We tried an experiment in ISWC 2016, Kobe, as a practical example of data integration of sensor data through IoT devices and semantic information. We provided small GPS devices to some volunteers from participants of ISWC 2016 and collected trajectory data during the conference. Then, we integrated the data with Points of Interest (POI) information collected through existing open data such as open government data by Kobe city, DBpedia and Wikidata. This paper presents the method to integrate GPS trajectory data and existing open data with consideration on usefulness of them as sources of POI information for practical analysis.
Integration of sensor data through IoT devices and semantic information is an important technology in various fields. We tried an experiment in ISWC 2016, Kobe, as a practical example of such a semantic data integration. In the examination, we provided small GPS devices to some volunteers from participants of ISWC 2016 and collected trajectory data during the conference. Then, we integrated the data with Points of Interest (POI) information collected through some existing open data. As the result, we published the integrated trajectory data as Linked Open Data with SPARQL endpoint. It enables us to analysis trajectory information from various aspects and granularities.
This paper presents the method to integrate GPS trajectory data and existing open data from various aspects with multiple granularities. Then, we consider how much useful are existing open data to analysis real GPS trajectory.
There are some methods for collecting trajectory data such as Global Positioning System (GPS) [1], radio frequency identifier (RDFID), location recognitions from image [2], Monitoring of WiFi for smartphones [3] etc. In our experiment, we used GPS because it does not need any special equipment for target area, and it can measure location more accurately than WiFi monitoring and recognitions from image.
We conducted an experiment for collecting GPS trajectory in International Semantic Web Conference 2016 (ISWC2016), Kobe, Japan, October 17-21, 2016. The purpose of this experiment is to demonstrate of the possibility how we can combine GPS trajectory data and existing open data.
We ask some volunteers to receive and bring a small GPS device during the stay in the ISWC2016 conference. Their movements are recorded on the GPS devices. We used i-gotU GT-600 which are GPS devices offered commercially for the experiment. The devises recorded their location per 1 minute. The collected information includes its position (latitude, longitude), the height above sea level, movement speed and moving distance. As the result, we collected trajectory data from 11 volunteers during the ISWC2016. The average of recoded period was 91.8 hours (3.83 days). 3
The collected GPS trajectory data is converted into RDF using POI information obtained through existing open data. We collected POI information in Kobe city since moving range of the collected all GPS trajectory data, which is shown in Section 2, was limited in Kobe city. As the result, we collected six datasets. They are classified into two kinds as follows.
We collected four datasets related to sightseeing from open data portal1 by Kobe city
because our experiments aim to analyze interests of participants for the international
conference. They are dataset about (
We extracted data which has position information (latitude and longitude) in Kobe city using SPARQL query.
We supposed that the former contain POIs closely related to the local area because
they are provided by the local government. On the other hands, we supposed that the
latter cover wide broad kinds of POIs because they are general open data which
everyone can edit on the web. In addition to them, we prepared two kinds of merged datasets;
merged dataset of open data by Kobe city (
Table 1 shows POI information which we collected. The merged dataset of (
Choose one dataset from multiple datasets of POIs according to the purpose of the analysis
POI information (common to all users) Lat/Long label1 Lat/Long label5 Lat/Long label24 POI‐1 POI‐5 POI‐24 poi poi poi user1 stayed next user1 stayed next user1 stayed next at POI‐1 at POI‐5 at POI‐24 ・・・ Year/Day/Time Year/Day/Time Year/Day/Time
GPS trajectory (for each user)
Fig. 1 shows the overview of RDF data model for GPS trajectory. This model consists two data models for POI information and trajectory of each user.
POI information is represented by two classes POI class and MergedPOI class. POI class represents a primitive POI extracted original six datasets, and MergedPOI class represents a merged POI discussed in Section 3.1. Each POI has properties such as rdfs:label, position information (geo:lat, geo:long), data source. Links to other LOD are represented using rdfs:seeAlso property. In addition to these properties, a merged POI has contains properties which represent primitive POI merged into it. One of these POI datasets is chosen and commonly referred by GPS trajectory data according to the purpose of the analysis.
The GPS trajectory is represented by a series of RDF resources of StayPOI class. Each StayPOI resource represents information about a stay that a user stayed a POI during a time interval. It is described using properties such as the username, POI, starting/ending time point of the stay, a link to its next stay information. The POI is a reference to an instance of POI class discussed the above. Based on the data model, a GPS trajectory is represented as a directed graph which looks like the bottom of Fig. 1.
We translated GPS trajectory data discussed in Section 2 into RDF based on data models shown in the Section 3.2. Position information are compared with each POI information in a selected POI dataset and judged which POI the user stayed. The following steps show how we judge each staying. 1. Each position data in GPS data and each POI information in the selected POI datasets are compared. If the distance between the position in GPS and the POI is less than the threshold d, and the distance is minimum among the position in GPS between each POI, then it is considered that the user stayed the POI. 2. When a series of GPS data records are judged that the user stayed the same POI, - The earliest date-time from these records is considered as the time which the user entered (started to stay) in the POI - The latest date-time from these records is considered as the time which the user leaved (stop to stay) in the POI 3. On the series of GPS data records, when the POI the user stayed changes, the new
POI is considered as the new POI that the user stayed in the next.
We translated GPS trajectory data collected from the eleven users using eight kinds
of POI datasets. We used threshold d as 100m. In the case that we used the merged
dataset (
Table 2 shows how many POI
obtained existing open datasets and Data # of # of POI # of GPS data
used to transform GPS trajectory sIDet Dataset name dPaOtaIsient sosmtaeyuedser jwudhgicehd citasnstbaey
into RDF. It shows that 68.58% of P1 (
This work was supported by JSPS KAKENHI Grant Number 17H01789. The authors are deeply grateful to volunteers who join the experiments in ISWC2016.