Skip to main content
españolES
European Commission logo print header

E-Commerce with guiding Agents based on Personalized Interaction Tools

Objetivo

The project aims at improving consumer-supplier relationships in future e-commerce interfaces featuring agents, which can converse with users in written natural language ("chatterbots"). We extend the capabilities of today's chatterbots by a unique combination of technologies:
User guidance in conversational interaction: Whereas contemporary graphical interfaces are usually "passive" and often place overly high cognitive demands on their users, project COGITO aims at a system that is not merely re-active to some user request, but pro-active and capable of engaging in a goal-directed conversation with the user, e.g. in order to
- specify the overall goal of a (complex) task the user intends to fulfill
- clarify uncertainties in the understanding of the goal/task and negotiate the best strategy
- monitor and interpret the user's behaviour (actions and information conveyed during interaction).
On this basis the system's subsequent behaviour is then adapted accordingly, e.g. by taking the initiative to recommend new information items or search strategies to the user if feasible. In addition, the naturalness of interaction, especially for casual users, will be enhanced by appropriate 2D and 3D animations that express appropriate emotional reactions by the agent.
Personalization and profile extraction: Even chatterbots with a sophisticated repertoire of conversational skills will fail to be more than entertaining if they do not treat a user as an individual having specific needs, preferences, etc. In COGITO, we combine content-based filtering, where user profiles are generated based on content features extracted from documents that users find relevant, with collaborative filtering which clusters users according to their expressed taste to generate recommendations within these virtual communities. Since collaborative filtering is purely based on users' opinions, it has proven to be especially useful in taste-based domains such as books, movies, music, or TV.
Machine learning techniques are used to associate user characteristics with tastes and purchases.
Pro-activeness through intelligent retrieval: In general, a user is unlikely to be able to retrieve appropriate database entries directly. To avoid repeated query modifications, the retrieval process in COGITO involves three components: the chatterbot, the content management module, and an automatic query expansion system accessing the user profiles called the "prompter". Applying a rule-interpreter, the prompter is able to use the structure of documents, e.g. given in XML, and, via comparisons with the profile terms, can expand the original query assumed by the chatterbot. Thus, a considerably more precise search in the product database is accomplished.

Objectives:
The rapid evolution of interactive Internet services has led to both a constantly increasing number of modern Web sites, and to an increase in their functionality, which, in turn, makes them more complicated to use. Thus, any attempt to enhance the consumer-supplier relationship in e-commerce has to meet the challenge of coping with two almost contradictory goals: A useful e-commerce application should not only mimic traditional catalogues, order forms and other printed material which used to be the basis of communication between consumers and suppliers. Instead, the inherent potential for interactive data processing and man-machine dialogue should be used by e-commerce applications to meet the user's need for immediate situation-specific response, instantly available problem-specific advice, and better ways to access and inspect the supplier's offer. However, the currently prevailing graphical user interfaces, which rely on menu selection and navigation, require a considerable cognitive overhead. This may be tolerable to frequent users, but will in many cases deter casual users, especially those who are not yet used to computers. Hence, we need to combine the usefulness of a value-added service with a high degree of usability, and dedicated measures to build up trust and confidence in inexperienced users.
To meet these conditions the interaction must be, at the same time, as natural as possible, thus enabling users to rely on their communicative skills, it must convey precise and relevant information and address the personal background of the individual user. The interface must use best practice solutions to achieve a high degree of dialogue intelligence and employ an appropriate graphical design.
The COGITO project aims at extending the capabilities of chatterbots by a unique combination of technologies

Work description:
The COGITO solution is based on "intelligent personalized agents" which represent virtual assistants or advisors (also visually) by modelling their ability to support customers. They could instruct customers in the use of a Web site, point out new offers, help sift through products etc. There have already been some efforts made in developing chat robots ("chatterbots") based on expert systems.
In most applications, chatterbots are used as guides. Virtual assistants must be capable of flexible behaviour if they are to be acceptable to users on a long-term basis. Simple chatterbots, such as the first system of this type, ELIZA, and most of its successors only simulate conversation without utilizing any knowledge about the individual users and their actual behaviour during online sessions. Such simple chatterbots are not powerful enough to serve as a medium for customer advice. This means that, in addition to some of the abilities already available (e.g. help question answering controlled by simple event-action rules), a further reaching dialogue management will be needed to help 1) interprete the individual dialogue situation and 2) more complex dialogues allow goal-directed strategies to be pursued (cooperative behavior, convincing argumentation) to achieve an adequate, non-stereotypical repertoire of reaction.

*Whereas an increase in general dialogue intelligence can be achieved by elaborate rule sets, the naturalness of the dialogue depends on the degree in which the system is able to adapt to individual users, whether it is able to learn about their preferences and attitudes during the dialogue, and memorize them for later use. For this goal, we will include learning mechanisms that extract permanent features of a given user from the dialogue (of course, the user must consent to this, and will be given an opportunity to check and change the data). The resulting user profiles will be further analysed to automatically extract usage patterns from the data given about user communities. This helps content providers to tailor their offers to the customers' needs, and can be used to generate assumptions about new users, when they start. Published research to date shows that a further development of personalized interfaces into more flexible dialogue-oriented interfaces could increase the acceptance of such personalized agents. Therefore, we will add a component for intelligent access to the supplier's repository, which will act as a "prompter" helping the chatterbot in problematic retrieval situations (too many, too few hits, etc). It will rely on a repository of search heuristics for automatic query construction, expansion, and modification, and exploit the profiles as well as domain knowledge provided by the content manager.

*The expressive visualization of a virtual advisor - e.g. as an animated cartoon "Persona" - can be a direct and useful complement to the proposed dialogue approach. By being able to take the initiative, rather than simply reacting to user input and commandos, a system can take on the role of an independent agent during dialogue. To make this role as a true counterpart transparent, it is helpful to visualize the agent - thus the agent is also visually present and can express questions, recommendations, warnings, etc by means of mimic and gestures.
In a concrete application, the dialogue and retrieval functionality of a chatterbot must be backed by appropriate content management capabilities. While COGITO will support contemporary commercial solutions to gain a higher exploitation potential, the project will also use future developments in content management, especially by integrating tools to handle XML-based Web pages. XML will be used as internal representation language in which knowledge bases (representing contexts of the dialogue system) are to be coded. The XML orientation of the content management will guarantee that the final system can be used with a broad range of back-end systems.

Milestones:
The project aims at innovative software components allowing e-commerce companies to conveniently set up and maintain Web sites which address customers in personalized and pro-active ways. As part of the project, a demonstrator will be installed at the BOL site and experimentally used in BOL's online media shop. The agent will be able to extract information from the user's dialogue behaviour, and, based on its rule bases, provide access to probably relevant product information, such as book and CD descriptions, or help the user in finding appropriate offers. The overall system architecture will be centred around a chatterbot system kernel - the Session Manager - providing the basic functionality, invoke appropriate rules from a Chat Rule Base. The dialogue is recorded for each user by a Log interpreter. The work on the User interface will result in a Client based on HTML, XML, and Java. It will incorporate an interactive chatterbox allowing for two-way conversational interaction in written natural language in German and English, and will employ a Visual Persona based on 2D as well as 3D-animated cartoons (the virtual assistant). The Connector to back-end systems comprises components enabling the system to access external services and knowledge sources. In addition to providing an interface to Application databases, the project will focus on XML-based Content Management: Here, we will provide a suite of tools supporting the creation, management, and distribution of XML documents. Specific document type definitions needed for the representation of knowledge inside a COGITO application, as well as DTD's referring to RDF-complying documents residing in a back-end system, or are retrieved accessing broker services, will be taken into account. Further additions to the chatterbot are components for Profile extraction and context-dependent prompting. For the first task, the automatic construction of a structured dialogue history from the logged data serves as a start.

Ámbito científico

Programa(s)

Tema(s)

Convocatoria de propuestas

Data not available

Régimen de financiación

CSC - Cost-sharing contracts

Coordinador

FRAUNHOFER IAF
Aportación de la UE
Sin datos
Dirección
Tullastr. 72
79108 MÜNCHEN
Alemania

Ver en el mapa
Enlaces
Coste total
Sin datos

Participantes (5)

Compartir esta página

Descargar

Última actualización: 12 Junio 2012

Mi folleto

Permalink: https://cordis.europa.eu/project/id/IST-1999-13347/es

European Union, 2024