How do I split contents in a text that would include two or more different themes (context) in NLP?

First reading the question I thought this is very easy and I started searching and trying out some libraries (i.e. nltk, and spacy). Here are my attempts and clearly showing that none of them works. I was thinking that these libraries will generate a parsing tree or a context-free grammar to detect sentences, but they are splitting based on a certain characters such as periods or commas.

So the other solution is to learn a classifier that learns the boundaries of sentences. A simple solution can be to use random forest classifier or SVM to serve as a binary classification task for each token. But the missing part is the data to train the model on and what to consider as labels (a BIO tagging scheme is a good idea here). I include two papers here that can be a good starting point for you to first read their methodology and use the datasets they are mentioning to do the task.

• Du, Jinhua, Yan Huang, and Karo Moilanen. "AIG Investments. AI at the FinSBD task: Sentence boundary detection through sequence labelling and BERT fine-tuning." Proceedings of the First Workshop on Financial Technology and Natural Language Processing. 2019.
• Rudrapal, Dwijen, et al. "Sentence boundary detection for social media text." (2015).