Machine learning based search engines are systems designed to find items and services through text or vocal input. Recommendation systems are systems that make suggestions related to search history, customer profiles, and inventory metadata. In short, while search engines help users find what they want, recommendation systems help users find more of what they like, or relevant alternatives.
In the world of online shopping, these systems are hugely important. Effective search engines have to be quick, smooth, and deliver accurate results. Recommendation systems have to make appropriate suggestions for the user. The quality of these systems will impact customer retention, time on site, and sales volume.
So how do search engines and recommendation systems work, and how do you ensure high quality results to drive sales and keep customers coming back? In this article, we’ll give a brief introduction to each, and take a look at how quality evaluation enhances the user experience.
The search bar is a quick and simple start to finding what you want. For this reason, basic site-search is now a mainstay across modern ecommerce platforms. Basic-level search engines break a query into separate words, and through text-matching, link those words to product titles, descriptions, and categories. More complex search engines also include auto-correct, fuzzy matching (e.g. showing the same results for both table and talbe), and synonym recognition. As they get more advanced, search engines take into account factors like popularity, product rankings, and word clustering for more refined results.
However, search engines also have one more important feature: related results. These results work by introducing the customer to accessories, add-ons, and similar items during their search. At the same time, related results offer the customer alternatives when what they search for is unavailable.
For example, if a customer is looking for a coffee maker, the related results might include other coffee makers, coffee beans, and kettles. These results, and their order, could be the difference between a customer staying on your site or going elsewhere.
Recommendation engines deliver personalized suggestions based on a user’s previous actions and the actions of similar users. These suggestions can take the form of:
- Recommendations based on past purchases
- Related search results
- Newer versions of items the customer has viewed or bought
- Automated newsletters
A good example of this is Amazon, whose recommendation engine suggests accessories based on the purchases of other customers. It also shows what similar users bought, and creates individualized bundles of similar items to encourage larger purchases.
Broadly speaking, recommendation systems can be separated into two types: collaborative filtering and content-based systems.
Collaborative Filtering: Collaborative filtering analyzes past interactions between users and the database to make predictions about what a user will like. For example, if a user searches for and buys two books, the recommendation system will suggest other books bought by people who bought the same two books. As more people use the system, a deeper net of past interactions can be used to make more accurate recommendations.
Content-based Systems: Content-based systems use additional information to develop suggestions for users. This information can include age, gender, occupation, location, and more.
In content-based systems, the systems factors in the features of both the item and the user to generate recommendations. For example, males aged 18-30 living in the Gold Coast might like surfboard recommendations, but males aged 60-80 in the same location might not. These systems grow more complex as more people use them, allowing for a more specific range of suggestions.
Where to Start
Creating quality machine learning based search engines and recommendation systems is a worthwhile investment, but making them truly effective requires planning, maintenance, and evaluation. The payoff, however, is an experience customers will keep returning to for its simplicity, accuracy, and ease of use. We saw these results in the project we did with Traveloka. The online travel company revitalized their customer experience through refining their search engine. The result is a single search bar that covers the entirety of their core product offerings.
So if you’re looking to enhance your search experience or improve its related results, get in touch. Lionbridge’s data science teams and qualified community of 1,000,000+ contributors will help you define the scope of your needs and plan around your specific goals. Our community of evaluation specialists can then test your search engines quickly, at scale, and in multiple languages.
Get started enhancing your customer experience now.