A method to forecast financial performance of companies is provided. The method comprises retrieving from a database news articles related to a number of companies published within a predefined time period. The news articles are classified as either ESG articles or non-ESG articles and then vectoriz
A method to forecast financial performance of companies is provided. The method comprises retrieving from a database news articles related to a number of companies published within a predefined time period. The news articles are classified as either ESG articles or non-ESG articles and then vectorized. A subset of relevant non-ESG articles are selected. The ESG and selected non-ESG articles are fed into a sentiment scoring model, which generates sentiment scores for the companies over the predefined time period. The sentiment scores are fed into an ESG forecast model along with historical market data and ESG data related to the companies. The ESG forecast model forecasts the financial performance of the companies in relation to ESG policies.
대표청구항▼
1. A computer-implemented method of forecasting financial performance of companies, the method comprising: using a number of processors to perform the steps of: retrieving from a database, through an application programming interface via JavaScript Object Notation format query, a number of news arti
1. A computer-implemented method of forecasting financial performance of companies, the method comprising: using a number of processors to perform the steps of: retrieving from a database, through an application programming interface via JavaScript Object Notation format query, a number of news articles related to a number of companies published within a predefined time period;vectorizing the news articles;creating an artificial neural network comprising a sentiment scoring model and an environmental, social, and governance forecast model implemented as nodes in the artificial neural network;feeding the news articles into the sentiment scoring model;generating, by the sentiment scoring model implemented in an artificial neural network that employs semantic orientation or Vader sentiment analysis, a number of sentiment scores for the companies over the predefined time period;feeding the sentiment scores, and historical market data and environmental, social, and governance data related to the companies into the environmental, social, and governance forecast model; andforecasting, by the environmental, social, and governance forecast model, financial performance of the companies in relation to environmental, social, and governance policies comprising ratings assessing the companies' behavior and policies regarding their environmental performance, social impact and governance issues. 2. The method of claim 1, further comprising: classifying the news articles as either environmental, social, and governance articles or non-environmental, social, and governance particles; andselecting a subset of relevant articles from the non-environmental, social, and governance articles. 3. The method of claim 2, wherein the subset of relevant non-environmental, social, and governance articles comprises a top number of non-environmental, social, and governance articles most similar to the environmental, social, and governance articles. 4. The method of claim 2, wherein the subset of relevant non-environmental, social, and governance articles have a similarity to the environmental, social, and governance articles above a specified threshold. 5. The method of claim 2, wherein selecting the subset of relevant non-environmental, social, and governance articles is performed via cosine similarity scoring. 6. The method of claim 1, wherein the sentiment scores are on a scale of −1 to 1. 7. The method of claim 1, wherein the historical market data comprises stock trading volume for the companies. 8. The method of claim 1, wherein the environmental, social, and governance data comprises at least one of: environmental score;social score;governance score;carbon emission volume; orgreenhouse gas emission volume. 9. The method of claim 1, wherein the news articles are extracted in text format. 10. The method of claim 1, wherein vectorizing the news articles is performed via term frequency-inverse document frequency technique. 11. A system for forecasting financial performance of companies, the system comprising: a storage device configured to store program instructions; andone or more processors operably connected to the storage device and configured to execute the program instructions to cause the system to: retrieve from a database, through an application programming interface via JavaScript Object Notation format query, a number of news articles related to a number of companies published within a predefined time period;vectorize the news articles;create an artificial neural network comprising a sentiment scoring model and a environmental, social, and governance forecast model implemented as nodes in the artificial neural network;feed the news articles into the sentiment scoring model;generate, by the sentiment scoring model implemented in an artificial neural network that employs semantic orientation or Vader sentiment analysis, a number of sentiment scores for the companies over the predefined time period;feed the sentiment scores, and historical market data and environmental, social, and governance data related to the companies into the environmental, social, and governance forecast model; andforecast, by the environmental, social, and governance forecast model, financial performance of the companies in relation to environmental, social, and governance policies comprising ratings that assess the companies' behavior and policies regarding their environmental performance, social impact and governance issues. 12. The system of claim 11, wherein the processors further execute program instructions to: classify the news articles as either environmental, social, and governance articles or non-environmental, social, and governance articles; andselect a subset of relevant articles from the non-environmental, social, and governance articles. 13. A computer program product for forecasting financial performance of companies, the computer program product comprising: a computer-readable storage media having program instructions embodied thereon to perform the steps of: retrieving from a database, through an application programming interface via JavaScript Object Notation format query, a number of news articles related to a number of companies published within a predefined time period;vectorizing the news articles;creating an artificial neural network comprising a sentiment scoring model and a environmental, social, and governance forecast model implemented as nodes in the artificial neural network;feeding the news articles into sentiment scoring model;generating, by the sentiment scoring model implemented in an artificial neural network that employs semantic orientation or Vader sentiment analysis, a number of sentiment scores for the companies over the predefined time period;feeding the sentiment scores, and historical market data and environmental, social, and governance data related to the companies into the environmental, social, and governance forecast model; andforecasting, by the environmental, social, and governance forecast model, financial performance of the companies in relation to environmental, social, and governance policies comprising ratings assessing the companies' behavior and policies regarding their environmental performance, social impact and governance issues. 14. The computer program product of claim 13, further comprising instructions for: classifying the news articles as either environmental, social, and governance articles or non-environmental, social, and governance articles; andselecting a subset of relevant articles from the non-environmental, social, and governance articles. 15. A computer-implemented method to generate sentiment scores, the method comprising: using a number of processors to perform the steps of: retrieving from a database, through an application programming interface via JavaScript Object Notation format query, a number of news articles related to a number of companies published within a predefined time period;creating an artificial neural network comprising a sentiment scoring model and a environmental, social, and governance forecast model implemented as nodes in the artificial neural network;classifying the news articles as either environmental, social, and governance articles or non-environmental, social, and governance articles;vectorizing the environmental, social, and governance articles and non-environmental, social, and governance articles;selecting a subset of relevant articles from the non-environmental, social, and governance articles;feeding the environmental, social, and governance articles and the subset of relevant non-environmental, social, and governance articles into the sentiment scoring model; andgenerating, by the sentiment scoring model implemented in an artificial neural network that employs semantic orientation or Vader sentiment analysis, a number of sentiment scores for the companies over the predefined time period comprising ratings assessing the companies' behavior and policies regarding their environmental performance, social impact and governance issues. 16. The method of claim 15, wherein the subset of relevant non-environmental, social, and governance articles comprises a top number of non-environmental, social, and governance articles most similar to the environmental, social, and governance articles. 17. The method of claim 15, wherein the subset of relevant non-environmental, social, and governance articles have a similarity to the environmental, social, and governance articles above a specified threshold. 18. The method of claim 15, wherein the sentiment scores are on a scale −1 to 1. 19. The method of claim 15, wherein the environmental, social, and governance articles and non-environmental, social, and governance articles are extracted in text format. 20. The method of claim 15, wherein vectorizing the environmental, social, and governance articles and non-environmental, social, and governance articles is performed via term frequency-inverse document frequency technique. 21. The method of claim 15, wherein selecting the subset of relevant non-environmental, social, and governance articles is performed via cosine similarity scoring. 22. A system for generating sentiment scores, the system comprising: a storage device configured to store program instructions; andone or more processors operably connected to the storage device and configured to execute the program instructions to cause the system to: retrieve from a database, through an application programming interface via JavaScript Object Notation format query, a number of news articles related to a number of companies published within a predefined time period;create an artificial neural network comprising a sentiment scoring model and a environmental, social, and governance forecast model implemented as nodes in the artificial neural network;classify the news articles as either environmental, social, and governance articles or non-environmental, social, and governance articles;vectorize the environmental, social, and governance articles and non-environmental, social, and governance articles;select a subset of relevant articles from the non-ESG articles;feed the environmental, social, and governance articles and the subset of relevant non-environmental, social, and governance articles into the sentiment scoring model; andgenerate, by the sentiment scoring model implemented in an artificial neural network that employs semantic orientation or Vader sentiment analysis, a number of sentiment scores for the companies over the predefined time period comprising ratings that assess the companies' behavior and policies regarding their environmental performance, social impact and governance issues. 23. The system of claim 22, wherein the subset of relevant non-environmental, social, and governance articles comprises a top number of non-environmental, social, and governance articles most similar to the environmental, social, and governance articles. 24. The system of claim 22, wherein the subset of relevant non-environmental, social, and governance articles have a similarity to the environmental, social, and governance articles above a specified threshold. 25. A computer program product for generating sentiment scores, the computer program product comprising: a computer-readable storage media having program instructions embodied thereon to perform the steps of: retrieving from a database, through an application programming interface via JavaScript Object Notation format query, a number of news articles related to a number of companies published within a predefined time period;creating an artificial neural network comprising a sentiment scoring model and a environmental, social, and governance forecast model implemented as nodes in the artificial neural network;classifying the news articles as either environmental, social, and governance articles or non-environmental, social, and governance articles;vectorizing the environmental, social, and governance articles and non-environmental, social, and governance articles;selecting a subset of relevant articles from the non-environmental, social, and governance articles;feeding the environmental, social, and governance articles and the subset of relevant non-environmental, social, and governance articles into the sentiment scoring model; andgenerating, by the sentiment scoring model implemented in an artificial neural network that employs semantic orientation or Vader sentiment analysis, a number of sentiment scores for the companies over the predefined time period comprising ratings assessing the companies' behavior and policies regarding their environmental performance, social impact and governance issues. 26. The computer program product of claim 25, wherein the subset of relevant non-environmental, social, and governance articles comprises a top number of non-environmental, social, and governance articles most similar to the environmental, social, and governance articles. 27. The computer program product of claim 25, wherein the subset of relevant non-environmental, social, and governance articles have a similarity to the environmental, social, and governance articles above a specified threshold.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.