ERCIM News No.25 - April 1996 - CNR
An Agent-Based Approach for Interacting with the Web
by Amedeo Cesta, Daniela D'Aloisi, Vittorio Giannini
The current wide diffusion of Web sites means that they are accessible
to different kinds of users with different levels of expertise and awareness
on the usage of network resources. At the same time, the quantity of tools
and information accessible is dramatically increasing, making their manipulation
difficult and time-consuming. In order to facilitate access to the Web,
interfaces should provide facilities to support users in dealing with complex
tasks, to perform repetitive and time-consuming operations automatically,
and to better exploit tools and services obtainable through the network.
A promising approach is the use of software agents.
In a collaboration between the Istituto di Psicologia (IP-CNR) and the
Fondazione Ugo Bordoni (FUB), Rome, we are now designing intelligent interfaces
which should help users to manage new network technologies by delegating
tasks to or being helped by software agents. The aim is to allow different
kinds of users to access the Web resources in an effective and efficient
way.
There are two main lines of development:
Implementing Active Interfaces
The interface must not be just a passive actor in the interaction but must
assume a proponent role, actively supporting the users according to their
needs. The interfaces must:
- provide reasoning mechanisms to understand the user's needs and how
best to satisfy them
- supply capabilities for autonomous reasoning in order to supply performances
not explicitly required by the user
- minimise the need for the user to acquire specific competence in using
the tools.
Designing Distributed and Agent-Oriented Architectures
The implementation must exploit recent progress in related technologies.
In particular, the interfaces should:
- have a distributed architecture, subdividing their capabilities among
connected and cooperating units
- adopt the software agent methodology: the software agent is an intelligent
entity designed to solve specific problems on the behalf of the users and
according to their directions.
The software agent technique allows an incremental development of complex
architectures and offers a uniform language to interface different modules.
An integrated environment has been developed in which an Interface Agent
interacts with the users, translating their requests and connecting them
with other local or remote agents and then presenting them with the results.
The Interface Agent is also in charge of defining the information needs
of the user by observing their behaviour and thus deducing the type of support
needed. The connection with other agents is user-transparent.
The other kinds of agents are:
- the Mail Agent that filters and stores the incoming e-mail messages
according to the user's interests
- the News Agent that selects only a small number of messages from the
huge amount delivered daily by news servers
- the Meeting Agent that automatically contacts other similar agents
to organise meetings and take appoint-ments on behalf of the user
- the Info Agent that retrieves data and documents of interest to the
user. All these agents are connected so that they can share common data
and infor-mation. Moreover they can cooperate and communicate using the
KQML (Knowledge Query and Manipulation Language) language/protocol.
An example of application of the general framework is the Info Agent, a
multi-agent system that helps users to manage information coming from their
work environment and from the Web. The Info Agent's task is to assist the
user in retrieving documents and other types of items in repositories, archives
and databases accessible through the network. The ultimate aim of this agent
is to disengage the user as far as possible from knowing:
- what is available on the network
- how the data and other types of infor-mation are structured and organised
- when new entries are available
- where the repositories are located
- what retrieval services are available
- the different access languages and modalities.
Since the Interface Agent is able to reason about the user's requests, it
contacts the Info Agent when users need its services and translates their
requests. The agent's architecture consists of two sub-agents p; the
Local Retriever and the External Retriever p; each devoted to a particular
aspect of the problem. There is an interactive stage between the Interface
Agent, the user and the other two agents in order to select the set of documents
that most satisfy the initial query.
The Local Retriever knows the structure of the archives available in a given
organisation: it is in charge of retrieving scientific and administrative
data. The External Retriever agent is in charge of retrieving documents
on the network. It can work in two modes: retrieval (or query) mode and
surfing mode.
In the first case, it searches for a specific document: this service is
activated by a direct request from the user. The agent uses a user model
and the user's past history to refine the set of retrieved documents so
that their content most reflects the user's needs. In the second case, the
agent navigates the network searching for documents that could interest
the user. The search is driven by a user's profile built and maintained
by the Interface Agent. At the beginning, the Interface Agent uses a setting
introduced by the user; this profile is then refined according to how the
user manages the data found/proposed by the agent.
Both the External and the Local Retriever utilise the same software tool
to perform their search: it is a public-domain software called Harvest,
an integrated set of tools to gather, extract, organise, search, cache and
replicate relevant information across the Internet.
Other search methods or systems can also be provided to be used alone or
along with Harvest; this is one of the advantages of the modular and distributed
architecture of the system framework and its immersion in the Web world.
The Info Agent system is part of an integrated environment in which traditional
software tools are interfaced with agent-technologies to flexibly support
the user in repetitive and tedious office type tasks.
Please contact:
Amedeo Cesta - IP-CNR
Tel: +39 6 86090-209
E-mail: amedeo@pscs2.irmkant.rm.cnr.it
or
Daniela D'Aloisi, Vittorio Giannini - Fondazione Ugo Bordoni
Tel: +39 6 54803422
E-mail: {dany, gvittori}@fub.it
return
to the contents page