An agent that matches workers and jobs, and which is capable in principle of gathering information from any database on the Internet, is described in this paper. The agent becomes capable of communicating with an unfamiliar HTML-based program or database on the basis of a single observation of a user interacting with that program or database.
This agent has been in use in Slovenia by the national employment office for seven years, and attracts many users; the number of accesses per month is equal to one tenth of the national population. Users can communicate with the agent in spoken Slovenian or English, but most seem to prefer to fill in forms. These forms can be close to natural language; slight variations between forms are dealt with by soft matching, and future research will deal with handling greater variations.
Part of the agent’s approach to combining many different data sources is to convert all information into plain text. The paper includes examples of the algorithms and data structures that support this, but not details of coding. Worthwhile comments on alternative approaches to the ones taken are included.