Brain News Topics Analysis with LLM
Product Summary
The Brain News Topics Analysis dataset exploits an internal and customized large language model to monitor specific topics and their sentiment within the financial news flow for stocks.
For example, an investor may want to identify all news related to the topic “innovation” for a set of companies and track their sentiment with respect to each specific topic. Similarly, another investor can be interested in tracking all news related to the topic “risks for the company” and their sentiment.
Metrics Provided for Each Stock and TopicFor each stock and each topic three metrics are provided using the news published within a given time interval:
1. The volume of news relevant to the topic2. The buzz, which measures the variation in the amountof news that are published for each topic.3. A sentiment score for the specific topic, rangingfrom -1 to +1.
All metrics are calculated based on the news published within a given time interval, e.g. the past 7 days that the model identifies as relevant for each topic.
List of Monitored TopicsThe topics monitored by the Brain Large Language Model in the news flow are the following:
1. Contracts, Licenses, and Partnerships2. Financial Results3. Investor Asset Transactions and Positions4. Governance and Management related Events 5. Innovation6. Price variations7. Rating and valuation estimates8. Risks for the company9. Legal
CoverageThe dataset covers the largest 1000 US stocks approximately corresponding to the Russell 1000 Index.
Metrics Provided for Each Stock and Topic
For each stock and each topic three metrics are provided using the news published within a given time interval:
1. The volume of news relevant to the topic
2. The buzz, which measures the variation in the amount
of news that are published for each topic.
3. A sentiment score for the specific topic, ranging
from -1 to +1.
All metrics are calculated based on the news published within a given time interval, e.g. the past 7 days that the model identifies as relevant for each topic.
List of Monitored Topics
2. Financial Results
3. Investor Asset Transactions and Positions
4. Governance and Management related Events 5. Innovation
6. Price variations
7. Rating and valuation estimates
8. Risks for the company
9. Legal
Coverage
The dataset covers the largest 1000 US stocks approximately corresponding to the Russell 1000 Index.
• Financial news are collected every few minutes from various financial media sources.
• A dedicated Large Language Model evaluates the relevance with respect to each monitored topic. If the news item is relevant the model assigns a sentiment with respect to the specific topic.
• Daily, for each stock, the news relevant to each topic are aggregated over various time periods into both a buzz and a sentiment for the topic.
• Repetition of news is taken into account during the aggregation phase.
• The dataset files can be shared via FTP or an AWS S3 bucket.
• Daily historical data starting from 2017 will be available for testing.