Automatic extraction of metadata using a neural network

Claim 1

A method of automatically extracting metadata from a document, the method comprising: (a) providing: a computer readable document including blocks comprised of words, an authority list, including common uses of a set of words, and a neural network trained to extract metadata from compounds; (b) locating authority information associated with the words by comparing the words with the authority list; (c) creating compounds, a first of the compounds describing a first of the blocks and including: first-block words, descriptive information associated with one of the first-block and the first block words, and authority information associated with one first-block word; (d) processing the compounds through the neural network to generate metadata guesses; and (e) deriving the metadata from the metadata guesses.

Claim 2

A method of automatically extracting metadata from a document, the method comprising: (a) providing: a computer readable document including blocks comprised of words, an authority list, including common uses of a set of words, and a neural network trained to extract metadata from compounds; (b) locating authority information associated with the words by comparing the words with the authority list; (c) creating compounds, a first of the compounds describing a first of the blocks and including: first-block words, descriptive information associated with one of the first-block and the first block words, and authority information associated with one first-block word; (d) processing the compounds through the neural network to generate metadata guesses; and (e) deriving the metadata from the metadata guesses.

Claim 17

A method of automatically extracting metadata from a document, the method comprising: (a) providing: a computer readable document including blocks comprised of words, an authority list, including common uses of a set of words, a neural network trained to extract metadata from compounds, and a document knowledge base including positioning information and size information for metadata in known documents; (b) locating authority information associated with the words by comparing the words with the authority list; (c) creating compounds, a first compound describing a first of the blocks and including: first-block words, descriptive information associated with one of the first of the blocks and the first-block words, the descriptive information including: a first bounding box information describing the size and position of the first of the blocks, and font information describing one of the first-block words, and authority information associated with one of the first-block words; (d) processing the compounds through the neural network to generate metadata guesses including: word guesses, a first of the word guesses indicating a possible word type for the one of the first-block words, compound guesses, a first of the compound guesses indicating a possible block type for the first of the blocks and including a compound confidence factor indicating a likelihood that the possible block type is correct, and document guesses, a first of the document guesses indicating a possible document type for the computer readable document and including a document confidence factor indicating the likelihood that the possible document type is correct; (e) deriving analysis data from the metadata guesses, the analysis data including: the first of the compound guesses and the first of the document guesses, proximate block type data derived by comparing the first of the compound guesses against a second of the compound guesses, the second of the compound guesses including a possible block type for a second of the blocks located on the computer readable document proximate to the first of the blocks, proximate block position data derived by comparing the first bounding box information against a second bounding box information, the second bounding box information describing the size and position of the second of the blocks, page position data derived from the first bounding box information and the second bounding box information, and font data derived from the font information; (f) comparing the analysis data to the document knowledge base to improve the metadata guesses; and (g) deriving the metadata from the metadata guesses.

Claim 18

A method of automatically extracting metadata from a document, the method comprising: (a) providing: a computer readable document including blocks comprised of words, an authority list, including common uses of a set of words, a neural network trained to extract metadata from compounds, and a document knowledge base including positioning information and size information for metadata in known documents; (b) locating authority information associated with the words by comparing the words with the authority list; (c) creating compounds, a first compound describing a first of the blocks and including: first-block words, descriptive information associated with one of the first of the blocks and the first-block words, the descriptive information including: a first bounding box information describing the size and position of the first of the blocks, and font information describing one of the first-block words, and authority information associated with one of the first-block words; (d) processing the compounds through the neural network to generate metadata guesses including: word guesses, a first of the word guesses indicating a possible word type for the one of the first-block words, compound guesses, a first of the compound guesses indicating a possible block type for the first of the blocks and including a compound confidence factor indicating a likelihood that the possible block type is correct, and document guesses, a first of the document guesses indicating a possible document type for the computer readable document and including a document confidence factor indicating the likelihood that the possible document type is correct; (e) deriving analysis data from the metadata guesses, the analysis data including: the first of the compound guesses and the first of the document guesses, proximate block type data derived by comparing the first of the compound guesses against a second of the compound guesses, the second of the compound guesses including a possible block type for a second of the blocks located on the computer readable document proximate to the first of the blocks, proximate block position data derived by comparing the first bounding box information against a second bounding box information, the second bounding box information describing the size and position of the second of the blocks, page position data derived from the first bounding box information and the second bounding box information, and font data derived from the font information; (f) comparing the analysis data to the document knowledge base to improve the metadata guesses; and (g) deriving the metadata from the metadata guesses.