Content-Oriented Composite Service Negotiation with Complex Preferences Reyhan Aydo˘ gan reyhan.aydogan@gmail.com Department of Computer Engineering Bo˘ gaziçi University Bebek, Istanbul,Turkey ABSTRACT In e-commerce, for some cases the service requested by the con- sumer cannot be fulfilled by the producer. In such cases, service consumers and producers need to negotiate their service require- ments and offers. Whereas some multiagent negotiation approaches treat the price as the primary construct for negotiation, we con- sider that the service content is as much important as the price. Therefore, this study mainly focuses on the content of the service described in a common ontology accessed by both agents for com- mon understanding. Acquiring user’s preferences and acting upon these preferences are crucial tasks for a consumer agent as far as the negotiation is concerned. Since the size of complete preference in- formation increases exponentially with the number of attributes and size of domain, it is required to keep these preferences in a com- pact way. There are a variety of ways of representing preferences and using these structures for automatic generation of consumer’s request. This research develops an automated negotiation approach in which the consumer takes the preferences of the user in an effi- cient way and uses these preferences in the generation of request. For this purpose, we design several strategies to generate requests to take the best offer by the producer. On the other side, in order to obtain a more effective negotiation results the producer tries to learn the consumer preferences from the bid exchanges incremen- tally in order to refine its offer over time. Furthermore, for some complicated services desired by the consumer, a single producer by itself may not meet the consumer’s needs. In such cases, the system should allow consumers negotiating with multiple service producers as far as composite services are concerned. 1. INTRODUCTION Service-oriented architectures (SOAs) are being used extensively to build multiagent systems in which autonomous agents request and provide services to each other [1]. In traditional SOAs, service consumers interact with a service provider to receive a predefined service, which is typically advertised by the provider though the registries. However, in many realistic settings, the content of the service would vary based on the receiving consumer, necessitating a negotiation between the consumer and the producer When the producer does not provide the exact service requested by the consumer because of lack of resources or some business con- straints over the service [2], the consumer and producer negotiate Copyright c  2008, International Foundation for Autonomous Agents and Multiagent Systems (www.ifaamas.org). All rights reserved. the content of the service [3]. The content of the service consists of multiple issues. Consider that the service that is being negotiated between producer and con- sumer is that of selling a wine. Possible issues constructing the service content may be color, region, grape, flavor, body and so on. The preferences may vary from consumer to consumer. Preferences represent the user choices when more than one alternative value ex- ists for a given issue. For example, alternative values for the color are red, rose and white. A preference may be strictly specified. For example, for the color attribute the consumer prefers only red wine. Any other wine whose color is different from red is not acceptable by this consumer. In our previous work [4], the preferences were in this form. Another important thing is the relative importance degree of the issues. For example, for some consumers the color of wine is more important than the grape whereas the grape may be more significant for other. The importance degree of each issue may be different for each consumer. Thus, assigning different weights to each service component can be useful for evaluation of the services. Some stud- ies take these weights as a priori and uses the fixed weights [5]. However, in many realistic settings, consumer’s preferences with the weight values are not known by the producer at the beginning of the negotiation. Hence, it is more convenient for the producer to learn these preferences from the interactions with the consumer. For more flexible negotiation scheme, more flexible preference representation can be used where the relative preference ordering over the values of the issue can be taken into account. A consumer may prefer red wine to rose wine and prefers rose wine to white wine. Of course, a preference may be in a more complicated form involving the dependency among features. For instance, the con- sumer may prefer red wine to white wine when the grape of the wine is Chardonnay. For other values of grape, the preference of the consumer may change. To obtain complete preferences from the user may require too many questions to be asked to user in most of the cases. There are some compact preference representa- tions, which require less question such as GAI-nets [8], CP-nets [9] and so on. Even though they are compact, they can represent most practical preference orderings. In our study [7], we use CP-net to represent preferences and construct a preference graph by inferring CP-net. By using a heuristics on this graph, we obtain a complete preference ordering and develop strategies to generate consumer’s requests in accordance with these orderings. 2. REPRESENTING PREFERENCES The concise representation of the user’s preferences plays an im- portant role in terms of generating requests and counter offers. Ac- quiring these preferences from the user in an efficient way and us- ing them in automatic generation of requests and offers are impor- Cite as: Extended Thesis Abstract for Doctoral Mentoring Program, R. Aydogan., Proc. of 7th Int. Conf. on Autonomous Agents and Multi- agent Systems (AAMAS 2008), Padgham, Parkes, Müller and Parsons (eds.), May, 12-16., 2008, Estoril, Portugal, pp. 1725-1726.