Discovering Functional Groups of an Area

Abstract

Disclosed herein are techniques and systems for discovering functional groups in an area, such as an urban area. A process includes segmenting a map of the area into sections, and inferring, for each section, a distribution of functions according to a topic model framework which considers mobility patters of users and points of interest (POIs) in the section. The topic model framework regards the section as a document, each function as a topic, the mobility patterns as words, and a POI feature vector for the section as metadata. The process may further include clustering the sections based at least in part on a similarity of the distribution of functions between each of the sections to obtain functional groups, estimating a functionality intensity for each of the functional groups, and annotating each of the functional groups.

1 - 20 . (canceled) 
     
     
         21 . A system comprising:
 one or more processors;   memory storing instructions executable on the one or more processors to perform acts comprising:
 segmenting a map of an area into a set of sections; 
 inferring a distribution of functions for each section in the set of sections according to a topic model framework which uses mobility patterns of users leaving from and arriving at each section and uses points of interest (POIs) in each section; and 
 determining functional groups of the area based at least in part on the distribution of functions. 
   
     
     
         22 . The system of  claim 21 , wherein the topic model framework is a Dirichlet Multinomial Regression (DMR)-based topic model. 
     
     
         23 . The system of  claim 21 , wherein the topic model framework stores a point of interest (POI) feature vector associated with the POIs in each section as metadata. 
     
     
         24 . The system of  claim 23 , wherein the POI feature vector for each section is based at least in part on frequency densities of different types of POIs in each section. 
     
     
         25 . The system of  claim 23 , wherein the topic model framework uses the set of sections as a set of documents, uses each function of the distribution of functions as a topic, and uses the mobility patterns as words. 
     
     
         26 . The system of  claim 21 , the acts further comprising estimating a functionality intensity for each of the functional groups based on a Kernel Density Estimation (KDE) model, wherein origin and destination data of the mobility patterns is provided as input to the KDE model. 
     
     
         27 . The system of  claim 21 , the acts further comprising annotating each of the functional groups on a visual representation of the area. 
     
     
         28 . The system of  claim 21 , wherein the inferring the distribution of functions further comprises generating a leaving transition matrix and an arriving transition matrix which represents leaving mobility patterns and arriving mobility patterns, respectively, obtained over a period of time. 
     
     
         29 . The system of  claim 21 , the acts further comprising determining one or more representative mobility patterns of each of the sections using a term frequency-inverse document frequency (TF-IDF) technique. 
     
     
         30 . A method of determining functional groups of an area, comprising:
 segmenting, by one or more processors, a map of the area into sections;   for each section:
 determining, by the one or more processors, mobility patterns of users leaving from and arriving at the section, 
 determining, by the one or more processors, one or more points of interest (POIs) that are located in the section, and 
 inferring, by the one or more processors, a distribution of functions for the section according to a topic model framework based at least in part on the mobility patterns and the one or more POIs; and 
   determining, by the one or more processors, a set of functional groups based at least in part on the distribution of functions for each section.   
     
     
         31 . The method of  claim 30 , wherein the topic model framework that applies the section as a document, applies each function of the distribution of functions as a topic, and applies the mobility patterns as words. 
     
     
         32 . The method of  claim 30 , wherein the mobility patterns comprise leaving mobility patterns and arriving mobility patterns determined over a period of time, each leaving mobility pattern describing a leaving time when the users leave from the section and a destination section that the users leave toward, each arriving mobility pattern describing an arriving time when the users arrive at the section and an origin section that the users arrive from. 
     
     
         33 . The method of  claim 30 , further comprising determining a number of the one or more POIs in each of a plurality of point of interest (POI) categories to obtain a POI feature vector for the section, wherein the POI feature vector is based at least in part on calculating a frequency density of each of the plurality of POI categories, and wherein inferring the distribution of functions for the section is further based on the POI feature vector. 
     
     
         34 . A method comprising:
 segmenting, by one or more processors, a map of an area into sections;   inferring, for each section and by the one or more processors, a distribution of functions according to a topic model framework which uses mobility patterns of users leaving from and arriving at the section and points of interest (POIs) in the section; and   clustering, by the one or more processors, the sections based at least in part on a similarity of the distribution of functions between each of the sections to obtain functional groups.   
     
     
         35 . The method of  claim 34 , wherein the topic model framework uses the section as a document, uses each function of the distribution of functions as a topic, uses the mobility patterns as words, and uses a point of interest (POI) feature vector of the POIs in the section as metadata. 
     
     
         36 . The method of  claim 35 , wherein the POI feature vector is based at least in part on calculating a frequency density of each of a plurality of POI categories. 
     
     
         37 . The method of  claim 34 , wherein the inferring the distribution of functions further comprises generating a leaving transition matrix and an arriving transition matrix which represents a frequency of the mobility patterns. 
     
     
         38 . The method of  claim 34 , further comprising estimating a functionality intensity for each of the functional groups based on a Kernel Density Estimation (KDE) model, wherein an origin and a destination of each of the mobility patterns is provided as input to the KDE model. 
     
     
         39 . The method of  claim 34 , wherein using the mobility patterns comprises determining one or more representative mobility patterns of each of the sections using a term frequency-inverse document frequency (TF-IDF) technique. 
     
     
         40 . The method of  claim 34 , wherein the segmenting the map is based at least in part on a road network of the area.