楽天レビューシグナル — レビュー信頼度チェック

You cannot prove it from a page, and this does not try to. It measures patterns that tend to accompany paid or solicited reviews and shows them as a 0-100 figure with the contributing signals listed, so you can read the reviews with the right amount of suspicion. The panel carries a note saying the result is heuristic and does not judge any individual review or shop.

It starts at 100 and subtracts weighted penalties. Sponsorship wording is the largest deduction, followed by short or duplicated text, date concentration, review-for-points wording, and last a five-star skew that only applies alongside many short reviews. Above 75 the panel reads 信頼サインは良好, from 45 to 74 it warns of caution signs, and below 45 it flags several suspicious ones.

It needs at least three reviews it can actually read out of the page. Category listings, search results and items with one or two reviews therefore show nothing. The panel is inserted just above the review section, and the analysis reruns by itself when the page content or the URL changes, which covers moving between items without a full reload.

Two things are declared: the storage permission and host access to *://*.rakuten.co.jp/*, nothing wider. Storage caches the most recent result under a single key in chrome.storage.local so the toolbar popup can show it again; the host access lets the content script read the review text, dates and ratings that Rakuten has already rendered.

No. The phrase lists that drive the sponsorship, incentive and generic-text signals are written for Japanese review wording, and the panel labels are in Japanese too. The content script runs only on rakuten.co.jp, so other Japanese marketplaces are out of scope. Star, date and duplicate-text signals do not depend on language, but the score as a whole is tuned for Japanese pages.