This paper describes MyWorkPlace, which uses personalisation of automatically generated ontologies to provide users with personalised information about new and invisible items within a ubiquitous computing environment. A key component of this system is the automatic generation of the ontologies, and models used to drive it. This data is being gathered from a number of sources including: building maps, the build manual, sta directory, student timetables, the departmental calendar and room bookings. We describe planned evaluation of the system in a deployment to a new building and its new inhabitants.
Ubiquitous computing aims to embed our everyday
environment with devices, sensors and services in such a way
that they are as unobtrusive as possible, to the point of
becoming invisible to common awareness [
To motivate this paper we consider a real life example of the invisibility problem, a University Department moving to a new building instrumented with a number of ubiquitous computing features. We examine the interactions and information needs of Fred, an academic.
Initially all facilities (ubiquitous computing and otherwise) within the building are unknown to Fred. He may have an idea of some types of facilities which are available, but is unlikely to know the full extent of the facilities available and details about them. A list of all the information about the building, its sensors, services, devices and occuPermission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. ubi-PCMM at UbiComp’06, September 2006, Orange County, CA, USA. . pants would have hundreds or thousands of items and clearly overwhelm the user.
MyWorkPlace solves this problem by modelling users, places, devices, sensors, services and objects to provide personalised views of the items within a user’s environment. An important part of the system is the use of an automatically generated ontology to assist with the selection of items to display for the user.
The ability for ontologies to facilitate human-machine and
machines-machine communications has gained wide
recognition in the development of the UbiComp. It has been used
in middleware to facilitate context management and
reasoning [
A novel aspect of our work is the use of ontological data generated using dierent sources which have dierent levels of reliability to personalise the information given to a user based on their context.
The eort in creating a comprehensive ontology is substantial. Partial or completely automated generation has the possibility to greatly reduce this eort. Depending on the degree and type of automation, the reliability of the ontology can vary greatly. To cope with this, we are examining multiple levels of ontologies.
Our automatically generated ontology in being built from a number of sources including: building maps, the build manual, sta directory, student timetables, the departmental calendars and room bookings.
The rest of this paper is organised as follows: We rst describe some related work in Section 2 to set the scene. Section 3 describes MyWorkPlace when used in the scenario described of Fred. The methods we use for automatically generating the ontologies are described in section 4. We conclude with a discussion of our proposed evaluation and future work in section 5. 2.
Weiser predicted that ubiquitous computing would
become a technology which disappeared and became
invisible [
There has been some work which addresses the issue of
informing users of ubiquitous computing systems what devices
and services are available to them. The AFAIK system is
a multimodal help system for an intelligent room [
The CONON system [
Outside of Ubiquitous Computing there has been work on
extracting ontologies from existing text sources. To do this
both concepts and relationships between them need to be
learned. ConceptNet [
The MyWorkPlace system provides users with
personalised views of places, sensors, devices, services, objects and
people in their environment. This advances on our earlier
work, MyPlace, described in [
We now return to the scenario from the Introduction for interactions of Fred, a member of sta interacting with MyWorkPlace shortly after moving into the newly constructed School of IT smart building.
Fred is an academic who does not know very much about the building. He knows the location of his own oce, and those of the students he supervises. He knows approximately where the the front counter is, but has not been there so is unaware of what facilities are available. He knows nothing of the seminar room, sta common room, pervasive computing laboratory, or undergraduate computer laboratories.
Figure 1 shows a screenshot from MyWorkPlace personalised for Fred, as it would be shown on a PDA while he is standing in the Foyer of the building.
The Status bar at the top tells Fred what the system believes his location and status is. In this case, his location is believed to be the Foyer, because the Mac address of his Bluetooth mobile phone has been detected there. There is a details button to allow him to scrutinise and correct the reasoning used for his location and status.
The content panel of the main screen consists of ve expandable headings. The headings are Devices at this location, Nearby Devices, Nearby Places, Services /Events, and People. Clicking a heading shows or hides the contents. The [Show all items] button displays all the items the user is allowed to use. It also allows the user to see why an item was included or excluded by MyWorkPlace.
The system must determine which of the myriad of devices, sensors, places, services, events to display to Fred. For each heading, it examines the evidence and places each item into one of a number of relevance categories : • Already knows - The user is believed to already know about this, based on either user feedback, or observations such as the use of a device, or being detected in a location. • Needs to learn - The information is thought to be useful to the user and the user is believed not to already know it. Whether information is useful to a user is determined based on manually entered stereotypes, or the generated ontologies. • Needs to know now - This is a special case of Needs to learn where the item is believed to be important based some aspect of the users’ current context. For example: If the user has stated they are on their way to a seminar in a room they are believed not to know the location of, then location of the the seminar room is very important. • Not relevant (Neutral) - The information is about something for which there is no information suggesting that it is useful to the user. • Doesn’t want to know - The user has indicated that they do not wish to be informed about this, or a very similar item.
The main screen shows items in the Needs to learn and Needs to learn now categories. The user can override this personalised selection of information to see all items and their relevance categories by choosing the [Show all items] button.
The Nearby Places category in Figure 2 lists a number of the places which Fred does not know about. There are many others which he is not informed about as the system does not believe they are relevant to him. For example with suitable information in the system in the system it might be able to omit details of the Undergraduate Laboratories, as it is semester break so he is not currently teaching classes.
Figure 2 shows the view when Fred returns to his oce. Here are more devices he may wish to learn about. Clicking on an items brings up more information about the object. It includes usage instructions and troubleshooting information. In addition to this, it includes links to related items, as suggested from the automatically generated ontologies. Each time Fred clicks a link requesting more information, a piece of evidence is added to his user model suggesting he knows about it.
It is important for the user to be able nd out why a certain item has been displayed to them, and others have not. We call this scrutability. When a user clicks the [Show all items] button the full list of items is colour coded according to which of the ve relevance categories dened above it belongs. An explanation of the reasoning used to categorise each item is shown if the user hovers the mouse over it. An example explanation might be You are teaching Algorithms 101, this room is used for Algorithms 101, and you have not yet been detected there. 4.
The data in our ontology is being built from a number
of sources, such as building plans, sta directory, the
building manual, student timetables, the departmental calendar,
room bookings, and a relatively small, handcrafted base
ontology. The degree of automation used and level of user
input required in generating ontological information from
these sources varies considerably. The reliability of each
source also varies for a number of reasons, such as input
errors and frequency of maintenance. MyWorkPlace takes
account for the inaccuracy problem with its evidence
accretion and delayed resolution approach [
Our initial source for location relationships were the building plans for each oor of the building. Features on the plans are grouped in the relevant layers. For example all the room number labels are in one layer, the room description texts are in another, another layer holds all the doors, one layer holds all the solid walls while another holds all the glass. There are over 100 dierent layers in total.
Analysis of the lines and text data in each layer allows us to use relatively simple reasoning to determine rooms and relationships between them (e.g. distance). Figure 3 shows a section of the plans (left-hand side sub-gure), and further extracts showing: a) only wall and door layers (top righthand sub-gure), and b) only room number and label layers (bottom right-hand sub-gure). The majority of the data generated from these plans is assumed to be very reliable as the building was built according to them.
The departmental sta directory is also an important source for automatic population of an ontology and user models. The sta directory yields a list of all academics, administrative sta and postgraduate students. It also gives relationships between people and their research groups (e.g. undergraduate coordinator, chair of a research group), oce or workspace locations, and contact information. A number of the student-supervisor links are also available in this database. This data can be extracted with minimal human interaction. However, it suers from some inaccuracies as the phone list is not always kept completely up to date, for example, when people move oces or change research groups.
Another source of ontological data is the building
manual. This documents various elements about the building
in a glossary and human readable format. Example entries
in the manual are an entry for fax that describes policies
for using a fax machine, and an entry for reception that
gives the opening hours and the location of the reception
desk. We believe we will be able to use Sago, a
descendant of Mecureo [
We propose to use other sources, such as: the university timetable that lists times and venues for every class in the university; the departmental calendar and room bookings that give scheduled times and locations for certain activities. Both sources may help generate more evidence that assists the MyWorkPlace system to deduce, say, current location and activity for a person, and suitable activities for certain venues. In future we may link databases of particular class enrollments to generate views of MyWorkPlace for undergraduate students. However, this information might not be as reliable as some of the other sources.
The nal source is a small, handcrafted base ontology. This serves as a base for other sources to build on. 5.
At the time of writing the new building for which the system has been implemented has only just been occupied. We plan to evaluate this system by two methods: rstly we will ask a number of users to perform a task requiring knowledge of the building, and compare the performance of a group of people given access to MyWorkPlace against those who are only given access to the data sources used to populate the ontology (probably building plans, building manual, and phone list).
The other form of evaluation we wish to perform is to make MyWorkPlace available to volunteer academic sta and postgraduate students, and record their use of the system. In addition to this we will request feedback from a sample of users.
One aspect we are particularly interested in is how any inaccuracies in the generated ontological data aect the user experience. As mentioned previously some aspects of this data are known to be very accurate, while others are expected to contain some out of date or incorrect items.
Another piece of future work is to investigate the use of a suggested template and categories for entries in the building manual to better facilitate information extraction. We would also like to include data extracted from existing, maturer ontologies.