Search relevance is the measure of accuracy of the relationship between the search query and the search results.

Online users have high expectations. Thanks to the high bar set by sites like Google, Amazon, and Netflix, they expect accurate, relevant, and rapid results. However, the reality is many sites do not have optimized results pages that understand the user’s intent and bring them to their needs with ease.

If you’ve ever searched a website only to be shown a bunch of useless, unrelated results, then you know just how your users might feel: frustrated and primed to go to a competitor’s site to find results. Search relevance is integral to the user experience.

Website owners can fine-tune their search relevance to order search results in the most helpful way to users. This can be based on a number of factors, such as search intent, business priorities, textual relevance, spelling accuracy, geolocation of the user, or proximity of the keywords in the content searched.

Fine-tuning search relevance for accuracy 

Relevance can be hard to get right, since it is highly dependent on context and a number of changing variables. For example, the type of site matters: the way things should be ranked on an e-commerce website versus an academic site won’t be the same. The type of searcher matters as well. A collector and a novice will have varied intents and search skills. Finally, different people will have different words to express what they are looking for, even for the exact same query. A result ranking formula has to work with these different needs.

When a user enters a query like “discount” on an ecommerce site they want a specific subset of records that match the query criteria to be returned. 

Why does search relevance matter? 

Optimizing search relevance is an extremely important, yet often overlooked, aspect of user experience design. Research shows 43% of website visitors go immediately to the search bar, and these searchers are about 2-3 times more likely to convert. When users are served results that align to their query and interests, they will be more satisfied, more engaged, and even more likely to convert. Thus, a good UX design should encourage users to start with the search bar and navigate through the search results.

In addition, modern online users have high expectations for website usability and thus speed, ease of use, and simplicity of design are important factors in how customers perceive a brand. 

A brief history of search relevance

The history of search relevance goes back to the earliest days of the Internet, when researchers were trying to find methods for information retrieval to manage the fast growth of content being created every day. This quickly led to the invention of the search engine.

Search engines

Early search engines and protocols such as Archie, created in 1990 by a postgraduate student at McGill University, and Gopher, created in 1991 by researchers at the University of Minnesota, were important milestones in the development of modern search relevance systems. They enabled academics to use search terms to search through the file systems of other institutions that they connected to over the Internet. 

Yet, these early search engines were still very technical, requiring users to have advanced knowledge of computers and low-level Internet concepts. Meanwhile, just a couple years later in 1993, the World Wide Web started to flourish as hundreds of websites started to go online, initiating a whole new wave of search systems.

Early web search engines

Soon came web crawlers that automatically loaded and updated web pages into the search engine indexes, allowing more content to be searched.

Web search engines such as Excite in 1993 and Yahoo in 1994 quickly gained popularity due to their ease of use. They even included some basic statistical models aimed at understanding user queries and how they related to content.

These novel early systems employed a workable but limited method to rank the most relevant search results to users. They based their relevance ranking on the number of times keywords appeared in web pages, however they didn’t consider any other criteria to assess the quality of the web pages.

Then the game-changer Google search box came onto the scene. Google, founded on September 4, 1998 in Menlo Park, CA, vastly improved search relevance, by building its cutting-edge search engine technology. In the 2000s, Google advanced its search algorithms using newer and more powerful machine learning models that offered even better relevance and predictive search features such as autocomplete and instant search.  

Measuring relevant document search   

With the number of sites expanding, the need to search for relevant documents within particular sites and databases grew as well. Document search was an important precursor to the way we search on sites today.

Traditional ranking systems would often look at the frequency of keywords in documents to predict their relevance. For instance, a classic algorithm known as TF-IDF would look at the number of times keywords appeared in the respective documents (Term Frequency) and at the number of times keywords appeared in all other documents in the repository (Inverse Document Frequency). The latter analysis helps to filter out common words that are typically noise, such as prepositions.

While TF-IDF and other such early approaches to relevance were good at general purpose document search, they failed to take advantage of the additional structure and metadata that most websites contain. Modern content has titles, descriptions, categories, tags, and more keyword-based information that can be used to interpret site content and improve search relevance. 

Over time, search engine companies have developed alternatives to TF-IDF, for example, by relying more on keyword algorithms than statistics. For our own search engine, we developed a tie-breaking and custom ranking system that makes it easier for businesses to see what’s happening under the hood, which enables them to tailor the relevance to their own needs. 

With the advent of newer semantic search capabilities, there are also more sophisticated ranking quality models to score search relevance including nDCG, the normalized discounted cumulative gain, which can determine the similarity between how well a set of query results is ordered for a particular query. The higher the score, the higher the relevance. We have added scoring with neural technology and vector search to expand the relevance of our keyword search to more use cases. There are other methods, too, such as MRR (mean reciprocal rank) and MAP (mean average precision), each with their own pros and cons.

For better relevance, the quality of the records in the search index matters. Keyword and semantic search is only as good as the quality of your data, which is why data cleansing to handle missing values or noisy data, structuring datasets from different sources so it can be better analyzed, and improving content such as titles, descriptions, tags, headers, and metadata can greatly influence search quality score. 

Implementing search result boosting and optimization

Today, as websites have grown their content and product offerings, optimizing search relevance is a major consideration for individual site search engines. Businesses producing their own relevance need to take into account their specific business needs to make their search useful.

For instance, an e-commerce brand may have thousands of diverse products and customers of very different demographics. Thus, when a customer searches for a product, the internal search engine must be able to provide results that are not only related to the query, but also contextually relevant to the specific user.

Today’s search engines must also know how to handle synonyms, typos, multi-word queries, or even questions. Natural language processing (NLP) is used by search engines to help read, understand, and make sense of human languages. 

Additionally, marketers may want to promote seasonal items similar to that of their in-store merchandising efforts, or business operators may want to push higher margin goods. Therefore, a relevant search system must also be able to take these factors into account and provide custom ranking that can be adjusted to meet these needs over time. 

Many of these algorithms are still clunky, however. To improve this, some algorithms take into account the importance of the matching attribute and the proximity of the keywords. By doing so, the search results are much more likely to be relevant to users than general purpose search algorithms.

More recently, to improve relevance, search designers have been working to build in more personalization and contextualization. This includes technologies like machine learning and natural language processing to enable more conversational search, tracking of user search and browsing history to allow for custom interpretations of queries, and automatic tagging and categorization of web pages for an understanding of content at a higher level than simply that of text keywords.

Fine-tune your search relevance with Algolia 

Optimizing a website’s search relevancy is a complex and ongoing process. It requires not only providing results that match users’ queries, but also provides them personalized results while also meeting your own specific business needs. Further, as users move more toward voice-enabled devices and digital assistants, businesses will have to figure out how to provide a new type of interface that can speak naturally with users.

To provide all of these features to your customers, you’ll need a search-as-a-service partner that provides all the industry best practices and state of the art capabilities out of the box. Better relevance makes a huge difference in whether a customer — on a website or intranet — walks away with a good search or bad search experience.