concept feedback in category nlp

appears as: feedback, feedback
Natural Language Processing in Action: Understanding, analyzing, and generating text with Python

This is an excerpt from Manning's book Natural Language Processing in Action: Understanding, analyzing, and generating text with Python.

Companies like to know what users think of their products. So they often will provide some way for you to give feedback. A star rating on Amazon or Rotten Tomatoes is one way to get quantitative data about how people feel about products they’ve purchased. But a more natural way is to use natural language comments. Giving your user a blank slate (an empty text box) to fill up with comments about your product can produce more detailed feedback.

In the past you’d have to read all that feedback. Only a human can understand something like emotion and sentiment in natural language text, right? However, if you had to read thousands of reviews you’d see how tedious and error-prone a human reader can be. Humans are remarkably bad at reading feedback, especially criticism or negative feedback. And customers generally aren’t very good at communicating feedback in a way that can get past your natural human triggers and filters.

At Talentpair, we experimented with matching resumes to job descriptions using the cosine distance between topic vectors for each document. This worked OK. But we learned quickly that we got much better results when we started “steering” our topic vectors based on feedback from candidates and account managers responsible for helping them find a job. Vectors for “good pairings” were steered closer together than all the other pairings.