Yellow pages catalogs and corresponding directory services help people to find companies providing services and selling products. We propose that semantic web technologies provides new possibilities for searching and browsing the Yellow page advertisements. As a first step towards this goal, an experimental implementation is presented.
semantic web, ontologies, directory services
Yellow pages (YP) is a widely used and commercially successful service model for matching the need of a customer with the corresponding products and services offered by companies. In this model, the YP service provider maintains a list or a hierarchy of product and service categories, such as ''Electronic equipment'' or ''Car Rental''. In spite of the versatility of possibilities, it can still be difficult to the end-user to map a need on the services offered. On the other hand, for the catalog advertiser, it may be difficult to index the service in such a way that the prospects would not miss the service.
The Helsinki area yellow pages (HYP) (2002) is a typical example of a printed yellow page catalog. HYP is based on an alphabetical category list of 1085 products and services. From the user's viewpoint, a problem of using the catalog is that a user's need may be related to several advertisement categories. E.g., if one needs help for a headache, the possibly relevant categories in HYP include ''pharmacy shops'', ''doctors'', ''medical centers'', etc. On the other hand, a category may be related to several needs. E.g., ''building equipment'' is potentially relevant in many building needs and problems. Furthermore, several categories are partly overlapping, such as ''barber's shops'' and ''hair dressers''. In the bilingual HYP, there is also the additional problem that many, but not all, categories have both Finnish and Swedish categories.
From the advertiser's viewpoint, the advertisement should be found in as many reasonable categories as possible for a better coverage. This is not always feasible to the company (e.g., due to the costs of advertising) and hence many useful categories in the list in HYP are practically empty. This is not what the user expects from a good service. E.g., the category ''garden furniture'' in HYP contains only one advertisement although many furniture shops sell also garden furniture.
Online directory services* can benefit the user in providing better searchability of contents (e.g., keyword-based search and simple hierarcial navigation of the categories, such as in yahoo), new information content (e.g., hyperlinks to up-to-date business web sites), and interactive features (e.g., online enquiry forms and feedback mechanisms). Despite of the advances of publishing the YP online, the current online implementations still suffer from the problem of finding the relevant categories to the customer's problem and finding categories that are potentially relevant to several problems.
The work presented in this article is related to research and standardization work on service description and matching on the web, such as [5] where DAML-S is proposed as an ontology for annotating service capabilities. However, the main focus there is on modeling service capabilities in terms of the input and output of a web service for a software agent. In our work [4], the problem of matching human user needs with advertisements is considered.
Our work is also related to the OntoSeek system [8], where a linguistic ontology (WordNet) is used for improving the search capabilities of a directory service. In our work, the problem of modelling the advertisement information in such way, that the rich semantical properties and connections between services can be used in retrieving related advertisements to the user. The linguistic ontologies (e.g. WordNet and EuroWordNet) may be useful in solving this problem. However, currently such linguistic ontologies does not yet exist in Finnish.
From the viewpoint of the user's need, indexing a company's advertisements according to a business category is not very useful unless the category unambiguously implies the services offered to the user. A service cannot be characterized extensively by a simple category label, because services have internal semantical structure. For example, transportation services are characterized by properties such as the cargo (people, furniture, oil, etc.), the instrument used (car, train, ship, airplane, etc.), and the area of transportation (city, country, international, etc.). The internal properties of services can be described by using ontologies [1]. Services are often related to each other and make combined services. These relations can be described in terms of ontological structures.
In our view, YP online services should be based on ontologies, that define the terminology and concepts to be used by the advertiser and by the end-user**.
From the advertiser's viewpoint, each advertisement in the system is a metadata description of some service that the company offers. A single company may offer many services. It is not enough for a camera shop to put one advertisement under category ''camera shops'' but, for example, one advertisement under camera ''selling services'', another under ''repair services'', and a third one under ''rental services'' depending on the company at hand.
To the end-user, the service ontology provides new possibilities to finding advertised services. The ontology can be used as a multidimensional semantic navigation directory for browsing the advertisements in the spirit of Topic Maps [6]. E.g., services related to the concept ''Piano'', such as ''Piano shops'', ''Piano transports'', ''Piano tuners'' etc. can be found through the product class ''Piano'', because this concept is used in different roles in these services.
The ontology also provides a kind of semantic glue that relates the service advertisements of different service classes with each other. If you want buy a flight ticket to Helsinki, you are likely to be interested in booking hotel accommodation there as well. Such semantic relations between service classes can be defined in general terms within the ontology, and be used for generating recommended links for the user. E.g., if the user selects the category ''flight services'' with properties such as destination and time of travel, then the YP system could automatically show links to related accommodation services and events taking place in the destination at the given time.
As a first step towards semantic web based YP services, we have implemented a prototype called ''Keltsi''. The Keltsi system provides the end-user with ontology-based searching and semantic instance browsing.
Ontology-based searching means that the user can query the advertisement collection by using the classes and properties of the the underlying ontologies***.
In Keltsi, the query is created by traveling the ontology three from general to more specific service classes until the service class that best describes user's need is found. The class can be a very specific class, such as ''Restaurants'', or a more general class, such as ''Eating and drinking establishments'', which contains also bars, fast food services, etc. in addition to restaurants.
The user can set property constraints that the searched service instances must fulfill. The properties and their possible values depend on the selected class. Therefore, if the selected class is ''Restaurants'', only properties related to ''Restaurants'' and its superclasses are shown. The selection of the value for a property is either based on traversing through an ontology hierarchy, by selecting the correct value from a list, or by writing the value in a text field. The result of a search is a list of links to the matching services which can be browsed with the instance browser.
The semantic instance browser shows information about the currently selected advertisement instance such as the topic and contact information for acquiring the service. In addition, the relations to the ontologies and the recommended links for the current advertisement are shown.
This paper proposed a promising new way to organize and annotate advertisements in online Yellow page directories by using semantic web ontologies. The underlying ontology can be used as a navigational device by which the end-user can map her needs onto relevant offerings (advertisements). After finding a suitable service and provider, the ontological structure relating services and their instances, the advertisements, provide a means by which recommendations to related advertisements of interest can be generated.
A demonstrational semantic web -based YP system was described. This system is, however, only a first small step towards the vision of semantic YP presented in this paper. Several advances are needed before an actual semantic YP service could be deployed. Firstly, full blown service and other ontologies need to be designed and tested on real customer data. Secondly, the advertisements should be annotated accordingly. This is more demanding and time consuming than current practices. However, annotation work is the price to be paid for better accuracy in information retrieval and for better servicing the end-user. We feel, that the advertising companies and the YP operators should be interested in providing such annotations since the matchmaking facility is the key point of the whole directory service business in the first place.
Thanks to Arno Aalto, Merja Jalava, Tuomo Kajava, Petri Lindgren, Vilho Raatikka, Pekka Raitala, and Irmeli Rämö.
* We investigated among others the following representative YP sites: http://www.superpages.com/, http://uk.yell.com/, http://www.brabys.com/, http://web.wlwonline.de/. Also other than YP directory services was investigated, such as http://www.yahoo.com.
** This idea has been applied before in other domains, e.g., in photo annotation and retrieval [7,3]