The truth is, this is not brand-new tech, and has been many decades in the making, although Artificial Intelligence (AI) models are currently breaking new ground in language (ChatGPT), art (Dall-E) and even music (Jukebox, Musico).
Whilst we are seeing the first unicorns built on these models (ChatGPT creator Open.ai valued at $29b, Jasper.ai valued at $1.5b), we see huge further opportunity for NLP models to access untapped ROI in the form of human knowledge and sentiment in existing content – news, opinion, reviews, and forecasts.
“Natural language processing (NLP) is a branch of artificial intelligence that focuses on helping computers to understand the way that humans write and speak.”
Why are NLP Models useful?
Once fed with sufficiently large training datasets and skilfully tuned, these models are highly adept at extracting qualified numerical data to develop benchmarks that can accurately compare two or more companies side by side across a range of different criteria, providing unique perspectives into any company’s competitive position against its peers. This is a fundamental part of what we do at Deltabase to flush out opportunities and risks for businesses.
When would we use an NLP Model?
For example, we may wish to analyse investment trends within the commodities sector based on thousands of articles from a news feed like this one:
INPUT:
“China’s Zhejiang Huayou Cobalt Co will invest $147.2 million to build a copper project in the Democratic Republic of Congo (DRC), as it extends the exploration of its mining assets in the country said in a filing to the Shanghai stock exchange on Tuesday.”
Performing this research task manually over a meaningful volume of news articles would incur a great deal of time and cost. Using NLP, however, we can extract information very quickly, consistently, objectively, and at huge scale.
How can NLP extract information from the news article?
NLP models trained to perform Q&A are great at working with the context and meaning of the text to answer questions and, as a bonus, they can also tell us how confident they are that they have found the answer in the text as illustrated:
OUTPUT:
Context: China’s Zhejiang Huayou Cobalt Co has allocated $147.2 million to build a copper project in Democratic Republic of Congo (DRC)
Question: How much will be invested?
Answer: “$147.2 million”, Confidence: 96%
We can apply this technique over thousands of data points and millions of articles to extract datasets like this:
| Investor | investment_usd_m | investment_country | investment_commodity |
| Zhejiang Huayou Cobalt Co | 147.2 | DRC | Copper |
| Southland Steel Fabricators Inc. | 18.0 | USA | Steel |
| Glencore | 500.0 | Brazil | Iron Ore |
Table 1 — Example Commodities Dataset
Notice that our question “Who made the investment” doesn’t need to exactly match the wording of the news article. Artificial intelligence language models are great at comprehending the text – they understand that allocating funds to a project is similar enough to make an investment in that project, even though the specific words differ.
What other ways can NLP models drive business value?
This example provides a small glimpse into how we can use NLP models to extract business value. Other examples include:
- Intelligent classification of reviews by topic and sentiment.
- Constructing models of company ownership from descriptions of M&A activity
- Analysis of public domain company reports and news articles data to build a picture of how a company performs in aspects of the social element of ESG.
In conclusion, we saw how the application of artificial intelligence NLP models can help us to rapidly extract value from the written word and accelerate research. At Deltabase, we use this technique and others to derive precisely quantified and qualified information from resources like news, reviews and social media.
If you want to explore how Deltabase products can help you to leverage the power of AI to accelerate company research and benchmark companies against their competitors, then get in touch to schedule a quick call with our team to find out more.