Supplementing structured information about entities with information from unstructured data sources

Patent metadata

Field	Value
Publication number	US-9251182-B2
Application number	US-201313798229-A
Country	US
Kind code	B2
Filing date	Mar 13, 2013
Priority date	May 29, 2012
Publication date	Feb 2, 2016
Grant date	Feb 2, 2016

Abstract

A method for supplementing structured information within a data system for entities based on unstructured data analyzes a document with unstructured data and extracts attribute values from the unstructured data for one or more entities of the data system. Entity records with structured information are retrieved from the data system based on the extracted attribute values. Entity references for corresponding entities of the data system are constructed based on a comparison of the retrieved entity records and the extracted attribute values. The entity references are linked to the corresponding entities within the data system, with the entity references including extracted attributes from the unstructured data for corresponding linked entities.

First claim

What is claimed is:

1. A computer-implemented method of supplementing structured information within a data system for entities based on unstructured data comprising:
analyzing documents with unstructured data specifying two or more entities of the structured information and interactions between those two or more entities;
identifying from the interactions within the unstructured data of the documents one or more relationships between entities of the structured information;
extracting attribute values from the unstructured data for one or more entities of the structured information base on a comparison of the unstructured data with one or more dictionaries each including values for a corresponding attribute of an entity within the data system, wherein extracting attribute values from the unstructured data includes:
generating tokens from the unstructured data and comparing the tokens to the values within the one or more dictionaries, wherein at least one value within a dictionary includes a plurality of tokens;
retrieving entity records with structured information form the data system based on the extracted attribute values;

constructing entity references for corresponding one or more entities of the data system based on a comparison of the retrieved entity records and the extracted attribute values;
linking the entity references to the corresponding one or more entities within the data system to supplement the structured information for the corresponding one or more entities with information extracted from the unstructured data, wherein the entity references include extracted attributes from the unstructured data for corresponding linked entities; and
linking entities of the structured information to each other within the structured information to indicate related entities based on the one or more relationships between those entities identified form the interactions specified within the unstructured data of the documents.

2. The computer-implemented method of claim 1 , wherein the data system includes a master data management system and the documents are received from a content management system.

3. The computer-implemented method of claim 1 , wherein extracting attribute values from the unstructured data further includes:
extracting the attribute values from the unstructured data based on an attribute value within the unstructured data and a dictionary value including a common portion of an attribute value and being within a certain distance.

4. The computer-implemented method of claim 1 , wherein an attribute of an entity includes a plurality of atomic attributes, and extracting attribute values from the unstructured data includes:
extracting attribute values for each of the individual atomic attributes from the unstructured data.

5. The computer-implemented method of claim 1 , wherein constructing entity references includes:
constructing entity references for corresponding one or more entities of the data system based on a fuzzy match of the retrieved entity records and the extracted attribute values.

6. The computer-implemented method of claim 1 , wherein linking the entity references includes:
merging the entity references with records of the corresponding one or more entities within the data system.

7. The computer-implemented method of claim 1 , wherein linking the entity references includes:
inserting the entity references into one of the data system and an external data source based on a comparison of matching scores for the entity references with corresponding thresholds.

Assignees

• IBM

Inventors

• Deshpande Prasad M
• Mohania Mukesh K
• Murthy Karin
• Padmanabhan Deepak S
• Reed Jennifer S
• Schumacher Scott

Classifications

• G06F17/30292 Primary

  Physics · mapped topic

• G06F17/30303

  Physics · mapped topic

• G06F16/33 Primary

  Querying · CPC title

• G06F16/215 Primary

  Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors · CPC title

• G06F16/93

  Document management systems · CPC title