Language Models and Structured Document Retrieval.
Paul Ogilvie, Jamie Callan · 2002
We discuss possibilities for the use of language models in structured document retrieval. We use a tree-based generative language model for ranking documents and components. Nodes in the tree correspond to document components such as titles, sections, and paragraphs. At each node in the document tree, there is a language model. The language model for a leaf node is estimated directly from the text present in the document component associated with the node. Inner nodes in the tree are estimated using a linear interpolation among the children nodes. This paper also describes how some common structural queries would be satisfied within this model. 1.