Search

WordPress AI Recommendations
- SEO, conversions -

Table of contents :

What are AI-powered Recommendation systems : A 2024 complete beginner’s guide

Image of CTR skyrocketing using AI recommendation systems

Table of contents :

In the past, you used to increase your website’s click-through/conversion rate by adding a high performance and relevant search experience. But recently, a new type of technology has emerged : recommender systems (or recommenders). Recommendation systems often operate more subtly, with users interacting minimally with the recommendations and not fully registering their presence. They rely on personalization and are closely related to personalized search. Some of the most prominent recommenders on the market are Recombee, Amazon Personalize, Algolia and Google retail.

Logos of the different recommenders

Why choose a recommendation system?

 

Screenshot of the Amazon homepage

Example of the Amazon homepage full of recommendations to guide the user

 

Recommendation systems offer an effortless browsing experience

 

Forcing your users to search for the products or articles they want adds unnecessary steps to a simple process, thus greatly reducing your CTR and possibly dissuading them from continuing their visit. This could potentially make you lose countless conversions. AI (or machine learning) recommendation systems instead predicts the users’s needs and wants to preemptively suggest the most relevant or attractive content to them before they even have the chance to lift a finger. In this situation, the user becomes a passive participant and can count on the system to guide their choices, fostering a seamless and personalized browsing experience.

This means that you can guarantee that when a users arrives on a page, they will have found and clicked on the relevant content before they even had the time to bounce, therefore greatly increasing your website’s CTR and SEO.

 

Interaction based and context-aware

 

Machine learning recommendation systems are interaction-based which means whenever a user does an action or ‘interacts’ with any of the website’s elements, user events are sent to the engine. The engine, often powered by AI, can then use the collected user events to learn more about the website’s visitors’s behaviours and preferences to suggest only the most relevant content.

Graphic demonstrating the differences between explicit feedback and implicit feedback

But why would recommendation systems use interactions (or user events)? Explicit feedback is when users volunteer to provide data that illustrates their satisfaction with the content (examples: ratings, thumbs up, reviews, etc…) whereas implicit feedback are interactions: the users provide data passively just by interacting with the product (examples: opened detailed view of content, hovered over recommended content, added to cart, etc…).

Recommenders prefer to use interactions (implicit feedback) because even though the data is less reliable (based on speculation rather than listening to the user) it is more plentiful and varied.

 

How this feedback is used then depends on the filtering algorithm used by your recommendation engine. There are a few types of filtering algorithms currently used for different purposes : content-based filtering, collaborative filtering, knowledge-based filtering and hybrid recommendation engines.

DIagram showing the differences between Content-based filtering and Collaborative filtering

Collaborative filtering (only possible using AI-powered recommenders) will oftentimes be used for e-commerce purposes since analyzing groups of users and determining a profile based on similarities is very useful to push products that don’t necessarily complement what customers have already bought but you predict they could want anyway. Content-based filtering could be used if you have a content website and you wish to suggest content similar to what the users have already read/listened/watched based on their individual history.

 

Oftentimes, the recommendation systems don’t rely entirely on the given data and user events but are capable of understanding the real life concepts (semantics) implied by the data by leveraging the power of AI. This means that weather, location, news, etc… could all come into play when suggesting content.

 

Screenshot of the Ebay product detailed view recommendationsInside the ebay product detailed view, suggest similar products to boost sales

 

Best for e-commerce and streaming services

 

Recommenders are incredibly useful for e-commerce websites and you could  even argue that for these purposes they are a requirement. Have you ever seen a big and profitable e-commerce website without recommendations? I bet that you haven’t.

A big reason why it is so widely used is that an AI recommendation system is the closest thing to a vendor/sales person on an e-commerce website. Instead of letting the customer search and check everything on his own the recommender will suggest products everywhere and on every page of the website. It will even suggest products based on criteria and preferences the customers don’t even know they have, closing more sales in the end.

Some recommendation engines are even specialized in e-commerce, with models that are trained to specifically cross-sell, up-sell and suggest products based on similarity.

All this will contribute to make more conversions for your e-commerce website which will in turn make you more money, recouping the considerable investment in recommendations and more. There are quite a few e-commerce platforms : Woocommerce being the most popular.

 

Another big domain of expertise where AI recommenders excel is streaming services. Whether you manage a video, movie or music service website, you could use of some type of recommender to keep your users entertained, thus driving up user engagement and in turn keeping your users subscribed.

 

No fine-tuning necessary

 

To have the most relevant search and in turn increase your CTR by a rather measly 2-5%, you could need to hire a team of relevancy engineers (check out our blog post on quepid : a search relevancy tool)  which it would be a very costly endeavour. Whereas simply programmatically feeding the AI recommender model your own data will allow it to learn (by training) on it’s own and improve the experience greatly. Although it is far from free, it is way less expensive and demands less effort and paperwork than hiring consultants (depending on your recommendation system provider of choice, dataset size and productions website or platform traffic). And this is without taking into account the far greater conversion potential of a decent recommendation system vs an optimised search experience (Recombee brags about a 40% increase to CTR in some cases).

Some providers (like Recombee) can even train a model on demand so that it perfectly fits your needs.

 

 

Though machine learning recommendation systems have a few drawbacks

 

Although recommendations have all the benefits explained above, they have a few drawbacks that may deter some otherwise interested prospects.

 

Price

 

As mentioned above, a lot of the AI recommendation system providers can be very costly (even though they offer impressive results) which can deter quite a few smaller companies and businesses and will be even more so if you develop your engine yourself. So before implementing recommendations into your website is important to confirm whether it will be profitable with your traffic and within your industry.

 

You could check out our guide on which solutions you should use based on price.

 

Cold-start problem for recommendations is similar to cars on a snowy day

 

Cold-start

 

A lot of the recommendation systems rely on some sort of collaborative filtering algorithm and despite it’s strengths (predictively suggesting content based on user group profiles established through a collective dataset of user interactions) the recommendation system can’t work if the collective dataset of user interactions isn’t big enough. This is what we call a “Cold start”: needs a large number of user events to work. Some of the recommender providers use a hybrid algorithm to avoid the problematic cold-start but even if it technically works with a small number of events nothing guarantees that the recommendations are accurate.

 

 

Which AI recommendation system provider is the right one for you ?

 

The algolia logo

 

Algolia

 

Algolia is a company that used to specialize in search but now also does recommendations (Algolia Recommend). It can use both content-based filtering and collaborative filtering algorithms (separately) so it can suit all your needs. This means that even if you don’t have enough interactions/events on your website, you could make use of the content-based filtering algorithms (rather than collaborative filtering to avoid the cold start problem. You can also choose to send a wide variety of events. The biggest inconvenience that can deter potential customers would be pricing.

 

Advantages

  • Content-based filtering, collaborative filtering and image-based (looking similar) models are all available.
  • Wide selection of events guarantee that no matter the website (e-commerce,  informational, etc…), your recommendations will be as relevant as possible.
  • Personalized search included (same subscription tier as recommendations : in Premium and Elevate).
  • Optimize your recommendations with admin-defined rules in the Algolia backend, no additional code needed in your application (example : only show items where “category” is “Shoes”).

 

Drawbacks

  • Price on request and rather pricy.
  • Annual commitement, so no free trial for recommendation (and all personalization in general).
  • No hybrid filtering model, so collaborative search is unusable if your website does not have enough traffic (cold start).
  • Rules can only be set graphically, so they are rather limited.

 

Results

You can check out the official Algolia customer stories yourself do determine it’s effectiveness.

 

You can learn more about Algolia using the official documentation.

 

 

The recombee logo

 

Recombee

 

Recombee is a very well-rounded recommender. You can check out our complete Recombee review. It hosts a very wide variety of models, is very inexpensive for small websites and businesses (since pricing is based on usage) and uses ReQL to easily set rules for recommendations. This means that Recombee is good for e-commerce, streaming and all general purposes.

 

Advantages

  • Very comprehensive dashboard with detailed analytics (displays your CTR as well as it’s gain or loss overtime, how many recommendations displayed, etc…).
  • Uses hybrid search, no cold-start problems (content-based models for users with no prior interactions then collaborative filtering enabled after an interaction is sent).
  • Image filtering model available (to recommend similar content based on thubnail appearance).
  • Free tier for low traffic websites and 1-month free trial for everyone.
  • API clients available for a multitude of languages (PHP, Node.js, Pyhton, Ruby, Java, C#) or just use plain REST.
  • Personalized search is included.
  • Pricing displayed publicly and a price calculator is available.
  • ReQL (Recombee’s very own query language) enables perfect tweaking of recommendations per scenario in the Recombee backend. This ensures no code modifications are needed in your web application.
  • Item segmentation : Recombee can suggest groups of items based on their properties (example: user seems to like 80’s movies so create a recommendation containing only 80’s movies). This feature can also be used to automatically personalize the the homepage for each user.

 

Results

You can check out the official Recombee case studies yourself to determine it’s effectiveness.

 

You can learn more about Recombee using the official documentation.

 

 

Image of the WordPress logo

 

How to use it on WordPress

 

Adding recommendation systems to your web applications is a difficult task but if you use WordPress I have good news for you! Their exists quite a few WordPress recommendation plugins and WPSolr is the most complete. It does all the heavy lifting for you, no dev work or coding needed, letting you focus on tasks that impact your bottom-line and saving you money (you could check out our documentation on Algolia and Recombee).

WPSolr also adds search (keyword search, AI/vector/semantic search, personalized search) and is compatible with a host of search engines (Apache Solr, Elasticsearch, Opensearch, Algolia, Weaviate, etc…) so you can fulfill all your needs and greatly increase your CTR with one single plugin. Also offers re-ranking.

 

Conclusion

 

Recommendation systems are the latest trend when it comes to online content delivery systems. Contrary to search, the information, content or whatever data you host is suggested or “recommended” to the users instead of having them search for it. User events stored in the AI recommendation systems are used to train the AI so that it can accurately predict the users’ behaviours and actions to then recommend the most relevant content. It is a must-have for all large e-commerce websites and is so effective that it will probably easily reimburse all the costs with interest. One of recommenders’s biggest benefits are that they basically optimize themselves making it far less costly than search relevancy optimizations. It is recommended to use a recommendation system with hybrid algorithm so that you avoid the biggest pitfall of collaborative filtering algorithms: cold-start.

Related posts ... not powered by WPSOLR 😊