Using Shallow Natural Language Processing in a Just-In-Time Information Retrieval Assistant for Bloggers

Just-In-Time Information Retrieval agents proactively retrieve information based on queries that are implicit in, and formulated from, the user’s current context, such as the blogpost she is writing. This paper compares five heuristics by which queries can be extracted from a user’s blogpost or other document. Four of the heuristics use shallow Natural Language Processing techniques, such as tagging and chunking. An experimental evaluation reveals that most of them perform as well as a heuristic based on term weighting. In particular, extracting noun phrases after chunking is one of the more successful heuristics and can have lower costs than term weighting. In a trial with real users, we find that relevant results have higher rank when we use implicit queries produced by this chunking heuristic than when we use explicit user-formulated queries.