A search engine should have both high precision and high recall. When you search for a product on an online store, you should expect the top 5 or 10 results to match what you were thinking.
Search relevance is turning a search engine into the equivalent of a helpful sales rep: able to predict your needs. Good search amplifies a customer’s user experience; bad search is just bad service.
Let’s say you’re looking to buy a brown leather couch online. You go to a furniture store’s website and enter ‘brown leather couch’ in the search bar. A search engine with high search relevance will show you all the products listed as brown leather couch. One with low search relevance might show you vastly different results – brown chairs, black leather sofas, or even garden chairs – something you weren’t looking for.
Why is search relevance important?
Search relevance is the very first impression a company leaves on a customer. That’s because search is often the first thing most people use. Good search relevancy keeps customers on site, and helps retain users.
Think of bad search engines – if you consistently engage with a website that makes it difficult to find a ‘brown leather couch’, you’d eventually stop shopping there. Companies are looking for ways to improve search relevance so that customers can actually find their products in their search engines.
Search relevance is now more important than ever because expectations of search have increased. Think of Google: most people only click on results on the first page. First page results on Google garner 92% of all traffic from the average search, with traffic dropping off by 95% for the second page.
If your search engine doesn’t retain users on the first try, there is little chance it will manage to engage with the user later.
How does search relevance work?
Search engines are sophisticated analytical systems that rely on a variety of functions:
- Semantic Annotation: tagging different product titles and search queries.
- Text Analysis: recognizing different variations of the same word, to allow for fuzzy matching. For instance, shopped, shopping, and shopper all match up to the word shop.
- Query Weights: weighting the importance of different fields based on search requirements.
- Concept Tags: understanding the query in terms of specific concepts (instead of just matching terms).
- Natural Language Processing: understanding the grammatical structure of text in the query and search result.
- Statistical Processes: statistically detecting the relationship between different words that are related. For example, cutlery and dining table should be detected as related words.
- Click Tracking: determine which result is statistically most likely to be the best result for a query, given past user behavior.
Search Engine Evaluation
How do you know if your search engine is working properly i.e. delivering results that lead to customer retention and not abandonment? One of the best ways is to utilize a human relevance evaluation. This works by creating a representative sample of a few thousand or more search terms that your website is expected to get, and then noting the top results for each query. Then you have a set of humans rate the quality of the results by a simple metric of how useful they are. The actual parameters of what counts as useful is up to you (and your human raters), but this is one of the fastest ways to form a baseline definition of search quality.
This is where Lionbridge AI proves useful. We have a large, flexible workforce of over 500,000 native speakers across 300 languages, that can easily be set up to evaluate search queries. We can also build datasets to help you predict which categories fit best to a given product to make product classification easier, faster and less error-prone.
Search relevance elevates an engine from just a simple search tool to providing a customer the best possible online service. Lionbridge AI can easily help you achieve search results that match customer expectations.