Collect companies' Earnings Call Transcripts by web scraping and then apply NLP methods to their contents to find the best sentiments analyser.
To test this project, we will apply Microsoft Corporation earnings call as an example.
-
US Corporate Earnings Call Transcripts can be reached from https://news.alphastreet.com/
-
Stanford NER is a Java implementation of a Named Entity Recognizer, the package we used in this project can be download from https://drive.google.com/drive/folders/1mS0y92w65f4R7u6dNP8KFFAtdmNmI-v9?usp=sharing the file named 'stanford-ner'
- FinBERT
- PySentiment
- TextBlob
- Vader
Finbert enjoys high accuracy and an acceptable error rate at the same time.
WordCloud plot shows that 'currency', 'growth','revenue','cloud' are the most common words mentioned in the latest 12 Microsoft earning calls.
FinBERT model introduction:https://paperswithcode.com/paper/finbert-a-pretrained-language-model-for