Blog — 14 Jul, 2023
A Professor Leverages Kensho NERD to Analyze Thousands of Textual Documents for Trading Strategy Development
Highlights
THE CLIENT: A large Asia-based university
USERS: A finance professor
AI applications are delivering enormous benefits to users by helping them quickly scan large numbers of unstructured datasets to uncover insights previously not available given the time required to digest a massive volume of information.
Textual data, such as transcripts from earnings calls, can be a rich source of information revealing insights not readily apparent in a company’s financial reports. However, the volume of information in these unstructured types of documents can be overwhelming for any analyst that needs to follow a portfolio of companies, and there are several challenges that make it difficult to zero in on relevant information. For example, companies can be:
- Mentioned in numerous documents, not just their own.
- Have names that include commonly used words, making them hard to identify.
- Referred to by acronyms, ticker symbols or previously used names.
Ensuring that organizations are maximizing their use of data can be a challenge. The quick rise of artificial intelligence (AI) and machine learning (ML) in all facets of business has many asking how they can capitalize on these technologies to enhance their decision-making.
A professor at this Asia-based university wanted to evaluate how media co-coverage (i.e., where multiple firms are simultaneously mentioned in the same news article) impacts market reactions and stock price movements. He realized that it would be a massive undertaking to identify co-mentioned firms across thousands of textual documents to establish causality, and wanted to see if there was an AI capability that would make this a much easier task.
Pain Points
The professor at this large Asia-based university wanted to identify co-mentioned companies across thousands of textual documents to establish causality between co-mentioned and stock market reactions. To support this, he needed a capability that would enable him to:
- Identify company names mentioned in earnings call transcripts and other textual data with speed and accuracy.
- Link the mentioned companies with stock prices and other company-related data to analyze whether co-coverage drove market behavior, or if there was another explanation.
The professor realized this was an impossible task without the use of an AI application to speed up the analysis. He was aware that Kensho Technologies was an S&P Global company and had been accessing vast amounts of the firm's data to train and develop ML learning algorithms to build a wide range of data-centric solutions. He reached out to S&P Global Market Intelligence ("Market Intelligence") to learn more about the AI capabilities that were available.
The Solution
Specialists from Market Intelligence described Kensho NERD (Named Entity Recognition and Disambiguation), a cutting-edge ML learning system that identifies organizations, people, places and events that appear in text ― from news to emails to call transcripts ― and links each entity that is found to its matching entry in the S&P Capital IQ and Wikimedia[1] knowledge bases. Powered by Kensho AI, Nerd's context-aware natural language processing technology unlocks the full potential of textual data by linking it to these existing sources of structured knowledge that represent two of the world's richest databases. Kensho NERD would enable the professor to:
|
|
Extract entity information from text |
NERD uses technology that combines state-of-the-art entity recognition techniques with context-aware linking. It resolves ambiguous entity mentions, including abbreviations, acronyms, historical names and aliases. It is also a probabilistic framework. Each possible match is given a score to indicate the likelihood of the text being an entity (i.e., the “NER score”) and the confidence that the link to the S&P Global proprietary company ID is correct (i.e., the “NED score”). This enables users to identify cases with the highest-quality links. |
|
|
Automatically link textual data to stock prices and other relevant information |
NERD is the only technology on the market trained on business-related documents and specifically optimized to identify financial entities. NERD has learned from the patterns in millions of documents, from news articles to earnings call transcripts. It also leveraged NERD's Wikimedia engine, which links to a broad knowledge base containing tens of millions of entities. Many of these documents are hand-labeled by Kensho’s domain experts in the financial and business worlds to teach NERD to extract accurate information from this type of text. |
|
|
Get started in minutes |
NERD can be used via a simple REST API for real-time and batch processing, enabling users to quickly and efficiently analyze unstructured text. For example, they can investigate suppliers and competitors mentioned in company filings and transcripts, or find mentions of entities relevant to other workflow needs. |
Key Benefits
Kensho NERD was exactly what the professor was looking for to expedite his analysis, and he subscribed to the offering. He is now able to:
- Uncover the companies, subsidiaries and other financial organizations that appear in co-covered textual data and link them to stock prices and other information in the S&P Capital IQ database to assess the impact of co-coverage on market behavior.
- Count on a highly reliable capability, since NERD has been specifically, and uniquely, optimized to extract financial entity information from text documents.
- Utilize an API that is research friendly, enabling him to load a large volume of textual data and receive output in a speedy and structured manner.
- Obtain ongoing support from the NERD product team to fully utilize the capabilities.
[1] Wikimedia is a global movement whose mission is to bring free educational content to the world.