It is often the case that there is no clear winner discernible from an array of algorithm options. In a sentiment analysis problem, for instance, there are several possible approaches and it is not often clear which to take. You can choose from a Naive Bayes classifier with embedded negations, a Naive Bayes classifier using bigrams, an LSTM RNN, a maximum entropy model, and several other techniques.
If the format and form decision point doesn't help you here—for instance, if you have no requirement for a probabilistic classifier—you can make your decision based on your available resources and performance targets. A Bayesian classifier is lightweight with quick training times, very fast evaluation times, a small memory footprint and comparatively small storage and CPU requirements.
An LSTM RNN, on the other hand, is a sophisticated model...
