Value Mining: How Entity Extraction Informs Analysis
Download as PPTX, PDF•1 like•1,246 views
The document discusses how entity extraction can inform document analysis by turning unstructured data into a structured data model. It provides a case study of a federal agency that used entity extraction and open analytics tools to analyze over 6,000 documents, extracting entities like people, dates, and topics. This automated the previously manual process, reduced analysis time from hundreds of hours to a few hours, and ensured only relevant documents were included by identifying duplicates and missing metadata. The results provided the agency with key insights for their research.
Value Mining: How Entity Extraction Informs Analysis
- 1. Value Mining: How Entity Extraction Informs Analysis June 2012 | Andrew Strite
- 2. Agenda • Big Data and Document Analysis • Case Study: Federal Agency – Problem Definition – Open Analytics & Entity Extraction – Reporting and Visualization – Results Assessment • Questions
- 3. The Big Data Problem Data is becoming the new raw material of business: an economic input almost on par with capital and labor. “Every day I wake up and ask, „how can I flow data better, manage data better, analyze data better?” Rollin Ford, the CIO of Wal-Mart
- 4. Solution: Document Analysis "Document Analysis refers to computer-assisted analysis of large numbers of documents in order to answer questions about the content of a document set.” Source: http://www.text-tech.com/docanalysis/definition.html
- 5. Document Analysis • The goal is to: – Extract Entities (people, places, things) – Create Associations between entities (in the form of noun-verb-noun), e.g.: • John Doe lives in Washington, D.C • John Doe is married to Jane Doe • John Doe is a Virgo • John Doe traveled to Mexico on July 6th, 2011 • And…
- 6. Document Analysis • Turn Who, What, When and Where into a unified data structure that supports data analytics and visualization. Who When people, organizations, past, present, future facilities, company dates What Where events, summaries, city, state, country, facts, themes coordinate
- 7. Document Analysis Case Study: Federal Agency
- 8. Overview A Federal client produced reports for other DoD components and wanted to know: “Did our reports meet customer needs?” First step: assess historical reporting “What were teams writing about and when?”
- 9. Problem: Unstructured Data • Plenty of raw data, but no way to get at it – 6K+ unstructured documents – 15+ file types – No standard formats • Teams (Who) • Dates (When) • Topics (What) – Some content not relevant
- 10. Early Attempts • Initial client attempts to solve the problem mostly involved manual review – High document volume = labor intensive – Assessing relevance = skilled labor • Total process tied up skilled analysts for hundreds of man-hours. • Manual review prone to error – Incomplete attempts corrupted data
- 11. Solution: Open Analytics • Process to design and implement analytical solutions • Joins open tools and agile engineering techniques • Goal is to enable organizations to quickly deliver smart analysis and enable top line growth
- 12. Mechanism: Infinit.e Infinit.e is a scalable framework for Visualizing Analyzing Retrieving Enriching Storing Collecting Unstructured documents & Structured records
- 13. Infinit.e Concept 20% Structured • Log files • Databases • Apps • Documents • Presentations • Spreadsheets • Meeting notes • Email • IM chats • Reports • Social 80% Unstructured • Entities • Events • Facts • Sentiment Unstructured and Structured Data • Geospatial • Temporal • Themes
- 14. Infinit.e Data Model Duke and Progress announced merger plans in January 2012 Bernanke, 57 said in his testimony price increases “have begun to moderate” after a jump in oil costs Who people, organizations, earlier this year facilities, company What Tablet ownership levels hit 18% in China, the events, summaries, UK and US versus 3% in November 2010 facts, themes When <Incident> past, present, future <uid>20101043423</uid> dates <subject>1 person killed in armed attack by suspected Boko Haram in Maiduguri, Borno, Where Nigeria</subject> city, state, country, <multipleDays>No</multipleDays> coordinate <eventDate>06/04/2011</eventDate> </Incident>
- 15. Applying Infinit.e “What were teams writing about and when?” Open Analytic and Agile Intelligence architecture
- 17. Reporting and Visualization • Queries performed on the data, providing breakouts by team, topic, and dates • Flexible visualization – Built-in visualization framework – Multiple export options
- 18. Finding Value • Over the course of 2.5 weeks, we applied the entity-based data model to our client’s document analysis problem • Major advantages to this approach were: – Agility – Precision – Relevance
- 19. Agility • Automation reduced processing time: – Manual processing time: ~480 hours – Automated processing time: 2-3 hours • Speed enabled iterative development – Extraction adapted alongside analysts’ understanding of data – Positive feedback loop
- 20. Precision • Entity definitions created from original data – Definitions improved based on feedback • Automation ensures uniform application across data set entity1 TOPIC1 TOPIC2 TOPIC1 entity2 entity3 entity1 TOPIC2 entity3
- 21. Relevance • Entity extraction informs quality control – Duplicates identified based on similar entities – Exclude documents based on missing entities – Minimizes risk of data corruption – Reduced need for analyst review Duplicates Missing Meta-Data
- 22. The Results • Extracted entities became key meta-data 6K+ unstructured documents became… …3.5K documents with value to the study
- 23. The Results • Our client was able to complete the research shortly after final extraction • Confidence in methodology and results bolstered the value of recommendations • Considering similar approaches for future projects
- 24. Bottom Line Using document analysis significantly… … reduces the time to ingest data. … cuts right to relevant information. … builds a framework for future analysis.
- 25. Thank You! Andrew Strite www.ikanow.com [email protected] 301.513.1384