At first glance, the design and development of search engines appear to be concerned mainly with computer science — particularly information retrieval, machine learning, and natural language processing.
But search isn’t just a computer science problem. Effective search depends on the interaction between searchers and the search engine. Searchers turn to search engines with an intent in mind, invest effort into the search process, and expect to obtain something useful in return. While computer science plays an essential role in each component of the search process, it’s important to think about search as a whole from a human perspective.
People generally use search engines to satisfy a particular information-seeking intent. They primarily express their intent through the search query, in some cases refining the query with a sort or filter.
It’s common to evaluate the effectiveness of a search engine using behavioral signals, such as click-through rate, click position, and conversion rate. But all of these measures signal conflate three things: searchers’ effectiveness at expressing their intent, the search engine’s effectiveness at understanding their intent, and the relevance of the search results.
If a search engine can’t understand what searchers are looking for, it’s going to have a hard time delivering relevant results to them. Moreover, if searchers are unable to express their intent, the search engine is going to have a hard time figuring it out from their search queries.
So, if you’re developing a search application, make sure to invest effort not only into query understanding, but also into ensuring that searchers express their intent as well as possible. Look at search logs, and if you can’t figure out what searchers are looking for, then your search engine probably can’t either. Invest in autocomplete and in guidance on the search results page, such as faceted search, to help and encourage searchers express their intent in a way that the search engine can understand.
Searchers are people. People are lazy. All else equal, searchers tend to invest the minimum effort that a search engine requires of them. They type as little as possible, perhaps taking advantage of autocomplete if they see a suggestion that aligns with their intent. They only look at the first few search results, rarely scanning past the first page presented to them. And most don’t invest effort into learning how to be better searchers.
There’s nothing inherently wrong with searchers’ desire to minimize their effort. Even if there were, it’s unlikely that a search engine can change something so fundamental to human nature. Indeed, powerful search interfaces that require extensive training have mostly given way to simpler interfaces that require as little as possible from searchers.
But the downside of searchers’ laziness — and of search engines’ eagerness to cater to it — is that sometimes it actually ends up costing them more effort and reducing their return on investment. In general, short queries are an ineffective means for searchers to express a clear, specific intent. When search interfaces are too simple, they deprive searchers of the transparency and control to communicate their intent effectively.
The foundation of human-computer information retrieval (HCIR) is that search engines help searchers who help themselves. The best search engines reward searchers’ incremental effort with a higher return on investment.
But searchers have been trained by simple search interfaces, and their laziness is compounded by a skepticism of anything that violates their expectations. In order to earn searcher effort, search engines have to provide simple, incremental, and effective steps that guide searchers — and that teach them through experience that the return justifies the additional effort.
Search is almost never an end in itself; rather, it’s a means to an end. When we evaluate the effectiveness of a search engine, it’s critical to keep that end in mind. If searchers aren’t able to successfully complete the tasks that motivated them to search in the first place, then all they’ve accomplished is to waste their time and effort.
Search engine developers generally maintain a variety of metrics to measure the search engine’s effectiveness. Sometimes those metrics serve as proxies for searcher utility. Sometimes they are business metrics, intended to measure the contribution of search to the overall bottom line.
It’s important to know the difference, especially if searcher utility and business utility are not completely aligned. More importantly, thinking explicitly about measuring searcher utility forces you to ask tough questions about whether your metrics are adequate proxies for it. For example, does a click on a search result indicate search success, or does it mean that the searcher couldn’t obtain enough information from the search results page?
Without reading people’s minds, it’s impossible to be certain what searchers are trying to achieve — let alone to determine if they’ve achieved it. But we can perform user research, propose models that lead to testable hypotheses, and validate those models with data. Conversely, if we don’t think about what searchers are trying to achieve, we’re unlikely to help them achieve it.
Search is about searchers, and searchers are human beings. Even as we use computer science to improve each component of a search engine, it’s important to think about search as a whole from a human perspective. Searchers use search engines as a means to satisfy an intent, and search engines serve searchers by delivering utility as a return for the effort they invest into the search process. Framing search in these terms is the first step towards delivering effective and delightful search experiences.