Synonym Mapping: When Shoppers Use the Wrong Words for Your Products

A shopper searching for sneakers on a store that sells trainers will get zero results, and the store will conclude nobody wants sneakers. The shopper wanted exactly what the store sells, in different words.

Synonym mapping is the work of teaching a store's search engine that different words mean the same product. Shoppers do not use your vocabulary. They use theirs, and the gap between the two is where zero-result searches come from. A store that sells trainers will get searches for sneakers, running shoes, and kicks. A store that sells a sofa will get searches for couch and settee. A store that sells a jumper will get searches for sweater, and in the US the same shopper will search for a sweater and mean a jumper, while in the UK they mean something else entirely. The search engine that only matches the words in the product title will return nothing for most of these, and the shopper will conclude the store does not carry what they want. The fix is a synonym list, maintained per category, that maps the shopper's words to the store's words before the query runs.

Start from the zero-result report, not from a dictionary

The zero-result report is the list of queries that returned nothing, and it is the most honest document in the store. It shows the words shoppers actually typed, in the order they typed them, with the count beside each. The top of that list is the synonym work: the queries with volume that returned nothing are almost always vocabulary gaps, not catalog gaps. A query for sneakers with 400 searches a month and zero results is not a missing product. It is a missing synonym, and it costs one line in a mapping file to fix.

Do not build the synonym list from a thesaurus. A thesaurus will tell you that couch and sofa are synonyms, which is true and useless, because it will also tell you that a settee is a synonym, which is true in the UK and confusing in the US. Build it from the store's own data: the zero-result queries, the search terms in the analytics, and the words customers use in support tickets and reviews. The shopper's vocabulary is in the store's own logs, and it is the only vocabulary that matters.

Related reading on this site: RelevanceRail and Zero Results Pages Are a Catalog Report, Not an Error Page cover the neighbouring problems in more depth.

Map in one direction, and keep the map short

A synonym map should be directional: sneakers maps to trainers, not the other way around, because the store's canonical term is trainers and the mapping exists to catch the shopper's word and translate it. Bidirectional maps create loops and false matches, where a search for trainers starts returning sneakers the store does not carry in that sense. Keep the map per category, because the same word means different things in different aisles: a jumper is a sweater in clothing and a cable in electronics, and a single global synonym for jumper will pollute both.

Keep the list short and reviewed. A synonym file with 5,000 entries is not a strategy, it is a landfill, and nobody will maintain it. The working set is usually under 200 mappings per store, covering the top queries in the zero-result report and the top variant terms per category. Review it monthly, against the same report, and delete the mappings that never fire. A mapping that never fires is either wrong or unnecessary, and either way it is noise.

Test the mapping against the shopper, not the dictionary

The test is simple: take the top twenty zero-result queries from last month, apply the synonym map, and check whether the results now shown are products the shopper would actually want. A mapping that returns results is not enough. It has to return the right results, because a shopper who searched for sneakers and got hiking boots will leave just as fast as a shopper who got nothing, and with less patience.

The second test is the conversion rate on mapped queries against unmapped queries. If mapped queries convert at or above the store average, the mapping is working. If they convert below it, the mapping is returning the wrong products and needs narrowing. The synonym map is a living file, not a project, and the store that treats it as a one-time setup will watch it decay as the catalog and the shoppers' language both move on without it.

Frequently asked questions

What is synonym mapping in site search?

Teaching the search engine that different words mean the same product, so a search for sneakers finds trainers, and a search for couch finds sofas. The mapping translates the shopper's words into the store's vocabulary before the query runs.

Where do you find the synonyms a store needs?

In the zero-result report, the list of queries that returned nothing. The top queries with volume that returned nothing are almost always vocabulary gaps, not catalog gaps, and they are the synonym work in priority order.

Should synonym maps be bidirectional?

No. Map in one direction, from the shopper's word to the store's canonical term, and keep the map per category. A global bidirectional map creates loops and false matches, where one word means different things in different aisles.

How do you know if a synonym map is working?

By the conversion rate on mapped queries against unmapped queries. If mapped queries convert at or above the store average, the mapping returns the right products. If they convert below it, the mapping is too broad and needs narrowing.