Using Semantics to Identify Web Objects

Many common web tasks can be automated by algorithms that are able to identify web objects relevant to the user's needs. This paper presents a novel approach to web object identificalion that finds relationships between the user's actions and linguistic information associated with web objects. From a single training example involving demonstration and a natural language description, we create a parameterized object description. The approach performs as well as a popular web wrapper on a routine task, but it has the additional capability of performing in dynamic environments and the attractive property of being reusable in other domains without additional training.