project

Software Agents in E-Commerce

The term "software agents" here distinguishes it from hardware agents (robot) or human agents. A software agent stays in the computer, can move in the networks, can observe its environment and can take several actions to reach its goals. It also has certain level of intelligence, can "think", can learn from its environment and can make decision under uncertainty.

The definition of agent is unclear because there are many different definitions for it. To see how people define a software agent implicitly, click here, here, here and here. Click here to see why it is so difficult define an agent. And click here to see the tutorial for software agents by Pattie Maes.

Why use Software Agents in E-commerce?

E-commerce will become a part of the human life in 21st century. According to the research by CommerceNet, the growth of e-commerce during last 9 months is about 40%. Unfortunately, the complexity of the e-commerce will increase along with the increasing of the number of users and items sold. A huge number of unstructure information in the internet make the cybermarket so complex that people can't use it as efficient as before, especially for untrained consumer who also want to find the best deal in the cybermarket. Thus, a software agent is needed to help consumers in buying goods and services via e-commerce.

Beside, a seller doesn't want to lost its customers just because the price offered is higher than what customer willing to buy or what its competitors offered. And seller also want to get as much data as possible from its customer. Thus, in order to make more profit, a flexible price should be offered for different customer, and different advertisement should be sent too. To deal with this problem, a seller's negotiator-agent is required to negotiate with the customer's agents, and its advertisement-agent will visit or send ads to its customer frequently. A spy-agent will negotiate with its competitors' agents, a survey-agent will collect the data from customer, and so on.

How a software agent work in the transaction process? First, buyer tell his/her agent the description of items he/she want to buy. Then the agent will search the items, negotiate with the sellers' agents, choose the best deal, tell the user, and make the transaction. Click here to see its simple diagram. On the other side, seller's agent can also perform an autonomous task to attract customers by sending advertisements to customer's agents. Then customer's agent will check the ads, perform filtering, check the procurement database, check the prices offered by other sellers, tell the user or negotiate directly to the seller's agent.

To help both buyer's agents and seller's agents, a broker agents will help buyer's agents to meet the seller's agents, and receive the transaction fee. It will work like a data miner who actively search and store all information about the agents involved in the cybermarket. A chain of agents then will exist in the cybermarket. As the buyer's or seller's agent become more and more intelligent, it can find more and more efficient way to make transactions, and at last the broker agents will extinct.

Click here to see the tutorial of how software agents can help e-commerce (provided by Agents Group, MIT). Security problem is a major issue in the application of mobile agents in e-commerce.

Several Software Agents or projects developed for e-commerce:

BargainFinder Agent is an example of searching agent developed by Anderson Consulting.

Intelligent Agent Software for e-commerce from AmEC, MIT: Kasbah(see its screenshots here), Tete-a-Tete(Frictionless Commerce Inc.) and Market Maker (try it here).

Yenta is an example of agent developed by Lenny Foner.

Intelligent Agent for e-commerce project in Carnegie Mellon University Intelligent Software Agents.

Bargainbot is an agent for searching book developed by Curtin University of Technology, Australia.

SearchSugar to enhance searching in internet developed by Bits & Pixels.

IBM Intelligent Agents research project.

JAVA Agents for Meta-learning (JAM) from Columbia University.

Technology related to Software Agents

1. Article from British Telecommunication Lab. It includes agent communication languages, such as KQML, language for constructing agent applications, ontologies for agent applications such as KIF and ontolingua, agent and client-server computing, and agent and distributed objects.

2. Agents Construction Tools, both tools in research projects or commercial products. A table from Agent Builder.

3. Agent technology and standards. A site from LAIT-UMBC.

4. Software and tools for building agents, compiled by Gerhard Wickler.

There are still lots of sites related to the tecnology in software agents. Several major studies which may be useful in developing intelligent agents in the future are Soft Computing, Artificial Life, Machine Learning, Game Theory, Human-Computer Interface, Distributed Artificial Intelligence.