Most recent post:
09 Oct 2020
In [1] Wang, Lu and Zhai present a supervised machine learning model to solve a problem they name Latent Aspect Rating Analysis. To explain the problem, let’s start with some context. Sites like Amazon, TripAdvisor or Audible have a review system where the user provides a rating and a review text.
The overall rating might be too vague, so it’s desirable to break down this rating per what is called aspect. The list of aspects depend on the use case: for logding, it could be room, cleaningless or location; for Audible it has performance and story. The model tries to infer these implicit (latent) ratings from only the overall rating and the review text.
The authors define a model that learns from training data and show promising results with test data. In this post we’ll focus on the theory presented in [1].