Intelligent Enterprise

Richard Brown of Thomson Reuters delivered an illuminating talk, "News, Blogs, and Full-Tick Logs: Innovative Approaches to Quantitative and Event-Driven Trading," Tuesday at Gartner's Event Processing Summit. The summit and the Event Processing Technical Society symposium now underway feature many such use cases, descriptions of low-latency transformation and analysis of high-volume data and event streams as applied to diverse business problems. Brown's case study, which looked at exploiting information from unstructured sources to support financial-market trading, was of particular interest (to me) due to its combination of events, text sources, and sentiment analysis.

Brown's talk profiled the Reuters NewsScope Sentiment Engine, which "processes a stream of Reuters English language news items, producing sentiment data for a list of customer determined target companies." You can read more on line, including about theapplication of technology from Infonic to assess author sentiment. The application assigns sentiment scores to selected words and phrases but goes beyond simple numbers, per Brown's presentation, to assess topical relevance — Is a given item substantively about the subject company — and novelty — How unique (and fresh) is the news? In the financial sector, it is especially important to distinguish updates from new news.

Brown was speaking at an event processing conference because "integration of myriad of data sources requires a sophisticated event processing framework," that is, one that delivery low-latency (low-delay) processing of high-volume data streams from diverse sources. Complex event processing's raison d'être is the inability of traditional data warehousing and analytical methods to accommodate these needs. Thomson Reuters works with CEP vendor Streambase.

Beyond scoring the author sentiment and market sentiment, "reaction based on information in the article/text," Reuters provides metainformation that support article aggregation to company, sector, and market levels and has been studying the trading value implied by news propagation pathways. As in other applications of text data mining, findings are combined with independent analytical processes. Since Reuters operates in a trading domain, these include technical analysis of financial instruments and study of company and market fundamentals.

Part of the analysis is generation of 45 predictive indices that the company, working with quantitative research firm AlphaSimplex, determined apply to foreign-exchange and stock volatility. The point is to assess event-based risk. The violence index, for example, represents the percentile of violence topicality relative to "comparable" historical periods. A next phase of research will study applicability to equities, for instance, how a spike in the violence index affects defense stocks. You'll find more substantive information on the Web page of principal Andrew W. Lo, director of MIT's Laboratory for Financial Engineering, than you will on AlphaSimplex's site. Brown may be willing to provide interested readers with a copy of a Reuters NewsScope Event Indices technical paper.

There are additional analytics touchpoints. Reuters deploys subsidiary ClearForest's text-analytics software at client sites to support information extraction. On the publisher side, Reuters is promoting use of ClearForest's Calais Web service to tag articles with standard metadata. Calais is actually free; you can get a key for API use or paste your own text into a sample application to see how it works.

In sum, Richard Brown's Gartner summit talk was an illuminating example of the integration of leading-edge information management and analysis technologies to meet a set of very demanding business challenges.

E-MAIL  |