How much do individual ratings influence future ratings in networks?
This explores whether the ratings people leave are independent judgments or whether each rating tugs the ones that come after it — and how much of that pull compounds as it travels through a connected system of products, reviewers, and recommenders.
This explores whether the ratings people leave are independent judgments or whether each rating tugs the ones that come after it. The corpus is fairly unanimous that ratings are not independent — but the size of the effect is the interesting part. Moe and Trusov decomposed star ratings into three pieces: a product's baseline quality, a social-dynamics component driven by prior ratings, and noise — and found the social-dynamics piece is real but *small* per rating. The catch is that small effects compound: each nudged rating becomes the prior that nudges the next, so distortion accumulates over time even though no single rating moves the needle much. High variance in opinion can eventually dampen the drift, but the default direction is contamination, not correction Do online ratings actually reflect independent customer opinions?.
The mechanism behind the nudge is worth seeing up close, because it isn't simple herding. When people post in public after reading negative reviews, they lower their own ratings — even when their actual experience was positive — because negative reviewers read as smarter and more discerning. Private raters show no such shift. So part of what looks like "influence" is self-presentation: people are rating for an audience, not just recording a preference Why do online reviewers publish negative ratings despite positive experiences?. Layer onto that the finding that the same person rates the same item differently across sessions — swinging by multiple stars from temporal mood, anchoring, and personal rating style — and a single rating starts to look less like a fixed data point and more like a reading taken under shifting conditions [[explicit-user-ratings-are-noisy-temporal-inconsistency-and-rater-idiosyncrasy-co].
The "in networks" part is where the answer gets surprising. How much one rating influences future ratings depends on *which network* the product sits in. Frequently-bought-together and co-viewed recommendation graphs route products to different audiences with different prior expectations, and that routing — not the product — decides whether ratings on linked items converge or diverge Do different recommender types shape opinion convergence differently?. The recommender is effectively choosing who gets to influence whom. Seen this way, recommendation feeds aren't neutral plumbing; they're persuasion infrastructure where network topology drives opinion convergence and rating contamination compounds through the same feedback loops How do recommendation feeds shape what people see and believe?.
There's also a selection-bias twist that amplifies all of this. Only people with strong opinions bother to rate at all — small participation costs produce U-shaped distributions where the lukewarm middle stays silent and the average drifts away from true quality Why do people bother writing online ratings at all?. And the algorithms that learn from these ratings can lock the distortion in: ranking systems that don't explicitly model selection bias converge on degenerate equilibria that amplify their own past decisions Why do ranking systems need to model selection bias explicitly?.
So the honest answer: per rating, the influence is *small* — but "small" is the wrong frame, because ratings live in feedback loops. Each rating becomes a prior for the next, the network decides whose prior reaches you, and the algorithm trains on the result. The thing you didn't know you wanted to know: the corpus suggests the danger isn't that any one rating sways you, but that a system with no countervailing force will quietly compound tiny social nudges into a number that no longer reflects the product at all.
Sources 7 notes
Moe and Trusov decomposed ratings into baseline quality, social-dynamics influence, and error, finding that prior ratings meaningfully affect subsequent ones. These effects have both immediate sales impact and long-term compounding effects through future ratings, though high opinion variance can eventually dampen the distortion.
Posters systematically reduce their ratings in public when exposed to negative reviews, even with positive personal experience—because negative reviewers appear more intelligent. Private raters show no such shift, revealing a self-presentational mechanism tied to multiple-audience communication.
Amatriain et al. found that the same user gives substantially different ratings to the same item across sessions, shifting by multiple stars. This noise stems from temporal inconsistency, rater-specific biases, and anchoring effects—making ratings reflect both preference and rating-behavior rather than stable preference alone.
Research shows that frequently-bought-together and co-viewed recommendation networks produce different opinion convergence patterns. The mechanism: each recommender type attracts different audience segments with different prior expectations, shaping both who sees products together and how they rate them.
Research shows recommendation systems operate as political actors: feed weights influence producer behavior, network topology drives opinion convergence, and automation enables targeted persuasion at population scale. These effects compound through rating contamination and selection biases.
Lafky's experiments show raters care about both buyers and sellers rather than purely one or the other. Small participation costs create U-shaped distributions where only strong-opinion raters engage, biasing average ratings away from true quality.
YouTube's multi-objective ranker uses MMoE for conflicting objectives and a shallow position tower to remove selection bias from training data. Without both mechanisms, models converge on degenerate equilibria that amplify their own past decisions.