 1950-0
UNC Wilmington - ECN - 324
Financial I nst it ut ions &amp; M ar ket s ECN 324Phase 2 Exam ReviewCarcassonne, FR1M or t gage M ar ket s: Chapt er 9A. The US Gover nment insur es FH A and VA mor t gages. Why is t hat impor t ant t o t he lender ? B. What is a pass-t hr ough
UNC Wilmington - ECN - 324
Financial Markets &amp; Institutions ECN 324PHASE 1 REVIEWTOPICS FOR CH 1 What is the difference between Surplus vs. Deficit units? Who are the main providers of funds to the US Treasury? Who are the main buyers of government debt? What is the di
UNC Wilmington - ECN - 324
Financial Institutions &amp; Markets ECN 324Phase 2 Exam ReviewCarcassonne, FR1Mortgage Markets: Chapter 9A. The US Government insures FHA and VA mortgages. Why is that important to the lender? B. How does a second mortgage differ from a home equ
UNC Wilmington - ECN - 324
Year 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 1 2007 2 2008Energy GDP Energy consumption billion consumption, per GDP,
UNC Wilmington - ECN - 324
Oil &amp; Energy Consumption per GDP 2008 Annual Energy Review US Department of Energy Source: 2008 Annual Energy ReviewEnergy Consumption per GDP (2000 Chained $)20 18 16 14 12 10 8 6 4 2 1973 1977 1975 060,000 40,000 20,000 0 100,000 80,000 12
UNC Wilmington - ECN - 324
October 26, 2008Why the Mortgage Crisis HappenedBy M. Jay Wells Obama's economic narrative of the mortgage crisis ignores the facts. He has put free-market capitalism at the root of the current mortgage industry debacle, denying the real history o
Mich Tech - GE - 2800
The water needed to have the Dutch drink coffeeA.K. Chapagain A.Y. HoekstraAugust 2003 Value of Water Research Report Series No. 14UNESCO-IHE P.O. Box 3015 2601 DA Delft The NetherlandsContact author: Arjen Hoekstra Phone +31 15 215 18 28 E-mai
Mich Tech - GE - 2800
National Primary Drinking Water StandardsContaminant Acrylamide OC OC Alachlor Alpha particles R Antimony IOC Arsenic IOC IOC OC IOC Benzene OC Benzo(a)pyrene (PAHs) OC Beryllium IOC Beta particles and photon emitters R Bromate Cadmium IOC Carbofura
Mich Tech - GE - 2800
PrecipitationInception of precipitation Precipitation occurs when large masses of warm air steadily move higher and cool to the point where condensation occurs. Uplift of air masses can occur via convection: caused by differential solar heating
UCSC - A - 2414
# BLASTP 2.2.17 [Aug-26-2007]# Query: PAE2414# Database: /projects/compbio/data/pdb/dunbrack-pdbaa# Fields: Query id, Subject id, % identity, alignment length, mismatches, gap openings, q. start, q. end, s. start, s. end, e-value, bit score
UCSC - A - 2414
PAE2414.t04 CB8-sep93 2 12730405060E F A 3 2 1F D A F G DGD E F ED AE A E D E F G F I G E I F K I H I E DG E H70 120AL Q SANPV VKSIALAVAGESKAKAETP L SV V KY G V YY V Y E GG G W A LS A A PI F D778090100A 3 2 1
UCSC - A - 2414
PAE2414.t04 alpha3 2 12730405060H 3 2 1S B HB I B S BEA E A EDS E70AL Q SANPV VKSIALAVAGESKAKAETP L SV V KY G V YY V Y E GG G W A LS A A PS B778090100110I S B 3 2 1EB D S B I HB S B EG SEA D S E
UCSC - A - 2414
PAE2414.t04 bys4 3 2 12730405060H 4 3 2 1PE P EY EP T S P E70AL Q SANPV VKSIALAVAGESKAKAETP L SV V KY G V YY V Y E GG G W A LS A A PP100110H 4 3 2 1E PH EY PHT EP T GEY E120778090SA V STPQL YA
UCSC - A - 2414
PAE2414.t04 dssp-ehl22 12730405060H 2 1HEE100110E 2 1HEE127130140150160EEEE170DG A VAITH YLPMCVSGTDFRTETVGQY F TY K GV R F RV R L V II R C120778090SA V STPQL YAVGFGNCPADISGLYGRA Y AP G SN
UCSC - A - 2414
PAE2414.t04 near-backbone-112 12730405060G 2 1DJ DG I E D I J D I K I G AD I B G A DI D K G D I D J F J I G D A J E I K J D G A100110G A G 2 1D K I K J ID H AK DID GAG K G K G D I D G I J H D I D120778
UCSC - A - 2414
PAE2414.t04 o_notor23 2 12730405060H G N 3 2 1HNB N B N A B A BNB70AL Q SANPV VKSIALAVAGESKAKAETP L SV V KY G V YY V Y E GG G W A LS A A PA B N778090100110B P 3 2 1NHNG NBA B NPN127130140
UCSC - A - 2414
PAE2414.t04 str24 3 2 12730405060H 4 3 2 1HTSY Z A ZY Z AY TA70AL Q SANPV VKSIALAVAGESKAKAETP L SV V KY G V YY V Y E GG G W A LS A A PZ100110T 4 3 2 1Z AZ YSHTTSASE AY Z Y Z Y Z T127130
UCSC - A - 2414
PAE2414.t06 o_sep5 4 3 2 12730405060H D 5 4 3 2 1HDAK A G DP A T A D778090100110A 5 4 3 2 1DH B HD G DG DAD A120SA V STPQL YAVGFGNCPADISGLYGRA Y AP G SN V V HV T G C SI V V P SI A K ED K G D1271
Washington - M - 124
I2~sIqfWTIgzB2c}Ux2iGrpAep2cin2fmg 9 ~rW}ixc( V B 0 s v V f R 0 f { 0 R v Q Du D 0 v 0 0 f ( Q 0 V R &amp; 0 F RT uuT { B 0 f 0V R nQ2fmi 9 FdPR2DW2'G2Wgf2'!cr'W}2c}cGWgWr'4pzcrcWcCtah 0 T Q u 0 ` u v QT R 0 ` D uuT { B 0 f { B DT R u D `