Our AI timeline
Thomson Reuters has been innovating for its customers from day one - which for some customer segments goes as far back as the 1800's. Initially, technology was used to collect, organize, and enhance information for its customers. Later, it would employ artificial intelligence (AI) to improve the customer's ability to find the information they needed. Today, we use AI to better understand our customers and surface information and insights they need when they need it. Thomson Reuters, through its different businesses, has had a formal applied research and development group since 1992.
Follow our AI @ TR timeline to discover our innovation journey.
Westlaw was one of the first online legal research services. Attorneys used dumb terminals to dial up to a mainframe. The content was limited (disk space was expensive) and the search language simplistic.
Westlaw marked the beginnings of technology-driven innovation in many ways for legal sector.
Scientist Spotlight: Howard Turtle
As chief scientist, Howard Turtle helped found one of the first R&D groups at Thomson Reuters. He is a nationally known scholar in search engine technologies, having developed a formal retrieval model based on Bayesian Inference Networks that formed the basis of the University of Massachusetts’ INQUERY Retrieval System and of West Publishing’s natural language search product, called WIN. Howard led the legal R&D group until 1996. Howard Turtle retired from Syracuse University in 2016.
WIN (Westlaw is Natural)
Westlaw Is Natural (WIN) was the first commercially available search engine with probabilistic rank retrieval. Howard Turtle led the effort after completing his PhD at UMass. This was an innovation milestone for legal research because prior to that most search engines only supported Boolean term & connectors.
Scientist Spotlight: Peter Jackson
Peter Jackson was one of the founders of research and development (R&D) at Thomson Reuters. Peter joined Lawyers Cooperative Publishing (LCP) and formed the natural language processing (NLP) group in 1995. In 1998, Peter assumed leadership of the legal R&D group. Peter became Chief Research Scientist & Vice-President, Technology in 2005 at Thomson.
History Assistant was a large scale natural language processing (NLP) system that analyzed case law documents, extracted parties, judges and built the appellate chain for a particular case. The system found history relationships between court decisions by using a combination of information retrieval and machine learning techniques to link each new case to related documents that it may impact.
PeopleCite & Profiler
PeopleCite and Profiler extracted entities from American case law documents to create a knowledge base of judges, attorneys, and expert witnesses with links to all their cases and biographies. Machine learning enabled those systems to analyze millions of documents, a scale far beyond what could be done manually. [Info Today]
CaRE - Classification and Recommendation Engine
Using an ensemble of machine learning algorithms, CaRE has been used widely across the company to classify legal, tax and finance documents to large taxonomies, e.g., CaRE is used to classify millions of headnotes to a KeyNumber taxonomy with more than 90,000 categories. CaRE later formed the basis of ResultsPlus (2003) - a very successful document recommendation system. It is still in use today (2019).
ResultsPlus was a very popular document recommendation solution in Westlaw. Based in part on CaRE, the system made contextually relevant recommendations of secondary law documents, Key Numbers, briefs and more alongside search results. The system incorporated: natural language generation of summaries for briefs; user behavior analysis to enable personalized recommendations; and dynamic ranking of selections based on data from real-time A/B testing.
Built on the work done for the Profiler project to link attorney and judge names, this system identified law firms and companies, and used semantic parsing and discourse analysis techniques to infer the relationships among judges, attorneys, law firms, their roles, and the companies they represent. The metadata was stored in a data warehouse which in turn fed the Firm360 reports.
Dexter is a machine learning (ML)-based named entity extraction and resolution (NER)system focusing on news and legal content. It is used in many products including Reuters Insider (2011).
Westlaw Medical Litigator provided legal researchers with immediate, "one-stop" access to information about medical terms, procedures and devices related to medical malpractice, personal injury, and device liability. In addition, it provided an understanding of related medical issues, health care professionals and expert witnesses. It included natural language support and disambiguation tagging.
Thomson Reuters provides one of the largest and most diverse set of Public Records in the United States. Concord enables searches where there could be thousands of "John Smith"s, connecting the correct records among hundreds of millions of records, and many more possible connections. It was, perhaps, the first statistics-based record linking and resolution solution of this scope in the public records domain.
Building on all our previous experience, a wide array of AI technologies were leveraged to solve a wide set of challenges. It used machine learning (ML), clustering, classification, usage log analysis, citation network analysis, topic modeling, and natural language generation – the veritable "kitchen sink" of AI. The result was WestSearch. WestlawNext set a new standard for legal research solutions. AI enabled the system to go beyond traditional search.
Reuters Insider used CaRE classification and Dexter entity extraction to connect transcripts of live news shows to video; this enabled searching video-based news.
NewsPlus is a content recommendation platform used in Westlaw and Eikon. The recommendation algorithms incorporated information from multiple sources to retrieve, filter, and prioritize news given the context of a specific user and application. It analyzes, de-duplicates, and groups/clusters the content. Like its predecessor, ResultsPlus, it used a hybrid approach of content and collaborative filtering.
Checkpoint - Broadside
Intuitive Search on Thomson Reuters Checkpoint
Checkpoint Broadside applied many of the technologies and approaches used in WestlawNext to power the new "Intuitive Search" capability in Checkpoint, our market-leading research solution for tax and accounting professionals.
Magnet analyzed SEC filing for deviation in language that may merit further review. An analysts could then review and provide deeper insights into related plans, initiatives, prospects and challenges; in effect, expanding the news coverage of that company.
Reuters Tracer and Social Data Platform (SDP)
Separating Real News from Fake in 40 Milliseconds
A platform and tool created for Reuters journalists to monitor Twitter for breaking events. Tracer filters out social media noise and identifies potential breaking news events. It utilized natural language processing (NLP), content classification, clustering and machine learning. A special Tracer feed is provided to Eikon as a completely automated real-time news feed.
DPA - Data Privacy Advisor
Data Privacy Advisor
Data Privacy Advisor incorporates a next-generation question-answering feature built in partnership with the Thomson Reuters and IBM Watson.
Adverse media capabilities enable our analysts to perform adverse media screening from tens of thousands of news sources. In addition to extracting risk signals, these capabilities also perform concordance from unstructured sources. This capability helps analysts in performing ongoing background checks that support Anti-Money Laundering regulations. In particular, this new capability will perform interactive person name disambiguation and identify text/documents that potentially contain evidence of financial crimes. This service was being done manually by analysts. The solution searches news articles and leverages NewsPlus tagging capabilities. This capability is used in World-Check One: Media Check.
AutoMuni - Automated Municipal Bond Pricing
The accurate evaluation of approximately a million bonds daily is a big challenge. The muni valuation method is quite manual. Such an approach is not only time consuming and costly, but also only a small portion of bonds can be accurately evaluated due to the restriction of resources. The AutoMuni system helps scale the valuation process by using intelligent, machine learning, algorithms that can price the entire fixed income universe automatically and efficiently.
Inferno is a data ingestion and workflow tool to help analysts refresh content within the World-Check database. This helps to improve the accuracy and reliability of World-Check which is crucial in keeping the competitive advantage. It mines news feeds and then uses various techniques, including machine learning and natural language processing (NLP), to enhance the data ingestion workflow, increasing the speed and accuracy of identifying required updates to World-Check data.
Westlaw Edge: WestSearch Plus
WestSearch Plus on Westlaw Edge
WestSearch Plus answers customers' questions posed in natural language. Behind the scenes, it mines the rich analytical material in our headnotes as the source for answers. It uses editorial guidelines to divide Headnotes into frames/intents. Then it classifies both answers and questions (mined from the query logs) to those intents. WestSearch Plus uses search strategies based on questions and intents to assemble a headnote candidate pool and uses natural language processing (NLP) & discourse features in XGBoost model to classify/score answers.
Westlaw Edge: Litigation Analytics
Litigation Analytics on Westlaw Edge
Lawyers now have all the analytics from past cases at their fingertips and can shape their litigation strategy with Westlaw Edge's Litigation Analytics.
Many AI methods were used including deep learning. It extracts information from federal dockets, identifying names and relationships of all parties. It identifies case topics and adds all this information to a knowledge graph. The knowledge graph is continually updated. Customers use natural language to explore this knowledge graph. When returning results, the AI generates text narratives explaining the generated visualizations.
Westlaw Edge: KeyCite Overruling Risk
KeyCite Overruling Risk on Westlaw Edge
KeyCite is our market leading case citator system. It is important to identify cases that are negatively impacted by an overruling decision, and to flag them as possibly impliedly overruled cases that need further careful examination. The AI-powered solution identifies those cases using machine learning algorithms.
The problem is formulated as a binary classification of candidate cases into those that are negatively impacted (and possibly impliedly overruled) and those that are not (and hence are still safe to be relied on in a legal investigation). Our classifier uses a variety of features, including metadata information about the cases, citation paragraphs in the overruling and affected cases, as well as headnotes in the overruled case.
Deep Learning Center Launched in Boston
Thomson Reuters launched a deep learning center in the heart of Boston, Massachusetts, very appropriately addressed at 1 Thomson Place. This marked a growing commitment within the organization to expand capabilities across core technology domains. Deep learning methodologies can greatly add to the proficiency of related fields like natural language processing or image processing and analysis. The implications of this field are substantial across operational efficiencies as well as discovering new insights across key industry segments.
Checkpoint Edge introduced an entirely new feature called Concept Markers and made further enhancements to the Intuitive Search algorithms that assist and guide the research process. It helps users refine and/or elaborate on their queries to find answers. Natural language processing (NLP) and machine learning (ML) technologies were used to enable these features.
Westlaw Edge : Quick Check
Westlaw Edge Quick Check™ In the Making
Attorneys perform hours of legal research so that they can provide their clients with the best possible representation. They are under constant pressure to do their best work as efficiently as possible.
Quick Check is a Westlaw Edge feature that, given a legal document, finds additional supporting authority to research. This helps our customers complete their legal research faster and have a higher degree of confidence in their work. It also enables them to do a thorough analysis on their opponent’s work and in turn help them be more prepared for their clients.
Our first partnership with the Legal Tracker team looked to enhance the capabilities of the invoice review process using AI technologies. Using machine learning classification techniques to predict if an invoice narrative violates specific billing guidelines, we launched a feature to detect block billing on invoices and invoice items describing non-reimbursable work such as travel, or internal communication.
Released Practical Law Dynamic Search
State-of-the-art AI system designed to find the answers to free-form questions in Practical Law. It searches content-based on complex questions, as opposed to keyword queries, which poses unique and significant challenges. PLDS allows users to ask for exactly what they need without having to guess at a suitable keyword query.
Onvio: Predictive Tax Code Mapping
Onvio introduced a new AI-based capability that helps customers with the tedious steps involved in the classification of accounts for tax purposes. The predictive tax code AI model that powers the capability turned a very consuming manual process into an automated one, which allows our customers to spend more time on higher revenue-generating tasks.
"I just used the predictive tax code mapping feature when importing accounts and balances from a spreadsheet for a new client, and the tax codes predicted were 90% accurate. It was amazing."
-Satisfied customer of Onvio Trial Balance
AI platform launched to support enterprise AI solutions
AI Platform launched in 2022 to facilitate time-to-market for data science and machine learning teams. Core to the team’s principles is the implementation of data and model governance standards and the operationalization of the model release process. This internal platform is built upon both existing and new tools, continuously reimagining them to support wider organizational needs.
Concluding 2022, the platform consisted of:
Data Service, AI Annotation, AI Workspaces, Model Registry, Model Deployment, Model and Data Drift Monitoring, Bias Monitoring
Updated Practical Law Dynamic Search
An updated version of PLDS was released that not only retrieves relevant answers more often, but nearly eliminates non-relevant answers from appearing in the top 5 results.
In its on-going effort to make editors and their workflows more efficient, Thomson Reuters is introducing greater amounts of Artificial Intelligence (AI) and Machine Learning (ML) into its production processes. A recent example is the UK Judicial Cases Editorial Treatment Process, which predicts the importance of an incoming case, based on properties like court name, subject area, keywords, and even level of anonymization required, if any.
The automated process leverages AI in order to queue the work for editors and to help ensure that the most important cases are processed first by our editorial team. State of the art ML operations have enabled this project to be successfully integrated into the associated production workflow and to deliver significant cost savings.
Legal Tracker Advanced
Legal Tracker Advanced brings the power of AI analytics and industry insights to empower law departments to strengthen spend management and operations. We delivered interactive data visualization for the Rate Increases dashboard in Legal Tracker Advanced, which facilitates users to make the data-driven decisions for allocating spending.
Thomson Reuters Labs at 30: Celebrating Three Decades of Innovation
At TR Labs, we are committed to driving progress and making a positive impact on the world. We are proud of the innovation we have delivered over the past 30 years, and we look forward to continuing to push the boundaries of what is possible in the next 30 and beyond.
AI @ TR - We're Just Getting Started!
Today, Thomson Reuters is scaling innovation for its customers to new heights. How will AI transform our industries tomorrow? AI @ TR is working on that answer today!