MAIL AGENTS (MAILBOTS) These agents assist the user with e-mail (Figure 13.4). For example, Maxims (Maes,
E-MAIL AGENTS (MAILBOTS) These agents assist the user with e-mail (Figure 13.4). For example, Maxims (Maes,
1994) monitors what the user does routinely with e-mail and memorizes the user's situation-action pairs. These pairs are stored in a memory of examples. The situations are described in terms of a set of features, including the names of those who send mes- sages to the user and receive messages from the user. When a new situation occurs, the agent analyzes the features of the situation and suggests an action to the user (e.g., read, delete, forward, or archive). The process is similar to case-based reasoning. The agent communicates with Eudora, a Windows-based e-mail software system.
The agent measures the confidence (or fit) of a suggested action to a situation. If the confidence is high, the agent executes the suggestion with the approval of the user. Otherwise, the agent waits for instructions On what to do. The agent's performance improves with time as the memory of examples increases. Several commercial e-mail agents are available (e.g., Beyond Mail for intelligent messaging from Banyan Inc., www.banyan.com).
Several other e-mail agents help the user to handle large numbers of messaged. For example, Motiwalla (1995) developed an intelligent agent for prioritizing e-mail mes- sages based on the user's preferences or knowledge.
According to Murch and Johnson (1999), e-mail agents can • Control any unwanted, unsolicited e-mail.
• Alert users by voice if a designated message arrives. • Automatically forward mail messages to designated destinations. • Consolidate mail from several sources, the way the user wants it. • Search the Internet for certain sources and deliver them to the user by e-mail. • Distinguish business-related e-mail from private or personal mail. Automatically
answer mail and respond according to conditions, for example, "I am on vacation
C H A P T E R 13 INTELLIGENT SYSTEMS OVER THE INTERNET
PattieMaes 11:11 PM 6/2/93... 1
Read Eudora and the message assistant
Any User 7:08 PM 12/17/9.. 1 Delete Re: Annual ski trip Pattie Maes
11:32 AM 7/10/93.. 1
Read Demos and lunch
A User 7:03 PM 12/15/8.. 1 Delete Re: Annual ski trip Chuck "Thin Man"...
7:01 PM 12/14/9.. 1
Read Proposal lor longer summer break
PattieMaes 10:21 PM 7/14/93.. 1
Read Agent stutl fepository
Chuck "Thin Man"... 7:01 PM 12/14/9.. 1
Read Longer summer break in 1994
Chuck "Thin Man"... 7:01 PM 12/14/9.. 1
Read Longer summer break in 1994 - correction!!
PattieMaes 8:59 PM 7/12/93.. 1
Read Week ending 7/9
Some User 7:00 PM 12/14/9.. 1 Delete Annual ski trip Chuck "Thin Man"...
7:01 PM 12/14/9.. 1
Read Proposal lor longer summer break
The Daft Stranger 6 2 2 PM'12/22/9.. 3
Re: Radio Shack mixer
The Oaik Stranger 6:22 PM 12/22/9.. 3
Re: Radio Shack mixer
Ian Smith 8:51 PM 12/23/9.. 2 •>bpm Re: Stanton 680 cartridges
14/14K/10K I
V• Trash
F I G U R E 1 3 . 4 E-MAIL A G E N T
The e-mail agent makes recommendations to the user (middle column). It predicts what actions the user will take on messages, such as which messages will be read and in which order and which messages will be deleted, forwarded, archived, and so on.
Source: Maes (1994), p. 34. Maes, P. (1994, July). "Agents That Reduce Work and Information Overload," Communication of the ACM, Vol. 37, No. 7.
• Perform regular administrative tasks involving desktop e-mail (e.g., backing up files, archiving, indexing). i WEB BROWSING ASSISTING AGENTS
Some agents can facilitate browsing by offering the user a tour of the Internet. Such an agent, known as a tour guide, works while the user browses. For example, Web Watcher
(www.cmu.edu) helps in finding pages related to the current page, adding hyperlinks to meet the user's search goal and giving advice on the basis of user preferences.
Another example is Letizia (www.media.mit.edu/research/softwareagents). This agent monitors the user's activities with a browser and collects information about the user's behavior. Using various heuristics, the agent tries to anticipate additional items
that might be of interest to the user. A similar agent is Netcomber Activist from IBM (activist.gpl.ibm.com). This agent monitors a user surfing through the Yahoo catalog.The
user can build an interest profile, customize newspapers for daily reading, and so on. FREQUENTLY ASKED QUESTIONS AGENTS
These agents guide people to the answers to FAQs. People tend to ask the same or sim-
P A R T I V INTELLIGENT DECISION SUPPORT SYSTEMS
language, thus asking the same question in several different ways. The agent addresses the problem by indexing large numbers of FAQ files and providing an interface where people can pose their questions in a natural language. The agent uses the text of a ques- tion to locate the answer. Because of the limited number of FAQs and the semistruc- turedness of the questions, the reliability of FAQ agents is very high. GTE Laboratories has developed a FAQ agent that accepts questions from users of Usenet news groups in natural language and answers them by matching question-answer pairs (Whitehead, 1995). FAQFinder can deal with complex FAQs that change over time. The agent can also deal with multiple FAQs in the same domain. For details see O'Leary (1996). Both www.askjeeves.com and agents.umbc.edu are available to the public (see Internet Exercise 10).
INTELLIGENT SEARCH (OR INDEXING) AGENTS Web information seeking and retrieval are increasingly important and suitable for soft-
ware agents (Detlor and Arsenault, 2002). Web robot, spider, wanderer, and similar names describe agents that traverse the Web and perform such tasks as information retrieval and discovery, validating links or HTML, and generating statistics. These search engines (or indexing agents) are very popular, and thousands of them, many of which are very specialized, are available (www.searchengineguide.com). Representative names are InfoSeek, Lycos, Excite, and Hotbot. To achieve better results there are metasearch engines that combine several search engines and other methods of search. Metasearch engines (e.g., Spider, Savvy Search, Metacrawler, All-
in-One, Web Compass) integrate the findings of the various search engines to answer queries posted by the user.
Indexing agents carry out massive autonomous searches of the Web. First, they scan millions of documents and store an index of key words and words from document titles and texts. The user can then query the agent by asking it to find documents con-
taining certain key words. For details, see Etzioni and Weld (1995) and Wooldridge (2002). Indexing agents were developed for knowledge sharing and acquisition in large databases and documents. For example, see Jones et al. (1995).
INTERNET SOFTBOT FOR FINDING INFORMATION The search agents described above suggest locations on the Web to the user. Their sug-
gestions are based on a weak model of what the user wants and what information is available at the suggested location. An Internet softbot attempts to determine what the user wants and to understand the contents of information services. EtziOni and Weld (1994) discuss the pioneering work on these at the University of Washington. Early softbot agents were able to work only with structured information, such as stock quotes and weather maps or the Federal Express package-tracking service. Therefore, early agents relied on a simple model of the information service for the precise seman- tics associated with information provided by the service, increasing the reliability of the search. Also see Chen et al. (1997). Recently, Internet softbots such as google.com and www.askjeeves.com have become more powerful.