for Real Estate Appraisal Yanjie Fu Joint work with Hui Xiong Yu Zheng Yong Ge Zhihua Zhou Zijun Yao Rutgers the State University of New Jersey Microsoft Research ID: 1022179
Download Presentation The PPT/PDF document "Exploiting Geographic Dependencies" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1. Exploiting Geographic Dependencies for Real Estate AppraisalYanjie FuJoint work with Hui Xiong, Yu Zheng, Yong Ge, Zhihua Zhou, Zijun YaoRutgers, the State University of New JerseyMicrosoft Research, UNC Charlotte, Nanjing University KDD2014@New York City, NY
2. Background and MotivationProblem StatementMethodologyEvaluationConclusionsAgenda2
3. Housing Matters3Call for an intelligent system to rank estates based on estate investment value ZillowYahoo HomesWin your loverSettle down your family Make extra money as investmentRealtorshousing consultant services
4. What Special in Estate Investment ValueLocation! Location! Location!Urban Geography features the geographical utility of housesHuman Mobility reflects neighborhood popularity of housesProsperity of Business Area shows influence on houses4Road NetworksSubway StationsBus StopsPlaces of interestsTaxi GPS tracesBusinesses area
5. Quantifying Estate Investment Value5We don’t predict future priceWe predict growth potential of resale value (long term thing)Estate investment return rate of a given market period Prepare the benchmark investment values of estates for training dataIdentify rising market period and falling market period of BeijingCalculate the investment returns of each real estate during rising market period and falling market periodValue-adding capabilityinvestment returns of rising market period Value-protecting capabilityinvestment returns of falling market period represent Investment return
6. Geographic Dependencies (1)6Individual dependencythe investment value of an estate is determined by the geographic characteristics of its own neighborhoodPoint of InterestTransportation AccessibilityTaxi Mobility Your personal profiles tell your research interests
7. Geographic Dependencies (2)7Peer dependencyInside a business area, the estate investment value can be reflected by its nearby estates.AttributeA BDistance to Level2 road network 156 meters143 metersDistance to subway station1385 meters1585 meters#Restaurants 34#Transportation facilities88ABABYour friends tell your research interests
8. Geographic Dependencies (3)8Zone dependencyThe estate value can also be influenced by the values of its associated latent business area.Rising Market Period (02/2012-05/2012)Average regional prices Average regional prices Depressed business areaProsperous business areaDepressed business areaProsperous business areaYour associated research groups tell your research interests
9. Problem DefinitionGivenEstates with locations and historical pricesUrban geography (poi, road networks)Human mobility (taxi GPS traces)ObjectiveRank estates based on their investment valuesCore tasksPredict estate investment value using urban geography and human mobility Jointly model three geographic dependencies as objective function to learn estate ranking predictor9
10. Methodology Overview10Estate Investment ValueGeographic UtilityNeighborhood PopularityRoad NetworksSubway StationsBus StopsPlaces of interestsUrban Geography Cell -Tower TracesTaxicab GPS TracesCheck-insBus TracesHuman MobilityBusiness AreaInfluence of Business Area’s ProsperityPrediction ModelObjective FunctionIndividual DependencyPeer DependencyZone DependencyLocation
11. Modeling Estate Investment Value (1)Geographic UtilityFeature extraction by spatial indexinglinearly regress geographic utility from geographic features of the neighborhood of each estate11
12. Modeling Estate Investment Value (2)Influence of business area (A generative view)There are K business areas in a cityEach business area is a cluster of estates; estates tend to co-locate along multiple business areas The more prosperous, the more easier we identify an estate from business areaThe prosperities of K business areas (K spatial hidden states ) show influence on estatesThe inverse influence of geo-distance between estate and business area centerGaussian Mixture Model + Learning To Rank12Prosperities of K business areas
13. Modeling Estate Investment Value (3)Neighborhood Popularity (A propagation view)Propagate visit probability to POIs per drop-off pointAggregate visit probability per POIAggregate visit probability per POI categoryCompute popularity score Spatial propagation and aggregation from taxi to house13Point of InterestsHouseTaxi Passenger(drop-off point)Categories of POI
14. Modeling Three DependenciesIndividual DependencyCapture prediction accuracy of investment values , locations and business area assignment14Peer Dependency (pairwise analysis on estate level)Capture ranking consistency of predicted investment value of each estate pairZone Dependency (pairwise analysis on business area level)Capture ranking consistency of learned prosperities of the corresponding business area pair of each estate pairThe more accurate, the higher likelihoodOverall likelihood
15. Parameter EstimationGiven where Y , and L are the investment value, ranks and locations of I estates respectivelyTo maximize the posterior Parameters PriorsExpectation Maximization (EM) to learn the parameters by treating latent business area of each estate as a latent variableGeo-clustering updates the latent business area by maximizing the posterior of latent business areaAfter business area assignments are updated, maximizing the posterior of model parameters 15
16. Experimental DataBeijing real-world DataBeijing estate data2851 estates with transaction records from 04/2011 to 09/2012Falling market(04/2011 to 02/2012) and Rising market (02/2012 to 09/2012) Beijing transportation facility data including bus stop, subway, road networksBeijing POI dataBeijing taxi GPS traces16
17. Evaluation Methods and MetricsBaseline algorithmsMART: it is a boosted tree model, specifically, a linear combination of the out puts of a set of regression treesRankBoost: it is a boosted pairwise ranking method, which trains multiple weak rankers and combines their outputs as final rankingCoordinate Ascent: it uses domination loss and applies coordinate descent for optimizationListNet: it is a listwise ranking model with permutation top-k ranking likelihood as the objective functionEvaluation metricsNormalized Discounted Cumulative Gain (NDCG)Precision RecallKendall’s Tau Coefficient17
18. Overall PerformancesWe investigate the ranking performances by comparing to baseline algorithms in terms of Tau, NDCG, Pecision and Recall18Our method mines urban geography and human mobility to predict estate investment value and exploit three dependencies, and thus outperform traditional LTR methods
19. Study on Geographic DependenciesWe study the impact of three geographic dependencies by designing different variants of objective function (posterior likelihoods)19Individual dependency can achieve good overall ranking performance;Peer and zone dependency can achieve good top-k ranking performance;We recommend combination usage of three dependencies
20. Hierarchy of Needs for Human Life (1)20Triangle Need Structure of Human BeingWe hope we go shopping, work, eat, access transportation quickly and easily, children go to schools near our homeWe DONOT need hotels/hospitals/sports/spots located near our homeGood house is always balance people’s needThe POI density spectral of estates over multiple poi categories
21. Hierarchy of Needs for Human Life (2)21The triangle need structure of human life in Beijing
22. Conclusions Housing analysis is funny Use urban geography and human mobility to model estate investment valueCapture geographic individual, peer, and zone dependencies to better learn estate ranking predictorBusiness applicationsDecision making support for homebuyers Improve price structure for housing agent/broker/consultantOptimize site selection for housing developer22
23. 23Thank You !Email: yanjie.fu@rutgers.eduHomepage: http://pegasus.rutgers.edu/~yf99/