Learning to understand web site update requests

Although Natural Language Processing (NLP) for requests for information has been well-studied, there has been little prior work on understanding requests to update information. In this paper, we propose an intelligent system that can process natural language website update requests semi-automatically. In particular, this system can analyze requests, posted via email, to update the factual content of individual tuples in a database-backed website. Users' messages are processed using a scheme decomposing their requests into a sequence of entity recognition and text classification tasks. Using a corpus generated by human-subject experiments, we experimentally evaluate the performance of this system, as well as its robustness in handling request types not seen in training, or user-specific language styles not seen in training.