“Lovely app, but it deleted the only copy of my recording.”
The sentence contains praise. The incident deserves urgent attention. A sentiment score that averages the two into mild disappointment has missed the important part.
Give consequence its own field
Separate emotional tone from operational impact. Data loss, an inaccessible core workflow, incorrect financial output and a cosmetic annoyance should not compete solely on how angry the reviewer sounds.
For each issue, record what failed, whether a workaround exists, whether the loss is reversible and which users could be affected. Keep the evidence level visible: reported by one reviewer, repeated in a sample, or reproduced under known conditions.
Avoid a fake severity calculator
A numerical score can help organize work, but the inputs remain judgments. Multiplying uncertain frequency by uncertain impact does not create an objective truth.
Use a small set of explained categories instead. For example: prevents the core job; creates recoverable extra work; causes a minor inconvenience. Escalate potentially serious reports for investigation even when the sample is small.
Distinguish research from incident response
When studying a competitor, you usually cannot confirm internal causes or total affected users. Describe the report carefully and avoid turning it into an accusation. When the report concerns your own product, preserve evidence, investigate and communicate appropriately with affected customers.
Positive wording can reflect a loyal customer trying to be helpful. Angry wording can reflect a small issue encountered at a terrible moment. Neither should be ignored, but neither is a substitute for understanding the consequence.
When you review the next batch of feedback, hide the star rating for a first pass and read what actually happened. Then restore the rating as context. You may discover that the most important problem in the pile was asking for help quite politely.