How Computers Rank Financial News for Analysts

This patent describes a computer system that automatically collects financial news, groups similar stories, and ranks them for financial analysts using advanced text analysis and machine learning.

What it covers

This system first "ingests" (takes in) news articles from many different sources. It then "extracts named entities" (like company names or people) from each article to create a "one-hot vector" for initial grouping. The articles are then "clustered" based on these vectors. For each cluster, a "representative news article" is chosen. The system then uses a "machine learning model" with "character embeddings" and a "convolutional layer followed by a max-pool layer" to understand the meaning of words and sentences in these representative articles. Similar clusters are then "merged" based on their semantic meaning. Finally, the system generates a "set of ranked clusters," which it "digitally displays" in a user interface, allowing analysts to interact with the ranked news. For example, a financial analyst could see a cluster of news about a specific company's earnings report, with the most important articles ranked at the top, and then filter these results.

What it doesn't cover

• —Does not cover news ranking systems that do not use 'one-hot vectors' generated from 'named entities' for initial clustering.
• —Does not cover systems that do not employ 'character embeddings' to create 'word representations' for representative articles.
• —Does not cover systems that do not use a 'convolutional layer followed by a max-pool layer' to generate input representations for articles.
• —Does not cover ranking news articles for general audiences, as it specifically targets 'financial analysts in the capital markets'.
• —Does not cover ranking articles within a cluster solely based on publication date without considering 'trustworthiness and linking volume' of the news sources.

The clever bit

The novelty lies in combining specific natural language processing techniques, like named entity extraction for initial clustering and character embeddings with convolutional neural networks for deeper semantic understanding, to create a hierarchical, ranked view of financial news tailored for analysts.

Why it matters

Financial analysts need to quickly process vast amounts of news to make informed decisions. This system automates the complex task of sifting through, categorizing, and prioritizing financial news. By providing relevant, clustered, and ranked information, it helps analysts save time and potentially identify critical market movements or risks faster, which is crucial in fast-paced financial markets.

Real-world examples

• 1.S&P Global Market Intelligence platform
• 2.Bloomberg Terminal news feeds
• 3.Refinitiv Eikon news analysis
• 4.FactSet Research Systems news aggregation