Perceptual Feature Extraction

Feature Extraction mostly uses Natural Language Processing tools to encode the perceptions of users from the reviews into values for a system. Information Systems (or computers in general) struggle to understand our messy language, so we need to transform it into values, often numerical, so it can process them correctly.

In the text of my thesis, I talk about Explicit and Implicit Perceptual Features, this is from the "human interpretation" point of view. Explicit algorithms have features such as "acting" and "directing", that we can relate to and understand what they mean, for example if "acting" has a high value, it means it is good. The implicit algorithms are usually based in mathematical and statistical modelling, aided by machine learning for the processing. This means that the resulting dimensions have no name, and we cannot interpret what it means if their values are higher or lower. In the prototype we only use implicit features, to allow for more features than the usually selected in explicit methods, check the code here for more.

To go more into detail about feature extraction: