Update of time-invalid information in Knowledge Bases through
Mobile Agents
Ilaria Tiddi, Enrico Daga, Emanuele Bastianelli, Mathieu d’Aquin
@IlaTiddi
Research ProblemFocus
Query and update of knowledge bases (KB) containing statements that lose validity with time
Knowledge in KB• static (always valid)• dynamic (can expire)
ExampleThe KMi knowledge basePodium hasCoordinates [X, Y]Podium hasTemperature 22.5°C
Common Solutions• continuous KB update• sensors deployment
Problems
Costs• hardware deployment• power consumption• data collection
Flexibility • only querying at sensor locations
Usefulness• constant update of rarely queried info
Research Problem
Applications• large-scale areas • requiring monitoring• sensors cannot be used• things must be valid in time
Example Domains
Agriculture/Botanicmonitoring plants’ growing and problems in large-scale gardens
Urban Developmentmonitoring progress of large construction sites
Military/Humanitarian scenariosmonitoring movements/infrastructure status in conflict zones
Research Motivation
Solution
• move a sensor on request at query time • re-collect the “expired” info• update the KB the KB always returns time-valid info when queried
ExampleQ: “Which room is the most comfortable for my meeting?” A:• info about some temperature is too old• move a mobile sensor• update the KB• return time-valid results
Research Proposal
AimUpdate the time-invalid information of a KB using a mobile agent on demand and guarantee the results’ time-validity
ChallengesDetecting which is outdated information in the KB w.r.t. the query
Representing time-validity
Instructing the robot correctly Answering the query with time-valid statements Planning by favoring the information time-validity
Research Challenge
ApproachUse Semantic Web technologies as a knowledge representation framework
1. Time-validity representation• Resource Description Framework (RDF) to model the
time-validity of statements • SPARQL query language to decide the duration
based on a set of time-validity rules2. Planning
• focus on guaranteeing the time-validity of all the collected info
3. KB update • get data at query time with a robot• update the KB during collection
Proposed Approach
1. User query “Which room is the most comfortable for my meeting?”
2. Collection of expired statements in the answer“temperature of the MediaLab is valid,but in the Podium it needs to be re-sensed”
3. Planning the right sequence of actions“sense temperature in the Podium before temperature in the MediaLab expires”
4. Update the KB at each sensing“temperature in the Podium is 22.5 and will expire in 300secs”
5. Answer with time-valid statements“Right now, the Podium is the most
comfortable”
Implemented Process
1. RDF representation of static and dynamic info
Resource Description Framework a directed, labeled graph representation language connecting resources and literals of various datatypes
Basic components• RDF triple• a labeled link between two resources• resources are identified by their unique id (URI)• format <subject,predicate,object>
Representing Time-validity
22.5ºC<http://data.open.ac.uk/robo/location/Podium>location:Podium
<http://data.open.ac.uk/robo/properties/hasTemp>robo:hasTemp
2. Annotate RDF triples with their expiry-date
Extend triples into RDF quads• quad : <subject,predicate,object,namedGraph>
Named Graphs • cut general graph into sub-graphs• graph to which the triple belong • identified by a resource
Representing Time-validity
3. Decide the duration of a new information
Time-validity rules r={p,d}• rule : associates a triple pattern p to a duration d. • pattern : a triple with some variables (wildcards)• duration : how long a triple matching p will be valid
r={<?x robo:hasTemp ?y>, 300secs}
Matching a new triple t• Select the rules for which the triple matches the pattern• If more than one, take the most specific• Create new quads : <subject,predicate,object,currentTime+d>
Representing Time-validity
4. Assessing the time-validity of a query’s results
Handling user queries• Execute the SPARQL query onto the KB• Collect the KB portion (triples) that match the query• Identify triples whose graph is no-longer valid
Representing Time-validity
Naïve implementation
• no focus on efficiency• constant robot speed• no obstacles
Strategy• a best-first search in a tree of states and transitions• states: state of the KB (quads) and the location of the
agent• transitions: operators creating changes between states
Objective Find the best path to a state where all the quads are time-valid
Planning
Plan evaluation
Cost of a transition between states • min. time-to-invalidity of the least fresh quad in the
generated state• aim: maximise the time the quads will be valid
Optimality• Best-first search guarantees optimality• valid plan : information will be valid for the longest
time • no valid plan : collecting takes longer than expiring
Planning
Aim Approach feasibility
Scenario
Querying the KMi knowledge baseMoving a simulated robot and update the KB at query time
Settings• [map] SLAM/gmapping• [platform] : iRobot Create 2 • [sensors] : Kinect, humidity/temperature sensors• [KB] RDF graph representing KMi• [rules] manually defined
Experiments
Queries1. Which activity area has the best wifi signal?
simple scenario asking for 4 wifi signals
2. Which is the temperature of Room 22 and Room 20? plan designed to favor the validity in time
3. Which meeting room is the most comfortable?impossible plan, it would take too long (no plan)
4. Which meeting room is the most comfortable?same query simplified (plan)
Experiments
Query and update of knowledge bases (KB) containing statements that lose validity with time using a mobile sensor on demand (at query time)
Semantic Web technologies as KR framework• RDF model to represent the time-validity (time-
stamped annotations)• SPARQL query language to assess the validity of
new information based on a set of time-validity rules.
Conclusions
Future work
Realistic large-scale application (e.g. smart-cities)Planner optimization• include variable speed• time-constraints (e.g. time to answer the query)
Multiple coordinating robots scenarios
More complex time-validity rules
Conclusions
Thank you very much
[email protected]@open.ac.uk
Ilaria Tiddi Enrico Daga Emanuele Bastianelli Mathieu d’Aquin
Advantages
1. RDF paradigm • Schema-less database• More flexible • no specific data model (vs. temporal databases) • simpler therefore more reusable
2. SPARQL• detect the relevant portion of data to answer a query.• simplification of the process implementation
3. Autonomous Mobile Agents • no unnecessary data-flow management• no unnecessary sensor deployment.
Proposed Approach