Sentimental analyser

I am working on sentimental analyser to identify positive, negative n neutral emotions from the set of user comments from a dataset. As if now I use labelled dataset as a training data and uses ML.net logistic regression algorithm for the results. But I am happy with the results. What would be the best algorithm or way to derive perfect accuracy and also segregate those comments under different pre defined topics

If the dataset has datapoints that are incorrectly labelled, it’s almost impossible to reach perfect accuracy.

Please take up deep learning specialization by Andrew Ng to get a better understanding of what it means to develop a model and how to perform error analysis.