# US 2019/0171872 A1 · Semantic normalization in document digitization > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2019/0171872 A1 - Document ID: US20190171872A1 - Status: Published application - Kind: A1 - Application number: 15/829,078 - Filed: 2017-12-01 - Published: 2019-06-06 - Earliest claimed: 2017-12-01 - Assignees: International Business Machines Corporation - Applicants: International Business Machines Corporation - Inventors: Kevin NORTHRUP; Craig M. TRIM; Terry HICKEY; Abisola ADENIRAN; Kenji NORTHRUP - Web page: https://patentgrep.com/patent/US20190171872A1 ## CPC - G06F 40/30: Semantic analysis - G06F 16/284: {Relational databases} - G06F 16/35: Clustering; Classification - G06F 16/93: Document management systems - G06F 40/109: Font handling; Temporal or kinetic typography - G06N 5/022: Knowledge engineering; Knowledge acquisition - G06N 20/00: Machine learning - G06V 30/274: {Syntactic or semantic context, e.g. balancing} - G06V 30/413: Classification of content, e.g. text, photographs or tables - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING ## Abstract Methods, computer program products, and systems are presented. The methods include, for instance: identifying a candidate key corresponding to an object in a document image with a key in key ontology data, based on that the candidate key is semantically interchangeable with the key. A context, position, and style of each objects of the document image is represented in the document metadata. The candidate key is normalized into a normal form. A key class corresponding to the normal form is determined and a confidence score indicating a likelihood of the key class being representative of the candidate key is assessed. A semantic database is updated with the key class upon verification for enhanced processing of future documents. ## Claims ### Claim 1 A computer implemented method for normalizing a key in a document image comprising: obtaining, by one or more processor of a computer, document metadata of the document image, wherein the document metadata include a context, a position, and a style respective to each object appearing in the document image; identifying a candidate key corresponding to an object in the document image with a key in key ontology data, based on that the candidate key is semantically interchangeable with the key; normalizing the candidate key into a normal form; determining a key class corresponding to the normal form, wherein the key class is associated with the key in the key ontology data; assessing a confidence score for the key class, wherein the confidence score indicates a likelihood of the key class being representative of the candidate key, based on the document metadata; and updating a semantic database with the key class, based on verifying the key class according to a preconfigured manner of verification, such that the key class may be efficiently associated with a semantically interchangeable text appearing in other document images. ### Claim 2 (depends on 1) The computer implemented method of claim 1, further comprising: deriving zero or more alias of the key class; comparing the derived alias with aliases in a key alias set corresponding to the key from the semantic database; discovering that the derived alias is not present in the key alias set and that respective confidence score corresponding to each of the derived alias is greater than a preconfigured threshold; and updating the semantic database with the derived alias. ### Claim 3 (depends on 1) The computer implemented method of claim 1, wherein the key in the key ontology data is equal to the key class. ### Claim 4 (depends on 1) The computer implemented method of claim 1, the assessing comprising: comparing a context of the candidate key as specified in the document metadata and a context of the key class, wherein the context of the candidate key indicates other objects appearing together with the candidate key in the document image, and wherein the context of the key class indicates other typical objects appeared together with keys of the key class in past document images; and adjusting the confidence score proportional to a level of similarity between the context of the candidate key and the context of the key class. ### Claim 5 (depends on 1) The computer implemented method of claim 1, the assessing comprising: comparing a position of the candidate key as specified in the document metadata and a position of the key class, wherein the position of the candidate key indicates relative portion of the document image in which the candidate key appears, and wherein the position of the key class indicates relative portions of past document images in which keys of the key class typically appeared; and adjusting the confidence score proportional to a level of similarity between the position of the candidate key and the position of the key class. ### Claim 6 (depends on 1) The computer implemented method of claim 1, the assessing comprising: comparing a style of the candidate key as specified in the document metadata and a style of the key class, wherein the style of the candidate key indicates a font type and size of the candidate key in relation to styles of other objects as appearing in the document image, and wherein the style of the key class indicates font types and sizes of keys of the key class in relation to styles of other objects as typically appeared in past document images; and adjusting the confidence score proportional to a level of similarity between the style of the candidate key and the style of the key class. ### Claim 7 (depends on 1) The computer implemented method of claim 1, wherein the normal form is an extended Backus-Naur form (EBNF) notation, and wherein the document metadata is expressed in JavaScript Object Notation (JSON) format. ### Claim 8 (depends on 1) The computer implemented method of claim 1, wherein the key specified in the key ontology data is a name of a data field in a relational database, wherein the key is associated with a value as extracted from the document image, and the key and the value forms a key-value pair in the relational database for future computation. ### Claim 9 A computer implemented method for normalizing a key in a document image comprising: identifying, by one or more processor of a computer, a candidate key corresponding to an object in the document image with a key in key ontology data, based on that the candidate key is semantically interchangeable with the key; normalizing the candidate key into a normal form; deriving one or more alias of the candidate key from the normal form, wherein the one or more alias is not associated with the key in a semantic database; assessing respective confidence score for each of the one or more alias, wherein the confidence score indicates a likelihood of each alias being representative of the candidate key, based on document metadata of the document image; and updating the semantic database with the one or more alias, based on verifying the one or more alias according to a preconfigured manner of verification, such that the one or more alias may be efficiently associated with a text from other document images, based on that the text semantically matches with the candidate key. ### Claim 10 A computer program product comprising: a computer readable storage medium readable by one or more processor and storing instructions for execution by the one or more processor for performing a method for normalizing a key in a document image comprising: obtaining document metadata of the document image, wherein the document metadata include a context, a position, and a style respective to each object appearing in the document image; identifying a candidate key corresponding to an object in the document image with a key in key ontology data, based on that the candidate key is semantically interchangeable with the key; normalizing the candidate key into a normal form; determining a key class corresponding to the normal form, wherein the key class is associated with the key in the key ontology data; assessing a confidence score for the key class, wherein the confidence score indicates a likelihood of the key class being representative of the candidate key, based on the document metadata; and updating a semantic database with the key class, based on verifying the key class according to a preconfigured manner of verification, such that the key class may be efficiently associated with a semantically interchangeable text appearing in other document images. ### Claim 11 (depends on 10) The computer program product of claim 10, further comprising: deriving zero or more alias of the key class; comparing the derived alias with aliases in a key alias set corresponding to the key from the semantic database; discovering that the derived alias is not present in the key alias set and that respective confidence score corresponding to each of the derived alias is greater than a preconfigured threshold; and updating the semantic database with the derived alias. ### Claim 12 (depends on 10) The computer program product of claim 10, wherein the key in the key ontology data is equal to the key class. ### Claim 13 (depends on 10) The computer program product of claim 10, the assessing comprising: comparing a context of the candidate key as specified in the document metadata and a context of the key class, wherein the context of the candidate key indicates other objects appearing together with the candidate key in the document image, and wherein the context of the key class indicates other typical objects appeared together with keys of the key class in past document images; and adjusting the confidence score proportional to a level of similarity between the context of the candidate key and the context of the key class. ### Claim 14 (depends on 10) The computer program product of claim 10, the assessing comprising: comparing a position of the candidate key as specified in the document metadata and a position of the key class, wherein the position of the candidate key indicates relative portion of the document image in which the candidate key appears, and wherein the position of the key class indicates relative portions of past document images in which keys of the key class typically appeared; and adjusting the confidence score proportional to a level of similarity between the position of the candidate key and the position of the key class. ### Claim 15 (depends on 10) The computer program product of claim 10, the assessing comprising: comparing a style of the candidate key as specified in the document metadata and a style of the key class, wherein the style of the candidate key indicates a font type and size of the candidate key in relation to styles of other objects as appearing in the document image, and wherein the style of the key class indicates font types and sizes of keys of the key class in relation to styles of other objects as typically appeared in past document images; and adjusting the confidence score proportional to a level of similarity between the style of the candidate key and the style of the key class. ### Claim 16 (depends on 10) The computer program product of claim 10, wherein the normal form is an extended Backus-Naur form (EBNF) notation, and wherein the document metadata is expressed in JavaScript Object Notation (JSON) format. ### Claim 17 (depends on 10) The computer program product of claim 10, wherein the key specified in the key ontology data is a name of a data field in a relational database, wherein the key is associated with a value as extracted from the document image, and the key and the value forms a key-value pair in the relational database for future computation. ### Claim 18 A computer program product comprising: a computer readable storage medium readable by one or more processor and storing instructions for execution by the one or more processor for performing a method for normalizing a key in a document image comprising: identifying a candidate key corresponding to an object in the document image with a key in key ontology data, based on that the candidate key is semantically interchangeable with the key; normalizing the candidate key into a normal form; deriving one or more alias of the candidate key from the normal form, wherein the one or more alias is not associated with the key in a semantic database; assessing respective confidence score for each of the one or more alias, wherein the confidence score indicates a likelihood of each alias being representative of the candidate key, based on document metadata of the document image; and updating the semantic database with the one or more alias, based on verifying the one or more alias according to a preconfigured manner of verification, such that the one or more alias may be efficiently associated with a text from other document images, based on that the text semantically matches with the candidate key. ### Claim 19 A system comprising: a memory; one or more processor in communication with memory; and program instructions executable by the one or more processor via the memory to perform a method for normalizing a key in a document image comprising: obtaining document metadata of the document image, wherein the document metadata include a context, a position, and a style respective to each object appearing in the document image; identifying a candidate key corresponding to an object in the document image with a key in key ontology data, based on that the candidate key is semantically interchangeable with the key; normalizing the candidate key into a normal form; determining a key class corresponding to the normal form, wherein the key class is associated with the key in the key ontology data; assessing a confidence score for the key class, wherein the confidence score indicates a likelihood of the key class being representative of the candidate key, based on the document metadata; and updating a semantic database with the key class, based on verifying the key class according to a preconfigured manner of verification, such that the key class may be efficiently associated with a semantically interchangeable text appearing in other document images. ### Claim 20 (depends on 1) The computer implemented method of claim 1, further comprising: deriving zero or more alias of the key class; comparing the derived alias with aliases in a key alias set corresponding to the key from the semantic database; discovering that the derived alias is not present in the key alias set and that respective confidence score corresponding to each of the derived alias is greater than a preconfigured threshold; and updating the semantic database with the derived alias. ### Claim 21 (depends on 1) The computer implemented method of claim 1, wherein the key in the key ontology data is equal to the key class, wherein the key specified in the key ontology data is a name of a data field in a relational database, wherein the key is associated with a value as extracted from the document image, and the key and the value forms a key-value pair in the relational database for future computation. ### Claim 22 (depends on 1) The computer implemented method of claim 1, the assessing comprising: comparing a context of the candidate key as specified in the document metadata and a context of the key class, wherein the context of the candidate key indicates other objects appearing together with the candidate key in the document image, and wherein the context of the key class indicates other typical objects appeared together with keys of the key class in past document images; and adjusting the confidence score proportional to a level of similarity between the context of the candidate key and the context of the key class. ### Claim 23 (depends on 1) The computer implemented method of claim 1, the assessing comprising: comparing a position of the candidate key as specified in the document metadata and a position of the key class, wherein the position of the candidate key indicates relative portion of the document image in which the candidate key appears, and wherein the position of the key class indicates relative portions of past document images in which keys of the key class typically appeared; and adjusting the confidence score proportional to a level of similarity between the position of the candidate key and the position of the key class. ### Claim 24 (depends on 1) The computer implemented method of claim 1, the assessing comprising: comparing a style of the candidate key as specified in the document metadata and a style of the key class, wherein the style of the candidate key indicates a font type and size of the candidate key in relation to styles of other objects as appearing in the document image, and wherein the style of the key class indicates font types and sizes of keys of the key class in relation to styles of other objects as typically appeared in past document images; and adjusting the confidence score proportional to a level of similarity between the style of the candidate key and the style of the key class. ### Claim 25 (depends on 1) The computer implemented method of claim 1, wherein the normal form is an extended Backus-Naur form (EBNF) notation, and wherein the document metadata is expressed in JavaScript Object Notation (JSON) format.