Hands-On Machine Learning for Algorithmic Trading

Pros and cons

The benefits of LSI include the removal of noise and mitigation of the curse of dimensionality, while also capturing some semantics and clustering both documents and terms.

However, the results of LSI are difficult to interpret because topics are word vectors with both positive and negative entries. There is also no underlying model that would permit the evaluation of fit and provide guidance when selecting the number of dimensions or topics.