As a project, build an information extractions system for some knowledge domain to be used on the WWW. See Section 15.5.3 for suggestions.

Data from Section 15.5.3

The World Wide Web offers many exciting challenges as well as interesting opportunities for artificial intelligence and natural language understanding research. One of the biggest is the need to design intelligent software that can summarize “interesting” web-based materials. There are actually two parts to this problem: locating potentially interesting information on the web and then extracting (or mining) critical components of that information. Although the first problem of locating potentially interesting information is

critical (and many of the techniques of Chapters 10, 13, and 15 are important), we now consider a template matching and filling approach to address the second problem.

After locating information, perhaps by key-word match or use of a more sophisticated search-engine, an information extraction system takes as input this unrestricted text and then summarizes it with respect to a pre-specified domain or topic of interest. It finds useful information about the domain and encodes this information form suitable for report to the user or for populating a structured database.

In contrast to an in-depth natural language understanding system, information extraction systems skim a text to find relevant sections and then focus only on processing these sections. We propose a template-based approach similar to extraction techniques proposed by several researchers (Cardie 1997, Cardie and Mooney 1999). An overview of our information extraction system is presented in Figures 15.20 and 15.21. Suppose we want to find and summarize from the WWW information related to computer science universitylevel faculty positions. Figure 15.20 presents a target job ad and also presents a template of the information we want our software to extract from this and similar job ads.

Transcribed Image Text:

Semantic interpretation of natural language query: hire object İ person: "john smith" Expanded graph for query: hire object employee: "john smith" agent agent person: ? manager: ? manager of_hire: