Sensitive Words
The sensitive-word system intercepts forbidden words in business texts such as payment titles, product names, merchant names and announcements. Menu: System Management → Sensitive Words.
Three capabilities, each with its own role: the filter policy (master switch) under Payment Platform Configuration, the word library under Sensitive Words, and hit records (audit).
Filter Policy (Master Switch)
The entry is the sensitive-word config card in Payment Platform Configuration:
| Setting | Description |
|---|---|
| Enable sensitive-word filtering | Master switch; turn it off for overseas deployments that don't need it — when off, all text validation passes |
| Echo hit words in errors | When on, error messages include the specific word that hit — this risks probing of the word library; off by default |
| Record hit audit | Writes a hit record on interception (on by default); the data source of the hit-records page |
Word Library
- Query: word, category, status;
- Categories: politics / pornography / violence / advertising / custom;
- Match modes: contains (hits wherever the word appears in the text — the common choice) / exact (hits only when the whole text matches exactly);
- Add / edit fields: word (within 64 characters), category, match mode, status, remark.
The library uses Simplified Chinese
Before matching, the system automatically converts traditional to Simplified Chinese and full-width to half-width, so entering Simplified Chinese in the library also covers traditional and full-width variants — no need to store multiple copies.
Online Trial Check
The trial-check area above the list: enter a piece of text and click "Check" to immediately see whether it hits and which words hit; tick "write hit record" to send trial results into the hit-records page (useful for drills and verifying the audit chain). Before launching a new library, trial-check a few typical variants to verify the matching effect.
Hit Records
Every interception leaves a trace:
- Query: hit word, scenario, source, merchant number;
- List: hit word, text excerpt, scenario, source, client IP, merchant, time;
- Detail: the full text excerpt, merchant / app / operator, IP, request path and other traceability fields.
Scenario marks which business the text came from (payment title, product name, merchant name, store name, announcement, agreement, trial check, and ten-plus others); source marks which client triggered it (admin panel / merchant panel / open payment API / admin mini program).
Handling advice
High-frequency hits of the "payment title / product description" kind are usually legitimate merchant wording hit by an over-broad word — adjusting that word in the library (switching it to exact match or deleting it) works better than repeatedly explaining to the merchant; for real violations, handle them with the risk-control blacklist.
Related Links
- Payment Platform Configuration — the filter-policy master switch
- Payment Security and Risk Control — post-hit interception actions and blacklists