We don't often think about confirmation bias in data; data is supposed to be objective evidence. But it's not, always. "Ratings" are SO subjective and even the individual thought that provides the "logic" behind choosing a number is very different from one person to another. The real meat is in the text, and the TEXT will tell a different story. I know people who give five stars to make people (AND EVEN COMPANIES) feel good about themselves, and I know people who NEVER give five stars because "nobody's perfect." Same with the low ratings - often, a one-star and a two-star is the same thing. A one-star can be "I'm sorry, but I really hate this product" or "The product sucks, but I want to hurt YOU." A two star is generally more informative and likely to come with thoughtful text. But I don't know how you can assess that with numbers alone. I used to say that data science requires an interdisciplinary approach, and with massive amounts of human textual feedback, you would do well to add a linguist and a psychologist to the team.