How to aggregate sentiment labels in a long text

Good morning everyone,
I’m using the Roberta-based sentiment model (mrm8488/distilroberta-finetuned-financial-news-sentiment-analysis) to assign a sentiment category (positive, neutral or negative) to purely financial news, which typically have a medium/long text. The model works very well on individual sentences, however I’m looking for a clever way to assign a sentiment to the whole news, aggregating the result obtained in each individual sentence. Is there a standard way to do this? Taking the most frequent class doesn’t seem like a good idea because I would probably get the neutral class as the most frequent class often. Did you ever face a similar problem?

Thanks a lot for your suggestions!

1 Like