Financial Document Tagging and Analytics

MarketPsych NLP Engine for LLM-based Topic, Entity, Event, and Sentiment Classification (NLP-as-a-Service)

Download PDF

Why NLP Engine?

Used by Top Firms for Detailed NLP Analytics

Large hedge funds, central banks, regulators, and media firms use MarketPsych’s NLP service as an outsourced NLP team, allowing their staff to focus on more complex downstream tasks.

Your Text Goes In, Analytics Come Out

Clients access MarketPsych’s NLP Engine through manual document entry, API, or on-premises installations. Files are ingested in real-time and processed within seconds by layered, fine-tuned LLMs. Analytics templates are available for parsing and analyzing news, social media, transcripts, filings, internal emails, chats, and brokerage research.

Fine-tuned LLM classifiers

Tags and classifications include:

Entities: 20+ types (organizations, commodities, locations, etc.)

Topics: 1,000+ (accounting, litigation, sustainability, etc.)

Events: 4,000+ (dividend payments, factory construction, etc.)

Sentiments: Positive/negative/neutral for finance, ESG, commodities

Emotions: 14 types (anger, fear, optimism, etc.)

Sentiment Engine for Natural Text (SENT)

Figure: Client text is ingested in real-time and returned in JSON format with each sentence tagged.

MarketPsych NLP Engine Use

Tagging Entities in Global News

Yahoo! Finance requires a robust analytics suite to connect company names across global media in 28 languages. The named entity recognition (NER) labels from MarketPsych’s NLP Engine enable Yahoo! Finance to provide relevant and timely business news in 28 languages on their website and within the iPhone Finance App.

MarketPsych’s NLP Engine identifies millions of entities across twenty categories. Recognized entities include companies, currencies, commodities, central banks, cryptocurrencies, events, facilities, locations, people, sectors and industries, political and military institutions, and much more. Point-in-time named entity recognition (NER) is conducted on historical documents, accurately identifying top company names back to 1890.

Manage Potential Risks - Screen for Negative Media, Counterparty Risk, and Internal Chats

A large financial regulator utilizes MarketPsych’s NLP Engine to swiftly identify negative media reports. Specifically, the regulator monitors media articles and rumors questioning the creditworthiness of companies within their regulatory purview. By employing watchlists and alerts on global media flows, the regulatory team receives timely notifications of relevant news.

Parsing and Organizing Complex Documents

Developing document parsers for various types of financial documents and reports is a complex and challenging task. Document templates are updated every few years, some jurisdictions lack standards, and the process of parsing and organizing the source text is time-consuming.

MarketPsych’s team has addressed this challenge for numerous text types. A large hedge fund employs the NLP Engine to parse and organize text from various financial documents into structured files with standardized analytics.

Create Custom Data Feeds & Uncover Hidden Insights

Results are delivered in a data science-friendly JSON format, and a graphical user interface is available for generating customized charts and data feeds.

The analytics allow users to address queries based on their input text, such as:

− Compare Tesla sentiment in Chinese news versus EU news

− Plot social media sentiment surrounding the iPhone 16 launch

− Rank companies by the number of references to “robot” in earnings call transcripts and display each sentence

− Monitor sentiment regarding competing products like Ozempic, Wegovy, Rybelsus, and Mounjaro (below)

Figure: Daily volume of mentions to GLP-1 medications (bars) and their average media sentiment (lines)