Levels of machine learning in e-commerce product search

Technology is becoming more sophisticated than ever these days, particularly when it comes to artificial intelligence (AI). The most advanced systems are now able to do things that were once only possible for humans to achieve, and they are helping organizations make better business decisions than ever before.

One type of AI called machine learning (ML), in which computers can improve and adapt their processes without being explicitly programmed by a human, has become particularly helpful for people who do business by selling products or services online. If you run an eCommerce company, there are a number of ways you can tap into the power of machine learning to provide an enhanced user experience, become more agile and open up revenue streams that may have been previously inaccessible to you.

Machine learning improves on-site product search

Search is critical to the success of any eCommerce business. If your products fail to appear in front of potential customers, how will they ever know you exist? Machine learning algorithms can dramatically enhance eCommerce product search results, helping to boost those click rates, customer ratings and conversions.

Machine learning helps users get so much more out of the search experience and can pinpoint with precision the products or services they are looking for. With machine learning engaged strategically in the search process, search results become more meaningful and geared towards what the shopper actually needs rather than what they just typed into the search bar.

Levels of machine learning in on site-product search

The current offerings for e-commerce search and navigation enhancements fall into four basic levels of ML sophistication, starting from no machine learning whatsoever.

Level 1: Basic text-matching search – This is the most elementary form of search, where the search is set up to look in a database for what the customer enters. This is still used by a remarkably large proportion of e-tailers, and one of the obvious ways of noticing this is that the search results will be limited to products which contain the exact query in the their metadata (i.e. title, description, tags, etc.).
- Level 1 does something like the following:
  1. Break query string into words
  2. Compose the search query by targeting known data fields like title, description, features, word by word. The where statement will look like ‘where (description like ‘%red%’ or features like ‘%red%’ or title like ‘%red%’) and (description like ‘%bag’ … etc etc)‘
  3. Display results in rank order depending on where the query term appeared in the product metadata (e.g. a match found in the product title has more weight than a match found in the description).

Level 2: Query transformation and expansion – Classic Natural Language Processing (NLP) problems such as handling prefixes, suffixes and misspellings come into play. The user can enter a query and the engine will ‘fuzzily’ match the query, making a more forgiving search experience. The engine will still only find products with the ‘fuzzily’ matched words.

Level 3: Basic machine learning search - The engine claims to be doing machine learning, and it most certainly does. The retrieval is still done by fuzzy text matching, but logic can be applied on top of the search results to, for example, give popular products, or products of a certain brand or margin a boost in the search results.

In the best cases, a bit of personalisation comes into play at this level too. So that, for example, that bread you always buy at your favourite online grocery retailer will be at the top of your search results following a search for ‘bread’.

Or you may see features like “autocomplete”. A feature that tries to predict your query before you’re done typing by tracking previous popular queries that started with the same letters and led to a click.

Nearly all e-commerce search engines end at level 3. On top of this they often offer merchandising interfaces (drag and drop, IF-This-Then-That rules) and analytics tools for the marketing and merchandising teams to work with.

But there is so much more potential to unleash. Particularly as it relates to understanding true shopper intent.

Level 4: Advanced machine learning search – Imagine walking into a physical store and asking for a bag, and the store clerk hands over all the bags for you to look at, but not any of the products labelled “sack”, “backpack” or “duffel”.

Of course, this would never happen in a physical store because the salesperson would point you to the section of the store where all relevant bags types are found. Moreover, the salesperson may be able to tailor his/her recommendation based on information ascertained from your loyalty membership profile.

With advanced machine learning, the same customer experience is possible online and it doesn’t require massive amounts of user data either. Advanced machine learning models like Neural Networks can map those complex relationships between products so that the shopper is no longer inhibited by the words they use.

Instead, their search and navigation experience will be relevant from first interaction and become increasingly personalised with each subsequent visit. Exactly like the experience provided in a well-merchandised physical store with great ales personnel.
If we throw away the notion that all ML models are created equal, one can understand that this task requires much more sophisticated models than the basic ML re-sorting, but can also yield great returns in terms of increased conversion rates and order value.