Together with Leiden University we build a binary NLP (natural language processing) ‘interesting-not interesting’ classifier. Training of this model is based on existing emails/RFQs (marked as interesting by client). This model will classify the new emails as interesting or not, based on what it has seen before (your own preferences).
We use different techniques such as TF/IDF; parameters of the model are adjusted in such a way that KPI’s are best for this particular case. The set is shuffled, and the training of the model is repeated multiple times until the best parameters are chosen.
By using the power of machine learning, the computer can decide in split seconds if the new incoming mail is interesting or not, saving time and delivering new business opportunities.
As an initial classifier, we used the Random Forest classifier. Random Forest is a multiple decision tree algorithm, using numerous decion trees to predict if a word or combination of words is interesting or not. For this case it is the best machine learning technique available. By combining hundreds of separate decision trees this will lead to higher precision and better predictions of the classifier.
For more information or a demo meeting, please contact us.