# US 8,543,906 B2 · Probabilistic learning method for XML annotation of documents

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 8,543,906 B2
- Document ID: US8543906B2
- Status: Granted
- Kind: B2
- Application number: 11/170,542
- Filed: 2005-06-29
- Published: 2007-01-25
- Granted: 2013-09-24
- Earliest claimed: 2005-06-29
- Assignees: Xerox Corporation
- Applicants: Boris Chidlovskii; Jérôme Fuselier
- Inventors: Boris Chidlovskii; Jérôme Fuselier
- Web page: https://patentgrep.com/patent/US8543906B2

## CPC

- G06F 40/169: Annotation, e.g. comment data or footnotes
- G06F 40/143: Markup, e.g. Standard Generalized Markup Language [SGML] or Document Type Definition [DTD]
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A document processor includes a parser that parses a document using a grammar having a set of terminal elements for labeling leaves, a set of non terminal elements for labeling nodes, and a set of transformation rules. The parsing generates a parsed document structure including terminal element labels for fragments of the document and a nodes tree linking the terminal element labels and conforming with the transformation rules. An annotator-annotates the document with structural information based on the parsed document structure.

## Claims

### Claim 1

A document processor stored in a non-transitory medium comprising:
a probabilistic classifier that classifies fragments of an input document respective to a set of terminal elements by assigning probability values for the fragments corresponding to elements of the set of terminal elements;
a parser that defines a parsed document structure associating the input document fragments with terminal elements connected by links of non-terminal elements conforming with a probabilistic grammar defining transformation rules operating on elements selected from the set of terminal elements and a set of non-terminal elements, the parsed document structure being used to organize the input document, the parser including a joint probability optimizer that optimizes the parsed document structure respective to a joint probability of (i) the probability values of the associated terminal elements and (ii) probabilities of the connecting links of non-terminal elements derived from the probabilistic grammar;
a classifier trainer that trains the probabilistic classifier respective to a set of training documents having pre-classified fragments; and
a grammar derivation module that derives the probabilistic grammar from the set of training documents, each training document having a pre-assigned parsed document structure associating fragments of the training document with terminal elements connected by links of non-terminal elements.

### Claim 2 (depends on 1)

The document processor as set forth in claim 1, wherein the probabilistic grammar is a probabilistic context-free grammar and the joint probability optimizer employs a modified inside/outside optimization.

### Claim 3 (depends on 1)

The document processor as set forth in claim 1, wherein the computer is further programmed to implement:
an XML document converter that converts the input document to an XML document having an XML structure generated in accordance with the parsed document structure.

### Claim 4 (depends on 3)

The document processor as set forth in claim 3, wherein the XML document includes a DTD based on the probabilistic grammar.