Chris Baillie

Data Quality in the Web of Linked Sensor Data

Research

Brief Biography

I am a research student working within the dot.rural Digital Economy Hub at the University of Aberdeen. I graduated from the University of Aberdeen in 2008 with a BSc(Hons) degree in Computing Science.

Research Focus

The Web has evolved from a collection of hyperlinked documents to a vast eco-system of interconnected documents, services, devices and even people. However, the inherent open nature of the Web allows any 'thing' to publish data and as such information quality is a major issue.

I am exploring how semantic web technologies can be used to facilitate Information Quality (IQ) assessment in the Web of Linked Sensor Data. Sensor observations can be annotated with metadata describing the characteristics (e.g. accuracy, latency and frequency) of the sensor that created them. By defining a set of rules and applying them to the sensor's characteristics we can reason about the quality of the sensor's observations. For instance, observations generated more than 15 minutes ago could be considered not timely; comparing values against a trusted source can be used to assess accuracy. I am also exploring how sensor data provenance can be utilised in the quality assessment process and how the results of these processes can enrich the provenance graph in order to facilitate future quality assessments.

Sensor Network Testbed

I have developed a basic sensor network testbed using a number of hardware nodes implemented using the Arduino electronics prototyping platform. These nodes feature sensors that observe a number of phenomena such as temperature, vibration, motion and light. Each node is connected to a LAN using either Ethernet or Wireless connectivity and has its own URI so that it can be accessed as a Web resource.

Observations generated by the sensors are transmitted to a data server implemented using Apache Tomcat. This server provides a number of servlets for both incoming and outgoing data. When the server receives an observation it is stored in a MySQL database to allow easy access to the raw data. This database is wrapped with d2r server, which provides a SPARQL end-point to data stored in a relational database. Queries passed to this end-point are converted from SPARQL into SQL and the results are annotated with RDF according to d2r's mapping file.

Our SPARQL end-point is available at http://dtp-126.sncs.abdn.ac.uk:8081/sparql.

A web-based SPARQL explorer is available at http://dtp-126.sncs.abdn.ac.uk:8081/snorql.

Visualisation

I have also developed a set of visualisation tools for sensor data observations. These allow a user to select the sensor and enter a time window over which they wish to view observations. These parameters are used to query the MySQL database, and results are plotted on a 2D chart using flot.

These charts are fully interactive. By clicking on a data point the quality assessment process is triggered for that particular observation. The results of the quality assessment are then presented alongside the Linked Data describing the observation and sensor as illustrated by the screenshot on the right.

The visualisation web service is available at http://dtp-126.sncs.abdn.ac.uk:8080/SensorNet/graph.jsp.

Reasoning

At present our quality assessment process is hard-coded in Java. One of my future plans is to implement quality assessment using a reasoner. To date, I have experimented using SPIN rules in TopBraid Composer. These rules are used to assess accuracy and timeliness as follows:

• If the observation is with ±2°C of Met Office data, it is deemed accurate.
• The observation is deemed timely if it was generated within the past 15 minutes.

Future

We have identified a number of research questions to guide our future work:

1. Is it possible to reason about quality in the Web of Linked Sensor Data?
2. 1. Can quality indicators be represented in an ontology?
   2. Can quality indicators be assessed using an appropriate reasoning mechanism?
3. Is it possible to capture the provenance of sensor data?
4. What other elements of provenance can be described, i.e. policies or intent?
5. Is it possible to assess quality using provenance?
6. Can the results of quality assessment be represented in the Web of Linked Sensor Data?
7. Can the provenance of quality assessment outcomes be used to facilitate future quality assessment?