Users treat negative feedback as emotional expression; its semantics are imprecise

What it is

“Not interested” is a learning label in the product and, under a finger, often irritation, surprise, embarrassment, anger at the ad that just ran. One act covers many internal states. Affective negative feedback means the reject channel will be used as an emotional outlet, and its semantics cannot be treated as a stable proposition about an item or a class. Used as a label, it is noisy.

Performative preference is acting to change the next display. Here the act is to send the present emotion out, not necessarily to change the model.

Why it happens

The interface collapses a complex discomfort into one key. The fewer and more salient the keys, the more easily they take emotions unrelated to the content: slow load, too many ads, a privacy guess that landed, the day’s mood. Those pulses enter learning and write a momentary context as a long-term negative feature. If default scope is the class, one pulse can pause the whole class — the next layer’s over-response is often affective semantics meeting a default that is too large.

Precise semantics need extra friction or structure: pick a reason, pick a scope, allow undo. Friction will lower use; it will also split “I want to change the model” from “I want to swear.” Without structure the channel has high recall and low precision, and the model cannot tell whether the reject was of the content, the ad, or the system itself.

Studying it

Immediately after a reject, ask the reason (content / repetition / ad / guessed too well / just annoyed), and compare those reports with whether the person still rejects the class later. Independent variables: whether a reason is required, whether undo exists in a short window, whether the key is split from “report” and “hide ads.” Dependent variables: distribution of reasons, correlation with later stable rejection, rate of writing an emotional pulse as a class-level feature.

Do not take high use as channel health. High use may be the outlet working. Report label precision: what share of this tap still holds a day later.

Where it stops holding

People with low literacy or access needs struggle with reason-picking; friction will shut them out of the channel. Safety reports must be split from taste rejects, even if both look like “the key I hit when angry.” This entry is imprecise semantics. It does not set a default scope, and it does not treat immediate visibility as precision — visibility means the channel is still alive, not that the label is accurate.

Applying it

• Split “not interested,” “too often,” “this is an ad,” and “report” into different acts. Do not let one key eat every discomfort.
• Give a short undo window after reject so an emotional pulse can be taken back before it writes into long-term features.
• Check: sample rejects, and a day later ask “still do not want this class?” Heavy denial means the channel is collecting emotion. Split the acts, add undo, and see whether denial falls.

Related

• Same group: L6.13.1 With only positive signals the system cannot tell "not interested" from "never saw it" · L6.13.2 Negative feedback must state its scope: this item or this class · L6.13.3 Negative feedback must produce an immediately visible change, or users will judge it useless and stop · L6.13.5 Over-responding to a single negative can wipe out a whole class of content
• Nearby: L6.09 Feedback Loops and Preference Entrenchment · L6.06 Inferred Preferences and Their Limits · L3.13 User Feedback Loops on Generation Quality
• Search terms: affective negative feedback · noisy not-interested · emotion versus preference