Database and Information Extraction

Using RAPIER, JobQuery has learned to extract information from netnews job postings and automatically build a structured database of jobs. The current extraction system was trained on 300 computer-related job postings annotated with the labelled substrings to be extracted. The database contains information automatically and continuously extracted from the newsgroup austin.jobs. It attempts to extract the following information about each computer-related job it finds in this newsgroup:

• job title
• company name
• city
• state
• country
• recruiter
• salary
• required years of experience
• desired years experience
• required degree
• desired degree
• area of computing
• programming languages
• computing platforms (OS)
• computing applications
• post date

Query Processing

JobQuery provides two query interfaces to the resulting database, a natural-language front-end and a more traditional menu-based HTML-form interface. Using CHILL, the natural-language interface has been produced by training on about 100 examples of English questions paired with a logical form (a Prolog query which answers the question when executed). When answering a question, JobQuery prints an English gloss of the formal query it has constructed in order to inform the user of how it interpreted their question. Here are a list of sample queries it can currently answer.

The menu form allows users to specify values or constraints on the various job-information fields listed above. Looking at the menus in the form interface provides useful information on the type of data in the database. The user is encouraged to first use the natural-language interface, if it does not interpret the question correctly, the user can provide the correct interpretation using the form interface. The resulting pair of English and formal query will later be used to retrain the English parser. Therefore, you can help teach JobQuery to understand English!