I wrote to Solita’s data blog about a text analytics project. The goal was to automate manual classification of funding applications. Here is the link to blog.
FastText in text analytics project – Summary
The primary tool for text analytics was Facebook’s FastText package. The neural network behind FastText works especially well for languages like Finnish as it is able to detect different kinds of word endings.
The biggest challenges were structuring the data and class imbalance. The results were promising and our model beat simple methods such as selecting the most common label.