Introduction and Motivation
1
Introduction and Motivation On the Semantic Annotation of Places in Location-Based Social Networks Mao Ye 1 , Dong Shou 1 , Wang-Chien Lee 1 , Peifeng Yin 1 , Krzysztof Janowicz 2 Problem Description Location-based Social Networks E.g., Facebook Place and Foursquare User check-in places SAP Framework 1 Department of Computer Science and Engineering The Pennsylvania State University {mxy177,dus212,wlee,pzy102}@cse.psu.edu 2 Department of Geography University of California, Santa Barbara {jano}@geog.ucsb.edu Tags are important Business categorization Location search Place recommendation Data cleaning … Tags are missing In our Foursquare and Whirrl dataset, there are a lot of places missing tags Places missin g tags Places with tags Places missin g tags Places with tags Foursquar e Whrrl 67% 33% 68% 32% Place semantic annotation (SAP) problem Multi-label classification problem Input User check-in logs <who, where, when> Some places are tagged Output Infer tags for the rest places Check-in logs Plac e Feature Extraction (FE) Component Binary Classifier For tag t 1 Binary Classifier For tag t 2 Binary Classifier For tag t m Feature extraction (FE) Check-in logs features Features to describe a place FE- Explicit Pattern FE- Implicit Relatedness 00:00 23:59 Day 1 Day 2 Day 3 Day 4 Day 5 Day 6 Day 7 Day 8 Bars Bars Bars Restaura nt Restaura nt Restaura nt Restaura nt Restaura nt Restaura nt Restaura nt Restaura nt Restaura nt Restaura nt Shopping Shopping Shopping Gym Health Beauty Spa Places checked in by the same user at around the same time (not necessarily the same day) are probably in the same category ? Evaluation Whrrl.com. 5,892 users, 53,432 places and 199 types of tags According to Yelp, we map 199 tags into 21 categories. Comparison: EP, IR and SAP (EP+IR)
description
1 Department of Computer Science and Engineering The Pennsylvania State University {mxy177,dus212,wlee,pzy102}@cse.psu.edu. 2 Department of Geography University of California, Santa Barbara {jano}@geog.ucsb.edu. Tags are missing - PowerPoint PPT Presentation
Transcript of Introduction and Motivation
Introduction and Motivation
On the Semantic Annotation of Places in Location-Based Social Networks
Mao Ye1, Dong Shou1, Wang-Chien Lee1, Peifeng Yin1, Krzysztof Janowicz2
Problem Description
Location-based Social Networks
E.g., Facebook Place and Foursquare
User check-in places
SAP Framework
1Department of Computer Science and EngineeringThe Pennsylvania State University
{mxy177,dus212,wlee,pzy102}@cse.psu.edu
2Department of GeographyUniversity of California, Santa Barbara
{jano}@geog.ucsb.edu
Tags are important
Business categorization
Location search
Place recommendation
Data cleaning
…
Tags are missing
In our Foursquare and Whirrl dataset, there are a lot of places missing tags
Places missing tags
Places with tags
Places missing tags
Places with tags
Foursquare Whrrl
67%
33%
68%
32%
Place semantic annotation (SAP) problem
Multi-label classification problem
Input
User check-in logs <who, where, when>
Some places are tagged
Output
Infer tags for the rest places
Check-in logs
PlacePlaceFeature Extraction (FE)
ComponentFeature Extraction (FE)
Component
Binary ClassifierFor tag t1
Binary ClassifierFor tag t1
Binary ClassifierFor tag t2
Binary ClassifierFor tag t2
Binary ClassifierFor tag tm
Binary ClassifierFor tag tm
Feature extraction (FE)
Check-in logs features
Features to describe a place
FE- Explicit Pattern FE- Implicit Relatedness
00:00
23:59
Day 1 Day 2 Day 3 Day 4 Day 5 Day 6 Day 7 Day 8
BarsBars BarsBars
BarsBars
RestaurantRestaurantRestaurantRestaurant RestaurantRestaurant
RestaurantRestaurant
RestaurantRestaurant
RestaurantRestaurant
RestaurantRestaurant RestaurantRestaurantRestaurantRestaurant
RestaurantRestaurant
ShoppingShopping ShoppingShopping
ShoppingShopping
GymGym HealthHealthBeautyBeauty
SpaSpa
Places checked in by the same user at around the same time (not necessarily the same day) are probably in the same category
?
EvaluationWhrrl.com. 5,892 users, 53,432 places and 199 types of tags According to Yelp, we map 199 tags into 21 categories.
Comparison: EP, IR and SAP (EP+IR)
