New Features for a Better Personalization Experience
Like most of the world, the majority of 2021 was spent on home office or in isolation - which left us with all the time to be invested in work (and Netflix :) ) and continue improving UX for our clients. We are now happy to share what our team was able to build and what advanced features we can offer to reach new levels of personalization.
Advanced features for content recommendations (video, music, articles...)
Netflix-like and Showmax-like rows
As the popularity of streaming platforms spirals, so does the complexity of personalization. We have noticed multiple trends across the industry, where most platforms are trying to implement features alike to the global giant of video streaming platforms - Netflix, Showmax.
We collaborate with our premium clients on integrating their video streaming platforms (VOD, SVOD, AVOD) in a new way of showcasing content using Netflix-like rows (which is how Netflix achieved 80% stream time through personalization). Utilizing a two-tiered row-based ranking, the recommendations are organized vertically as well as horizontally. Meaning, within each row, the strongest recommendations are on the left, and across the rows, the strongest recommendations are on the top. Resulting in a fully personalized experience where each user’s homepage is assembled from the most relevant categories and titles based on their unique tastes and preferences.
Infinite Scroll Personalization
Besides Netflix, another platform that lately earned its name as a master of personalization - TikTok, inspired us to innovate the content recommendations game.
We have developed a unique technology enabling real-time infinite scroll personalization that will provide users with TikTok, Instagram, or Twitter-like experiences.
The more and more popular Infinite scroll enables bottomless content loading as the user scrolls down the page, eliminating the need for pagination. This technique makes every user feel like there are endless content options specific to their taste with every scroll.
Personalized search 2.0
Have you ever searched for a foreign artist, but weren’t quite sure how to spell their name? Or typed in “MaDDona” for the 6th time, and left wondering why there are no albums being shown? Not every AI is trained to understand typos and understand the user meant “MadoNNa”.
As we are working with many media houses, podcast platforms, and other sites providing content on their side that can be found with a search function, we ease the user’s experience with a personalized full-text search. Our personalized search 2.0 helps users find what they have in mind as long as the search query remotely resembles any piece of content that was uploaded in the catalog feed. This function now also includes Search synonyms.
Content personalization in action
The new features aim for users to enjoy their streaming experience with each login. Providing personalized guidance helps users find what they are looking for much faster and saves them from the frustration of aimless scrolling. With the competitiveness of the online landscape and hundreds of alternative options online, a tailored experience can convert unsubscribed users into happy loyal streamers. To see a fully personalized experience from the first click, try our media demo site.
Scenario set-up guide
Throughout the years we understood every client has different pain points and is trying to achieve different KPIs. That is also the logic of each scenario which can be easily configured in a codeless way and changed over time. Some of the most popular scenarios (with explanations) can be found in our content recommendations scenario set-up guide.
E.g. See a personalized site for a user interested in family content. All scenarios are personalized to a specific genre, even though the platform offers romance, thriller, comedy, and many others.
Advanced features for product recommendations (e-commerce, marketplaces, real-estate...)
Infinite Scroll Personalization
A lot of features from our content recommendations can also be applied to our product recommendations - for example, infinite scroll can be used in e-commerce or marketplaces, or even in real estate to increase the chance of finding the right product. The continuous load personalization is a feature that can be now easily integrated through our Admin UI.
Personalized search 2.0
For marketplaces or online stores with huge catalogs of products well personalized internal search is a game-changer! It does not only save users time but can also provide a competitive advantage through seamless UX. Recombee’s personalized full-text search enables Search synonyms and the inclusion of typos. For example, an online shop that offers both “coffee tables” and “nightstands” can show a user seeking “ a table” both options available, despite it not being the table seeker's exact command.
Product recommendations in action
Real AI powers all our scenarios and undergoes constant retraining to meet each client’s KPIs. As there are more numbers of e-commerce and marketplaces scenarios to choose from, we have prepared a demo store to see in practice some of our most frequently used ones (i.e. “just for you”, “popular & trending”, or “recently viewed”). For no coding configuration of the most popular scenarios, our clients can now follow our Product recommendations scenario set-up guide.
Even faster and more efficient integration
Understanding huge variability in our client's use-cases and technical capabilities, we always strive to provide multiple integration options, so each client can find the most suitable solution. To enable new clients to effortlessly implement Recombee, we have prepared new ways for integration - the fastest one can be done in a matter of minutes!
HTML widget & built-in user identifier using 1st party cookies
The simplest, fastest way to integrate Recombee with no coding is through our HTML Widget that allows for item synchronization using URL. When first released, clients preferring the HTML widget method were identifying users from their side. With the growing trend of concerns of user privacy, we have developed a built-in user identifier using 1st party cookies.
Working with first-party cookies enables Recombee to provide the highest level of personalization while adhering to the strictest privacy regulations. Switching to 1st party cookies is one of many behind-the-scenes steps Recombee initiates to help our clients fulfill current and upcoming data privacy policies (see differences between 1st and 3rd party cookies here).
Extended support for catalog feeds
We have completely redone the Catalog feeds section and besides the Google Merchant feed, we now also support RSS and custom XML or CSS feeds.
If you generate a catalog feed in almost any format, you can now just set its URL in the Admin UI, and Recombee will take care of periodical parsing and processing of the feed, ensuring that your items catalog is always up to date.
Integration Issues section
There is nothing more common during software development than introducing a bug. A new section of our Admin UI helps developers find & fix these issues by showing error messages from all the API calls in a single listing.
For clients using Segment, or wishing to use Segment, we have developed an easy, minimum coding needed integration through the CDP platform. The integration is done on Segment’s platform and can be configured in Recombee’s admin UI.
We chose Segment because it greatly simplifies data collection from digital properties, such as websites or apps, which helps clients gather clean data for multiple purposes. Following the integration steps, clients can send views, purchases of products, or video watch time information to Recombee, and start personalizing within a few clicks. Read more information in our Recombee Segment blog post.
Kentico Software integration
For clients looking for multiple marketing services, we have enabled easy integration through a DXP platform Kentico Xperience, where Recombee AI recommendations can be added through the Kentico Xperience Recombee module. Similarly Kentico Xperience, there’s a simple integration available through Kentico Kontent, a headless CMS for content management at scale, now available with Recombee personalization.
Continuous AI research
Our AI is the core of our system, and there is no expense spared when it comes to developing new models. Our researchers can be typically found on the grounds of our partner institutes, European AI hub prg.ai, or the Czech Technological University in Prague, where Recombee develops new models and algorithms.
Improving linear methods for recommendation
In our article on Linear Methods and Autoencoders in Recommender Systems, you can read about linear models and how we improved these models by a non-linear part realized by variational autoencoder.
You can check out our code from the repository.
Next basket prediction with neural networks
Continuing on our research into next basket prediction, Recombee reaped recognition at the annual Recommender System Conference, RecSys 2021 in Amsterdam. Our Machine Learner Developer Vojtech Vancura presented research into Neural Basket Embedding for Sequential Recommendation, whose goal is to develop models that would automatically refill customers’ shopping carts based on their habits. We are on the tip of our toes to be amongst the firsts to present these revolutionary models to our clients.
We have also organized the next round of challenges for senior data science students, where the number of active participants who submitted a solution was almost 50. We are currently supporting the research of the winning student, who is introducing new architectures of popular transformer models to handle long purchasing sequences. Read more in our Deep Learning for Recommender Systems: Next basket prediction and sequential product recommendation.
Interested in a custom personalization roadmap for your business? Meet our team and let’s talk about recommendations.
Linear autoencoders for collaborative filtering in recommender systems are simple and surprisingly accurate as we explained in our blogpost on how linear methods work. The critical disadvantage of methods like EASE is that they are not applicable to real-world problems...
At Recombee, we have always collaborated with academia — after all, five of our co-founders graduated from the Czech Technical University in Prague, one of the largest and oldest technical universities in Europe, and most of them hold a Ph.D. degree.
Linear regression is probably the simplest and surprisingly efficient machine learning method. It should be the method of your first choice, according to the famous KISS principle. Also, it often works better than sophisticated methods, because it is...