In this paper, we introduce a technique for adaptive presentation of itineraries in navigation systems based on semantic models. We enrich waypoints with semantic information and display only those waypoints to the driver that he is really interested in, hiding information that will most probably be distracting.
Present solutions aim at optimizing routes without taking the driver’s personal knowledge, experience, and preferences into account and, thus, are not personalized. However, incorporation of personal information could improve presentation of routes significantly. On the one hand, instructions should be rather short and abstract, if the user knows the particular
In this paper, we introduce a concept to enhance the presentation of the route by adapting it to the driver and his preferences and experience. For that purpose, we use semantically enriched models of the itineraries. In the end, the user should only see and hear necessary and helpful information instead of every single detail. Besides automated adaptation, the user has always the option to adjust the level of detail of the presentation manually.
Even if not focused on the particular problem depicted in the introduction, research has been conducted, in order to enhance presentation of itineraries.
A generalization technique that is geared to hand-drawn route
descriptions and tries to solve the visibility problem of minor
parts of an itinerary on a constant scale factor, is presented
by Agrawala and Stolte [
In addition, humans are mainly interested in information about
the main waypoints and not the connections between them.
They rather neglect the length of individual roads and instead
raise their visibility or specific route characteristics (e.g. a
big building or a roundabout) that they consider to be
relevant to the navigation process [
In [
The authors of that paper also point out, that if
environmental features are taken into account for structuring route
knowledge, a coarser perspective on the required way-finding
action than simple turn-by-turn directions can be provided.
Variable granularity in route directions is also focused in
[
Most users have at least some knowledge of the vicinity they live. Although most people are familiar with their hometown or parts of it, they receive detailed route instructions from their device. A personalized granularity in route directions regarding the special knowledge of a user about the routing environment could lead to a more intelligent navigation system.
Such a comprehension of the user’s knowledge about several parts of the route has been largely neglected by device manufacturers and suppliers of relevant web services, so far. The fact that such navigators would need an extended learning phase to provide customized assistance, is mostly seen as a major drawback.
To address this problem, Richter and Tomko present an
approach to generate adaptive route directions generated through
a dialog-based knowledge recognition process [
We argue that a system could automatically provide user specific route directions based on a learning process that primarily is supported by a dialog-driven approach. Therefore, we act on the dialogue suggestion by Richter and Tomko, which in a first step can enhance the learning process to solve the cold start problem of completely unknown user preferences and also avoids the user from unnecessary interactions while driving.
In order to personalize the route descriptions, we need a detailed and machine-readable model of the route in order to adapt it to the user’s knowledge and preferences. Thus, we have to encode all information that may be helpful to decide whether a particular part of the route should be displayed in detail, only briefly, or not at all.
Itineraries usually are described by a set of waypoints, which represent positions between origin and target location. The idea is now to semantically enhance the waypoints in order to use the semantic information for filtering.
We therefore propose a layer model where each layer represents a degree of granularity in the route presentation. The lowest layer contains the default route directions including all details of the itinerary, as known from conventional systems. All upcoming layers show, depending on the level of abstraction, only certain parts of the route and provide the related routing instructions.
To achieve this goal, we transform the route description into a semantic model, which allows us to characterize each waypoint on the basis of its properties comprehensively. In addition to the general information of an itinerary, such as location coordinates, street name and driving instructions, a semantic description includes further information, such as a classification of each route point on the nature and type of geographical conditions. This means that a place can either be characterized as town, city, region or even a country and the connection between two places as a street, road or motorway. This hierarchical distinction enables later filtering to distinguish the different levels of abstraction. For the transformation every route object provided by online web services like Google Maps1 can be used.
If a user has sufficient knowledge about the environment in a particular area, only a few instructions, limited to the issue of the next motorway link and the direction of the nearest town, may be appropriate, while in areas less or not at all familiar, a detailed route guidance without any abstraction will be a better choice.
For enriching semantic itinerary models with further
information, geo-services such as LinkedGeoData.org2 can be
used. Those services provide comprehensive background
knowledge related to spatial features of the ways, structures
and landscapes around the waypoints of an itinerary [
instruction latitude lbel a
type longitude Waypoint e d u litt a 36 m / 118 ft residential locatedIn hasSuccessor
City l e b a l Duisburg "Keep right at the.."
instruction locatedIn lbel a altitude Waypoint latitude type longitude
residential 39 m / 128 ft 51.416 6.790 51.409 6.799
An itinerary that has been semantically enriched in that way, finally, facilitates the applications of particular “views” on the route. This mechanism can be used to show or hide certain waypoints and create an optimal presentation based on the users’ preferences and experiences.
The navigation system could, for instance, only display prominent waypoints such as freeways (if the user already has basic knowledge of the area). In this case, the directive could simply be “Head for Freeway 1”, whereas other users would receive a set of detailed instructions leading the driver to the particular freeway. 1http://maps.google.com/ 2http://linkedgeodata.org/ 3http://wiki.openstreetmap.org/ 4http://www.geonames.org/ 5http://www.topocoding.com/ The use of semantic models has different advantages compared to traditional ways routes are displayed in navigation systems:
Standardization: As information comes from various sources, each with their own formats and specifications, we need a standard to cover all these information. Semantic models are flexible enough to import all information provided by the original sources and make them accessible in a unified way (e.g. via SPARQL).
Extensibility: The characteristics of semantic models mentioned in the last paragraph allow integration of new information sources as well, regardless of their format. Ease of data processing: If the models are encoded in a standardized language like RDF or OWL, they can be queried using the SPARQL Protocol and RDF Query Language (SPARQL).
Additional services: The use of standardized semantic
models lays ground for future services apart from classical
navigation. Recommender systems that incorporate
semantic data [
As an intermediate step towards a personalized presentation, we create a layered model based on the semantic route, so that the distinct layers reflect a particular level-of-detail. The bottom layer contains all waypoints, whereas the levelof-detail decreases on each layer (see Figure 2). SPARQL queries can be used as a filtering technique in order to show or hide certain waypoints for each layer.
The layered model now allows us to switch between the levels-of-detail, such as zooming in or zooming out details of the route presentation. We provide means of manually and automatically switching between the degree of detail as well as choosing the granularity based on user profiles.
A simple way of adjusting the presentation could be by
interacting with the driver. Initially the user should be able to
convey known regions dialogue based at the beginning of the
guiding process, where the route has been calculated.
Therefore, he can check the known parts of the itinerary step by
step. Such a procedure is necessary on each guidance where
no part has been marked as well known, yet. This approach
is similar to the dialog-driven process described by Richter
et. al. [
The significant deviation in our approach is that we use the dialogue initially to customize the whole route guidance on the users individual needs, while Richter provides abstract instructions by default and requires user interactions at any time the user needs more detailed ones. Nevertheless, that kind of interaction facility we will provide additionally. The user can use a simple widget such as a slider or a turning knob, which he can set up or adjust the level of detail manually.
This functionality is available at each stage of the guidance process to allow the user to react appropriately in any situation depending on his individual perception. The opportunity to interact with the systeWmDeentanatiSlsieanazlleeigaeunfdmeömchBteildns,cvheirrwmesnidcehntbcSaerieesnthe any time also enhan satisfaction, thus, the acceptancdeen Loinfk Dthrueckeanunteobemn daetreKdartep.rocess can be improved. Figure 3 shows an example of such an interaction widget. The user interface provides two buttons for changing the level of detail. If the user pushes the “More” button he receives more details of the itinerary presented on the screen and as driving instructions. A push on the “Less” button on the other hand causes a higher level of abstraction.
Each layer can be seen as a “view” on the itinerary showing or hiding certain details. The base-layer corresponds to the way traditional navigators would display a route; it simply contains every single waypoint. If a higher level of abstraction is selected (either automatically or by hand), the navigator hides certain waypoints and only displays more prominent ones. If the adaptation process is supposed to be automatically instead, the level of detail can be adjusted rulebased or by other means.
Take the A3 motorway
For more sophisticated adaptation effects, dedicated user profiles cam be maintained to keep track of the user’s knowledge and preferences. The system keeps track of all places and routes the user has marked as well known. It then can provide recommendation for the levels of detail on new calculated itineraries. In this way, a user knowledge model evolves from the users interaction in a step by step manner. Figure 4 shows a schematic example of a map representing the users area knowledge, where the dark regions are assumed as well known and the lighter ones as unfamiliar. well known area partially known area unknown area
In order to automatically switch between the levels of detail, knowledge about the user is necessary. On the one hand, the system could incorporate information explicitly entered by the user or, alternatively, keep track of his itineraries, in order to “learn” such a profile. The first case requires user interaction, for instance by tagging certain areas on a map as “well-known” or selecting them from a list of areas. If the system should learn and update the profile automatically based on the driver’s routes, it has to keep track of the waypoints on these routes and autonomously mark them as “rather known” or “well-known”. The level of detail then is based on the supposed degree of familiarity with the particular route section.
Combining explicit profile information with learning, of course, is an opportunity as well.
In this paper we presented an approach for enriching the waypoints of itineraries with semantic information, enabling a route guiding system to provide an adaptive user interface. If a driver already knows parts of the itinerary very well, the system presents the route by adapting it to the drivers preferences and experience.
During the workshop, we would like to discuss, among other issues, the following questions: What would be better? Tagging a route object with semantic information vs. converting the whole route as a semantic model? What could be alternatives for interactive definition of already known waypoints (e.g. marking them at the route planning process). How could a learning system look like that recognizes frequently used route parts or repeatedly visited places?