Why does your product search miss a Chinese name or an exact model code?

Separate names from identifiers and attributes

A customer may search for a product in Traditional Chinese, an English brand-neutral description, or a precise model code copied from a box. Those are different kinds of request. If the catalogue stores only one display title, an English translation may hide the term the customer actually knows, while a description stuffed with codes may produce irrelevant matches. Give each product variant a stable internal identifier, a manufacturer model code where one exists, bilingual display names, approved alternate names, category and structured attributes. Keep the exact original code as data, including its meaningful hyphens and leading zeroes. Do not silently replace an old code when packaging changes; record a documented alias if the same product is genuinely known by both codes.

Start by asking staff for failed real-world queries, with private customer details removed. Separate “the item exists but search missed it” from “the item is not stocked”, “the customer used an obsolete code”, and “the product was hidden because unavailable”. These cases need different fixes. One missing synonym should lead to a catalogue edit, while a systematic failure on Chinese names points to an indexing or query-processing problem. The search results should show the same variant and availability information that checkout will use. Otherwise a better search engine merely helps customers reach an item they still cannot buy. Treat internal naming and translation review as part of product data maintenance, with an owner for corrections.

Design Chinese matching deliberately

Many English search implementations split on spaces and then match whole words. That assumption is a poor fit for Chinese product names, which normally have no spaces between words. The Unicode text segmentation annex says reliable Chinese word boundaries require dictionary lookup or another tailored mechanism; its default word boundaries are not enough for all such uses. This does not prescribe one vendor or one tokenizer. It means the team must test how its chosen index handles a query such as “電熱水壺” against a title that contains a longer phrase, and how it handles a shorter query such as “水壺”. Character n-grams, language-aware tokenization and curated synonyms are possible tools, each with a different balance of recall and irrelevant results.

Do not apply every expansion to every field. A broad Chinese name match can help discovery; a model-code field usually needs exact or carefully defined prefix matching. If the shopper enters “AB-012”, a near match to “AB-021” may be a dangerous recommendation if accessories only fit one model. Show an exact-code hit first, label related alternatives clearly, and make any code normalization explicit. English case differences may be harmless for a descriptive name, but punctuation and digits in an identifier can carry meaning. Search should use ranking rules that reflect these differences: exact identifiers, then exact approved names, then aliases and broader descriptive matches. Review the result set with staff rather than assuming a high match count means success.

Normalize for comparison without rewriting the catalogue

Visually similar strings are not always the same sequence of characters. The W3C Character Model explains that canonical Unicode normalization and case folding affect comparisons, while compatibility normalization can erase distinctions in meaning or appearance. A practical catalogue can keep the original approved names and codes for display, then create a separate search representation according to a documented field-specific policy. For a descriptive name, that might include a canonical normalization and appropriate case handling. For an exact model code, the shop should decide which differences are truly interchangeable with the manufacturer or product owner, and test every rule. Do not automatically strip all punctuation, convert every script or collapse full-width characters in identifiers on the assumption that it is harmless.

Traditional and Simplified Chinese forms introduce a further decision. A shopper may type a Simplified form while the catalogue shows Traditional Chinese, but automatic character conversion is not always one-to-one or sufficient to identify the intended product. Treat cross-script matching as an optional discovery route with reviewed terminology, not as a silent rewrite of official product names. Likewise, a Cantonese market term and a formal product name may describe the same appliance, yet an indiscriminate synonym can make two different items look interchangeable. Keep the official title visible in results, record why an alias exists, and let staff remove a misleading alias without editing the stock identity. The search index can be rebuilt; the authoritative catalogue and product identifiers must remain understandable after that rebuild.

Use a small, controlled query sheet

Imagine a retailer listing a kettle as “不鏽鋼電熱水壺”, with an English name “stainless steel kettle” and manufacturer code “KT-007”. The owner and staff create a query sheet: the full Chinese name, “水壺”, the English name, “KT-007”, a code with one digit wrong, an approved local nickname, and a query for a different appliance. For each query, record the expected first result, other acceptable results, and results that must never appear as an exact match. “KT-007” should identify the correct variant even if another product contains “007” in its description. A wrong digit should not be silently corrected into a purchase recommendation. “水壺” may return several kettles, with filters for capacity and availability.

Run the same sheet against desktop and mobile search, typed input and pasted code, and both language interfaces. Check whether the query is retained when there are no results and whether the page offers a useful next step: remove one filter, search by model code, browse a category or ask staff to verify a discontinued item. Do not invent a substitute model when fit or electrical specification matters. The W3C's accessibility guidance on status messages is relevant to dynamic result counts: people using assistive technology should be told when results update without having focus moved unpredictably. A catalogue search test therefore includes result relevance, understandable empty states, keyboard use and status announcement, while leaving the earlier screen-reader search-status question as its own accessibility topic.

Make relevance review a catalogue routine

After launch, review examples of failed searches, searches followed by a different product click, and exact-code searches that produced multiple plausible variants. These signals are prompts for investigation, not proof that the ranking is wrong: a customer may be researching an item the shop does not sell. Keep a short change log for synonyms, product names and ranking rules, with before-and-after queries. Assign a merchandiser to approve aliases and a developer to maintain indexing behaviour. When a model is discontinued or replaced, state that relationship explicitly in the catalogue instead of secretly remapping its code to a new item. If a recommendation depends on compatibility, have product staff confirm it.

Set an acceptance target based on the query sheet rather than claiming an unsupported industry benchmark. For example, every approved exact model query in the test set should resolve to its intended variant, while every intentionally wrong-code query should remain visibly unconfirmed. Review ambiguous Chinese terms with staff who know the local vocabulary. Repeat the sheet after catalogue imports, search-engine upgrades and major synonym edits, because a change that improves one query can worsen another. The owner should be able to answer why a particular result appeared, who approved the alias, and how to correct a harmful match. That is more useful to a shopper than a search box that merely responds quickly.

Record the reason for each no-result decision as carefully as a successful match. If a query contains both a Chinese noun and a model code, the code should usually narrow the candidates while the noun helps confirm the category; staff should test cases where those clues conflict. For example, a kettle term paired with a code belonging to a rice cooker should not be presented as an exact kettle result. Show the conflict or offer separate searches so the customer can correct it. Product codes printed on invoices may differ from manufacturer codes printed on packaging, so map those fields explicitly and label them in staff tools. A shopper can then ask a meaningful question about either code. Keep search analytics at an aggregate, privacy-conscious level and avoid storing unnecessary personal text from free-form queries. A weekly review of a few representative misses is more actionable than a single undifferentiated search-success percentage, especially when the catalogue changes often.