Urban Knowledge Extraction, Representation and Reasoning as a Bridge - - PowerPoint PPT Presentation

Urban Knowledge Extraction, Representation and Reasoning as a Bridge from Data City towards Smart City

Jaime De-Miguel-Rodríguez1, Juan Galán-Páez1 Gonzalo A. Aranda-Corral2, Joaquín Borrego-Díaz1

1 Dept. Computer Science and Artificial Intelligence. University of Sevilla-Spain 2 Dept. Information Technologies. University of Huelva-Spain

jdemiguel@us.es

Outline

• Motivation
• Formal Concept Analysis (FCA)
• Case 1: Self-City
• Case 2: Smart Pedestrian Mobility
• Case 3: Semandal
• Conclusions and future work

Motivation

• Massive availability of poorly structured data
• WWW, opendata, crowdsourced etc.
• Obtaining structured knowledge from digital

information aids to:

• Obtain information on cities structure and

dynamic

• Understand how citizens live and work within

the city

• Formal Concept Analysis (FCA) can be used to
• rganise knowledge and extract new concepts

from rear data

Formal Concept Analysis

• Automated conceptual learning theory
• Detects and describes regularities and

structures of concepts

• Also provides data reasoning methods
• Logical implications between attributes

(Stem Basis, Luxenburger Basis)

• Basic data structures:
• The Formal Context (O,A,I)
• The Concept Lattice

Formal Context

• A formal context (O,A,I) consists on:
• A set of objects (O)
• A set of qualitative attributes (A)
• A relation I between objects and attributes
• Basic operations. Extension and Intension

Intension of {Bream} is {Coast, Sea} Extension of {Sea} is {Bream, Sparus, Eel}

Formal Concept

• A concept is a pair (X,Y) where:
• X is a subset of O
• Y a subset of A
• The Intension (common attributes) of X is Y and the

extension (common objects) of Y is X

A concept

Concept Lattice

• The Concept Lattice contains all concepts within the context

New Classes

All concepts within the concept lattice: C1 := ({Escatofagus, Eel, Carp, Bream, Sparus},{}) [Any fish] C2 := ({Escatofagus, Eel, Carp},{River}) [River fish] C3 := ({Escatofagus, Eel, Bream, Sparus},{Coast}) [Coast fish] C4 := ({Escatofagus, Eel},{River, Coast}) [Estuary fish] C5 := ({Eel, Bream, Sparus},{Coast, Sea}) [Sea fish] C6 := ({Eel},{River, Coast, Sea}) [Euryhaline fish]

Estuary fish Euryhaline fish

Association Rules

• Stem basis is designed for true implications only.
• It does not take any exception into account.
• Association Rules (Luxenburger Basis):
• Support: Attribute set frequency (# covered objects)
• Confidence:

How we use FCA?

Case studies

• 1. Self-City: Estimating Social Perception on Housing Values
• 2. Exploiting Pedestrian Behavior for Smart Mobility
• 3. Semandal: Exploiting Real Time Government Information

SELF-CITY PEDESTRIAN SIMULATION SEMANDAL OBJECTS Houses Positions News ATTRIBUTES Size, price, trend, proximity values etc. Closer to destination?,

• bstacle, other.

Categories, keywords AIM Understand socio-economic dynamics Behaviour mining, pedestrian simulation in new scenarios Non-supervised clustering (news classification) KNOWLEDGE EXTRACTED Socio-economic patterns Patterns of pedestrian trajectories Semantic and hierarchical

• rganization of news

METHODOLOGY Apply FCA per street and compare lattices Use association rules as multi-agent behavior Use FCA lattice as a hierarchical structure of labels

Methodology

Case 1: Self-City

(self-city.com)

A context for Real State info

Attributes:

• Dimensions (small, medium, big)
• Price (very low, low, medium, high, very high)
• Price decreased/increased in the last 3 months
• Price with respect to other homes in the

neighbourhood (more expensive than average, average, cheaper than average)

• Amount of other homes for sale in the

surroundings (none, few, lots)

• Access to public transport

Objects:

• Houses

A Concept Lattice for Seville

• Collection of 6000 (approx.) for sale homes in the city of

Seville

• Global Concept Lattice with all the info aggregated
• Subsets (by streets, zones, ...) can be considered for a

more detailed analysis

Concept Lattices by streets

• Analysing street

dynamics:

• Comparison between

concept lattices associated to different streets

• Av. Kansas City
• Av. República Argentina

Similar lattices:

• A significant difference:

Home’s dimensions Idea:

• Analyse knowledge basis

Isolating differences

• In order estimate the influence of House

dimensions, Attribute associate to big flats is permuted in Avda. R. Argentina with the normal size attribute

• The resultant lattice is very similar to the

associated to Av. Kansas City

• However, it is interesting how similar these

implication basis are

Estimating true association rules in both

• Luxenburger (Kansas, 85%) ==> Lux(Republica’,

97%)

• Luxenburger (Republica’, 100%) ==> Lux(Kansas,

94%)

• That is, Knowledge about real estate of both streets

are essentially similar

Conclusion of the comparison

• Two different areas in the city, apparently very

different

• They have same behavior from a socio-

economic point of view of real estate markets (and the available information)

• The argumentation about why it occurs is aim of

urbanism specialists

Using the pattern within the District

Case 2

Exploiting Pedestrian Behavior for Smart Mobility

Motivation

• Av. Constitución, Sevilla
• Recently redesigned
• Potential problems for

pedestrian mobility

• Bike way and

tramway

• Terraces
• Temporary exhibitions

Observations Data on pedestrian mobility (non- aggregated) Artificial models of mobility Discrete Agent- Based model Formal Context Attribute selection & data collection

Aim & Methodolgy

Mobility patterns (implications)

Inference engine

Evaluation New scenarios

• Attributes -

Knowledge representation for pedestrians

• Qualitative observable features (attributes)

describing pedestrian neighbourhood

• Qualitative distances to destination
• Empty space?
• Obstacle/zone type
• Other features (social, environmental, ...)
• The feature selection is performed by an
• bserver in each case
• Similar to pedestrian’s perception of its

neighbourhood

P

++

• +
• +

= =

Destination

• Objects -

A Formal Context for Pedestrians

video

Goal: Assessment of urban planning

• Agent-based qualitative modelling of real urban

scenarios provides a simple but robust sandbox for:

• Detecting and isolating existing planning flaws
• Assessing the impact of hypothetical urban

planning changes before implementing them

• Simulating and understanding pedestrian

behavioural patterns

Case 3

Exploiting Real Time Government Information

Introduction

• Semandal is focused on the municipalities of

Andalucía, Spain.

• This scope was chosen to reduce the dimensions of

vocabularies, ontologies, and, even, databases.

• For this, we use Formal Concept Analysis.

Categories

• Attributes: categories + keywords
• Objects: news
• Refill all news adding all abstract concepts to

existent concrete concepts (superclasses)

Classification

• First step: Select the most

important words for each category and mostly in that category (not all)

• Creating a graph with resulting

words and categories.

• Some categories look like well

defined

Classification

• Build the formal context:
• Attributes are categories and words
• Objects are news
• Relations are words and categories previously extracted.
• We could build an emergent ontology from this. (out of

scope)

• Set of rules (association rules) obtained by means of FCA

Experiments

• We chose 2 news randomly

[Noticia 1] “El novillero de Écija Antonio David, proclamado triunfador de la V feria de novilladas de promoción la granada de plata”

Context A

Turismo Juventud

Context B

Turismo Cultura

Context C

Festejos

Experiments

Noticia 2] “El ayuntamiento da luz verde para la construcción de otras 75 viviendas protegidas”

Context A

Vivienda

Context B

Turismo Servicios sociales

Context C

Servicios sociales Obras

SELF-CITY PEDESTRIAN SIMULATION SEMANDAL OBJECTS Houses Positions News ATTRIBUTES Size, price, trend, proximity values etc. Closer to destination?,

• bstacle, other.

Categories, keywords AIM Understand socio-economic dynamics Behaviour mining, pedestrian simulation in new scenarios Non-supervised clustering (news classification) KNOWLEDGE EXTRACTED Socio-economic patterns Patterns of pedestrian trajectories Semantic and hierarchical

• rganization of news

METHODOLOGY Apply FCA per street and compare lattices Use association rules as multi-agent behavior Use FCA lattice as a hierarchical structure of labels

• Knowledge Engineering techniques can

enhance city services towards Smart Cities

• FCA is a qualitative analysis and reasoning tool

valid for urban, inter-urban and intra-urban contexts

• Future work is oriented to acquire better urban

knowledge mined from citizen’s sentiments and

• pinions

Conclusions

Contact: jdemiguel@us.es

Merci