This demo paper presents a travel advice system based on information extracted from social media and linked data.
The Twitter1 microblogging platform is widely used in the public transport
domain by passengers to communicate with transport operators and by
operators to provide customer service and passenger information [
This demo will show the TravelBot system developed to support bus users in the city of Aberdeen, UK. The demo will feature: a user registering a journey with the system; TravelBot monitoring Twitter for messages describing transport related events that may disrupt that journey; and when one is detected, sending a personalised message to the user warning them of the potential disruption2. The demo will utilise the datasets and system shown in Fig. 1. ? The research described here is supported by the award made by the RCUK Digital Economy programme to the dot.rural Digital Economy Hub; award reference: EP/G066051/1. The authors would also like to acknowledge the support of First Aberdeen in developing this work. 1 http://twitter.com/ 2 A video of this demo is available at https://youtu.be/ZAg6RnCQoUI.
The TravelBot system3 is supported by a linked transport information ecosystem (illustrated in Fig. 1 and further discussed below) that is based on a series of ontologies. Services provide the system functionalities by reasoning with data accessed via SPARQL endpoints. Figure 2 presents a sample of the data generated for a Tweet, a user journey, and an alert message sent to a user. User Interfaces Web Services
TMI Datasets
User Journey
Registration Ontotext
KIM
Tweet Processing
Annotation Triplifiaction
Event Inference
NextBus Interface
User Alerting EveMnat-tcJhoiunrgney Micro-NLG Twitter Data
Public Transport
Schedules
NaPTAN NPTG
Transport Infrastructure
Annotations
Journeys Transport Events Ontologies
Bottari FOAF SIOC Transit NaPTAN LinkedGeoData
Open Annotation
Transport Disruption
PROV-O Journey
Tweets published by accounts known to provide travel information for the geographic area, including bus operators, transport authorities and registered users, along with Direct (private) Messages to the TravelBot Twitter account are received and stored by the Twitter Monitoring Infrastructure4 (TMI). The message and associated metadata (including its unique identi er, author, and creation timestamp) are stored in the Twitter Data dataset and published as linked data using the Bottari5, FOAF6, and SIOC7 ontologies8.
Once stored, the message's URI is passed to the Tweet Processing component, which extracts, classi es, and contextualises a semantic representation of any transport event(s) described in the message. The Ontotext KIM9 platform, which is designed to identify semantic entities in text, is con gured to discover entities related to transport events in the message. To achieve this, the KIM knowledge base has been extended with RDF descriptions, including names, 3 The TravelBot system is available at https://github.com/SocialJourneys. 4 The TMI system is available at https://github.com/SocialJourneys/TMI. 5 http://purl.org/NET/bottari.n3 6 http://xmlns.com/foaf/0.1/ 7 http://rdfs.org/sioc/ns# 8 To comply with Twitter's terms and conditions, this data is only available within this system. 9 http://ontotext.com/kim commonly used abbreviations and slang terms describing: types of potentially disruptive transport events related to network operator actions, public transport, and tra c described by the Transport Disruption ontology10; open bus service and schedule data11 stored in the Public Transport Schedules dataset12; public transport access points from the NaPTAN dataset13 and settlements from the NPTG dataset14; and the road network extracted from openstreetmap.org and stored in the Transport Infrastructure dataset15.
The Annotation Tripli cation component generates annotations for each entity identi ed by KIM, represented with the Open Annotation ontology16. As shown in Fig. 2, each annotation links to the source message and the identied resource; these are stored in the Annotations dataset. The Event Inference module uses a set of hand crafted SPIN17 rules to create a semantic representation of transport events based on these annotations and represented using the Transport Disruption ontology. The rules attempt to determine if a transport related event is described, and if so, to associate a geolocation and time period with it, attribute values speci c to the event type (for example, in Fig. 2 the carriageway a ected by the roadworks) and link to any bus services that may be a ected. The inferred event(s) is (are) added to the Transport Events dataset, along with provenance information18 recording the creation timestamp and the message that it was derived from.
TravelBot users register any journey for which they wish to receive noti cations. Each journey is described in terms of the week days on which it is made, the time of travel, bus service(s) used, and boarding and alighting locations. At a user speci ed time before each journey, the NextBus Interface component retrieves the upcoming arrival times for bus(es) on the speci ed service(s) at the boarding bus stop19. This is privately communicated to the user via Twitter as a Direct Message and will be available through their usual Twitter client.
The Event Journey Matching component uses a series of quality metrics to determine if any events might disrupt a user's journey. The metrics include: temporal relevance, which considers if the event is ongoing during the user's journey; geospatial relevance, which considers if the event's location and user's 10 http://purl.org/td/transportdisruption# 11 Published at http://www.travelinedata.org.uk/traveline-open-data/ traveline-national-dataset/. 12 Represented using the Transit ontology - http://vocab.org/transit/terms/. 13 http://data.gov.uk/dataset/naptan, represented using the NaPTAN ontology http://transport.data.gov.uk/def/naptan/. 14 http://data.gov.uk/dataset/nptg, also represented with the NaPTAN ontology. 15 Represented using the LinkedGeoData ontology - http://linkedgeodata.org/ ontology/. 16 http://www.w3.org/ns/oa# 17 http://spinrdf.org/ 18 Represented using PROV-O - http://www.w3.org/ns/prov#. 19 This uses the NextBus API (http://www.travelinedata.org.uk/ traveline-open-data/nextbuses-api/), which provides real-time arrival information for all bus stops in the UK.
oa:hasTarget oa:hasTarget :KingStreet a lgd:Road rdfs:label "King Street" wkb:TimeInterval_Today a protont:TimeInterval rdfs:label "today" expected route of travel overlap, with more precise event locations (e.g. a road the bus travels along) being assigned a higher relevance than less precise locations (e.g. a locality the bus travels through); service relevance, which considers if the event is known to a ect a bus service that the user will travel on; and veracity, which is based on if the Tweet's author is included in a prede ned set of trusted users (e.g. the bus operator, transport authorities, or a local radio station)20.
If an event is rated with su ciently high temporal relevance and either geospatial or service relevance, then a tailored message is generated by the MicroNatural Language Generation (Micro-NLG) component, and sent to the user. The message is based on the inferred event resource and attempts to convey the level of certainty that the journey will be a ected, as indicated by the quality metrics. For example, in Fig. 2 although the roadworks are on a road that the bus travels on, no delays have been reported by the operator so the message is deliberately vague using the phrase \may be a ected"; if the operator does report delays on Service 1 in that area, this would change to a stronger phrase such as \is highly likely to be delayed".
A user study is planned to evaluate the TravelBot user experience and system performance with inferring event descriptions from social media posts.