The success of personal information agents depends on their ability to
provide task-relevant information.
This paper presents WordSieve, a new algorithm that generates context descriptions to guide document indexing and retrieval. WordSieve exploits information
about the sequence of accessed documents to identify words which
indicate a shift in context.
We have tested
WordSieve in a personal information agent, Calvin, which monitors a user's
document access, generates a representation of the user's task context, indexes
the resources consulted, and presents recommendations for other resources that
were consulted in similar prior contexts. In initial experiments, WordSieve
outperforms
term frequency/inverse document frequency at matching
documents to hand-coded vector representations of the task contexts in which
they were originally consulted, where the task context representations are
term vectors representing a specific search task given to the user.
