Understanding News Using the Bloomberg Knowledge Graph

News and the global capital markets are inextricably linked. Recent developments in machine learning, knowledge graphs, and language technology have enabled increasingly intelligent ways to obtain a market advantage based on real-world events. This talk details how Bloomberg uses these technologies to quickly understand and respond to major world events in order to predict when or how breaking business news will move markets – and why.

Transcript

1. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Understanding News Using the Bloomberg Knowledge Graph Big Data Innovators Gathering (BIG) @ The Web Conference 2019 May 14, 2019 Edgar Meij Graph Analytics Team Lead, AI Engineering
2. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Bloomberg • Bloomberg is just finance, right? ◦ A technology company ◦ Our strength and focus is data and analytics ◦ Provide community tools ◦ Both creator and consumer of news ◦ Increasing use of and contributions to open-source
3. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Bloomberg by the numbers • Founded in 1981 • 325,000+ subscribers in 170 countries • Over 19,000 employees in 192 locations • More News reporters than The New York Times + Washington Post + Chicago Tribune • Over 5,500 software engineers, including more than 200 engineers working on data science problems and nearly 100 data science experts
4. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Ticker Bloomberg Social Media Websites News Filings Mails/ Chats Survey Data Analyst Reports
6. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Market Impact Indicator
7. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Sentiment Analysis
8. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Sentiment Analysis
9. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Social Velocity
10. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Question Answering
11. What is the finance ecosystem? Market: Stock exchanges Buy side Investment Management firms (Vanguard, Azimut) Sell side Investment banks and other brokers (HSBC) Pension funds Wealth managers Insurance companies Sovereign Wealth Funds (Norway’s “Oil Fund”) Buy side Fund Rating agencies Central Banks Credit rating agencies Investment consultants Government and companies Sell side analysts Savers (Companies, charities, governments, individuals - me and you)
12. © 2018 Bloomberg Finance L.P. All rights reserved. FX has the most daily notational volume Vastly more structured products and derivatives than equities 50,000 40,000 30,000 20,000 10,000 03 04 05 06 07 09 08 10 11 12 13 14 0 Billions in USD Equities Bonds 5,000 4,500 4,000 3,500 3,000 2,500 2,000 1,500 1,000 500 0 Equities Bonds FX Commodities Billions in USD Municipal Bonds Govt + Corp Equities Structured Products Derivatives Sources: WCAUUS Index, http://www.sifma.org/research/statistics.aspx Bonds have more market value than equities
13. © 2018 Bloomberg Finance L.P. All rights reserved. News moves markets
14. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. SEC Announcement First Bloomberg Headline New York Times story ~12% …not all sources are the same
15. © 2018 Bloomberg Finance L.P. All rights reserved. Detection Extraction Interpretation Presentation New text arrives Extract structured data from text Relate new data to existing data Inform the user on what happened
16. © 2018 Bloomberg Finance L.P. All rights reserved. Document ingest and search
17. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Document ingest and search C US Equity HSBA LN Equity Banking (BNK) Federal Reserve (FED) United States (US) Mark Costiglio
18. © 2018 Bloomberg Finance L.P. All rights reserved.
19. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Over $136 billion was wiped out in minutes A disaster in the White House is huge
20. © 2018 Bloomberg Finance L.P. All rights reserved. Detection Extraction Interpretation Presentation
21. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Extraction from documents
22. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. [Yang, Irsoy, Rahman 2018] [Tsai, Roth 2018] How Bill O’Reilly leaving Fox fired up O’Reilly Auto Parts’ stock
23. Detection Extraction Interpretation Presentation
24. © 2018 Bloomberg Finance L.P. All rights reserved. JPMorgan Chase & Co. JPMorgan Chase (merged 2000) Bank One (acq. 2000) Chase Manhattan Bank (merged 1996) Chemical Bank (merged 1991) Chemical Bank (reorganized 1988) Manufacturers Hanover (merged 1961) Bank of the Manhattan Company (est. 1799) Chase National Bank of the City of New York (est. 1877) Chase Manhattan Bank (merged 1955) J.P. Morgan & Co. (formerly Morgan Guaranty Trust) (merged 1959) Banc One Corp. (merged 1968) City National Bank & Trust Company Farmers Saving & Trust Company First Chicago NBD (merged 1995) First Chicago Corp. (est. 1863) Guaranty Trust Company of New York (est. 1866) J.P. Morgan & Co. (The House of Morgan) (est. 1895) Louisiana’s First Commerce Corp. NBD Bancorp. (formerly National Bank of Detroit) (est. 1933) Washington Mutual (founded 1889) Providian Financial (acq. 2005) Dime Bancorp, Inc. (acq. 2002) Bank of United of Texas (acq. 2001) H.F. Ahmanson & Co. (acq. 1998) Great Western Bank (acq. 1997) Bear Stearns (est. 1923, acq. 2008) The Chemical Bank of New York (est. 1823) Citizens National Bank (est. 1851, acq. 1920) Corn Exchange Bank (est. 1852, acq. 1954) New York Trust Company (acq. 1959) Texas Commerce Bank (est. 1866, acq. 1986) Manufacturers Trust Company (acq. 1905) Hanover Bank (est. 1873) JPMorgan Chase Bear Stearns Financial named entity extraction
25. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. bbKG Universe • Entities include: ◦ Companies ◦ Industries, business segments, products and brands, etc. ◦ Financial instruments, including stocks, bonds, funds, etc. ◦ People ◦ Currencies ◦ Geographical locations, including countries, cities, factories/assets, etc. ◦ Literals • Relationships and properties include: ◦ Identifiers, including FIGI, CUSIP, etc. ◦ Multi-lingual aliases ◦ Fundamentals, such as market cap, revenue, earnings per share, debt, dividends, etc. ◦ People properties, including compensation, education, etc. ◦ Supply chain relationships and structural relationships, such as subsidiaries ◦ Industry hierarchies and company-industry membership relationships ◦ Issuers relationships, which connect companies to their financial products ◦ Employment relationships, which connect companies to people (boards, c-suite, etc.) ◦ Location relationships, which connect companies and assets to cities/countries, etc.
26. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Toyota Motor Tohoku Plant Cental Motor Miyagi Plant Kanto Auto Workers Iwate Plant Toyota Motor Hokkaido Plant March 11, 2011: Earthquake in Japan
27. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Avis Budget Group Inc Toyota Motor Corp Ford Motor Co *Automobiles Seating & Trim Hino Motors Ltd. Adient PLC *Automobiles Merck Tadanobu Nagumo Douglas G. DelGrosso Yasuhiko Ichihashi Kenneth C. Frazier Company Industry Person executiveOf hasIndustry supplierOf James Hackett
28. © 2018 Bloomberg Finance L.P. All rights reserved. Detection Extraction Interpretation Presentation
29. © 2018 Bloomberg Finance L.P. All rights reserved. Zev Siegl Starbucks 1971 Gordon Bowker Jerry Baldwin founder of founder of founder of date founded “In 1971, Siegl, Bowker and Baldwin established the Starbucks Coffee Company.” Explaining Knowledge Graph relationships [Voskarides, Meij, de Rijke 2017]
30. © 2018 Bloomberg Finance L.P. All rights reserved. [Voskarides, Meij, Reinanda, Khaitan, Osborne, Stefanoni, Kambadur, de Rijke, 2018] Representation Learning for Knowledge Graph Paths RNN RNN RNN sum ℝ' concat MLP-o Score s 4 5 s 4 a s 4 b t 4 a t 4 7 Feature vector … … RNN RNN sum Query fact Paths from source entity Paths from target entity
31. © 2018 Bloomberg Finance L.P. All rights reserved. Representation Learning for Knowledge Graph Paths • q: founderOf(Microsoft, Bill Gates) • f: profession(Paul Allen, Programmer) Bill Gates B&M G. Foundation c t1 Melinda Gates Microsoft Software Paul Allen Programmer 1975-04 1955-10 1964-08 Lanai 1994-01 Jennifer Gates c t2 CEO 1975-04 2000-01 spouse spouse founderOf industry founderOf profession profession dateFounded dateOfBirth dateOfBirth cerem onyAt marriageDate parentOf parentOf founderOf founderOf leaderOf leadership leaders from to Figure 3: Graph with a subset of the facts that are enumer- ated for the query fact spouseOf (Bill Gates, Melinda Gates). The entities of the query fact are shaded. Learning procedure. We train a network that learns the scor- ing function u(f , f ) end-to-end in mini-batches using stochastic
32. © 2018 Bloomberg Finance L.P. All rights reserved. Representation Learning for Knowledge Graph Paths • q: founderOf(Microsoft, Bill Gates) • f: profession(Paul Allen, Programmer) Bill Gates B&M G. Foundation c t1 Melinda Gates Microsoft Software Paul Allen Programmer 1975-04 1955-10 1964-08 Lanai 1994-01 Jennifer Gates c t2 CEO 1975-04 2000-01 spouse spouse founderOf industry founderOf profession profession dateFounded dateOfBirth dateOfBirth cerem onyAt marriageDate parentOf parentOf founderOf founderOf leaderOf leadership leaders from to Figure 3: Graph with a subset of the facts that are enumer- ated for the query fact spouseOf (Bill Gates, Melinda Gates). The entities of the query fact are shaded. Learning procedure. We train a network that learns the scor- ing function u(f , f ) end-to-end in mini-batches using stochastic
33. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. Automated News
34. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. In summary • Bloomberg aims to provide data, information, analytics, and insights for the entire finance ecosystem, with severe latency and accuracy constraints • Leveraging (un)structured alt data has become an essential component of modern trading strategies • Ranking and explaining relevant connections is essential when generating alerts based on market-moving news and events
35. © 2018 Bloomberg Finance L.P. All rights reserved. © 2019 Bloomberg Finance L.P. All rights reserved. More Info: emeij@bloomberg.net / @edgarmeij Thank You! https://www.TechAtBloomberg.com/NLP