The Power of Machine Learning: Unpacking Recommendation Engines

Across the web and on OTT platforms the most common terms are, “you might also be interested in”, “Because you watched”, “recommended for you”, and “People also searched for”. How does YouTube determine the next video that may capture your interest or how does Amazon anticipate and suggest products that align with your preferences? Thanos’s snap definitely ain’t the reason for these to happen, well in both cases, an ML-based recommendation model determines how similar videos/apps show up in the search results

Recommendation algorithms are most prominently recognized for their application on e-commerce platforms, leveraging customer interests to curate personalised lists of recommended items. A study conducted by Epsilon marketing revealed that brands offering personalised content have a higher likelihood of attracting 80% of customers towards making a purchase.. It is estimated that about 35% of Amazon sales in 2021 happened via recommendations. Also a study by PracticalEcommerce states that users who tend to click on product recommendations are 4 times more than those who do not use recommendations. 

A recommendation engine is a tool that uses machine learning to filter items that either use existing user insights i.e. what a user has searched for, what movie or show was viewed, what product was purchased, or predict a given customers preference on the basis of other customers

Recommendations can be classified as:

Collaborative Filtering: This is based on information from many users to identify the similarity in preference behaviour to predict future interactions. The idea is that if people have made similar decisions there is a high probability they will agree on additional future selections. 

                                                    (Img1: Source: nvidia/data science)

Content Based Filtering: Content filtering relies on the characteristics or attributes of an item (known as the content) to suggest similar items based on a user’s preferences. This approach focuses on identifying similarities between item features and user preferences, utilising information such as a user’s age, the cuisine category of a restaurant, or the average review for a movie. By analysing these details and the interactions between users and items they have engaged with, content filtering models can estimate the likelihood of a new interaction

                                             (Img2: Source: nvidia/data science)

Other Recommendation Engines ( mostly used by Ecommerce giants like Amazon)  

  • Cross – selling based on categories/product relationships, this segment is similar to the “Frequently bought together” recommendation. 
  • Recommendation based on browser history
  • Based on interest and offers 
  • Up sell recommendations i.e. to promote new versions of the product or any update or addons available for the product purchased. 
  • Based on popularity, 

Recommender systems play a crucial role in enabling personalised user experiences, fostering deeper customer engagement, and serving as powerful decision support tools across various industries such as retail, entertainment, healthcare, finance, and more. Companies implement recommendations engines to:

  • Improve customer retention 
  • Improve on overall sales 
  • Helping to form customer habits and trends
  • Increase pace of work 

According to Google, a significant portion of app installations on Google Play and a majority of the time spent watching videos on YouTube can be attributed to recommendations. Specifically, recommendations contribute to 40% of app installs on Google Play and 60% of watch time on YouTube.

Let’s take a look at how popular platforms use recommendation engines:

  • Netflix, uses a advanced personalised system that take into account descriptions, age range of movies, ratings and popularity 
  • Amazon has the most effective recommendation system in the market which works on models as illustrated above. Amazon applies filter grouping which later are converted to huge datasets that generate high quality real time recommendations. 
  • Spotify users are attracted to “Discover this week playlists’ ‘ that fits their music tastes. Spotify works on “Bandits for recommendations as treatments” or “BaRT” which works with three key algorithms: Collaborative filtering, natural language processing, audio path analysis. 
  • YouTube: YT’s recommendation engine works on two main networks, the first is to generate candidates i.e. it analyses millions of videos, narrows their collections based on the users viewing history, inquiries, demographic characteristics and the second is  the ranking system, which takes into account the data gathered from candidate generator system and considers users language, , language of video, users personal preferences and then sets the videos in the order of likelihood 

Recommendation engines play a significant role in media buying by providing valuable insights and recommendations that aid in making informed decisions. Here are several ways recommendation engines help in media buying:

  • Targeting and Segmentation: Recommendation engines analyse user data, preferences, and behaviours to identify relevant segments for targeting. By understanding the interests and preferences of specific audience segments, media buyers can optimise their advertising campaigns and deliver targeted messages to the right audience.
  • Personalization: Recommendation engines enable personalised advertising experiences by tailoring ad content based on user preferences and behaviours. Media buyers can leverage these personalised recommendations to deliver highly relevant and engaging advertisements that resonate with individual users, increasing the chances of conversion.
  • Cross-Selling and Upselling: Recommendation engines can identify opportunities for cross-selling and upselling by analysing user preferences and purchase history. Media buyers can leverage these recommendations to promote complementary products or services to existing customers, increasing the average order value and maximising revenue.
  • Predictive Analytics: Recommendation engines utilise predictive analytics to forecast user behaviour and preferences. Media buyers can leverage these predictions to make data-driven decisions regarding ad placements, content, and targeting strategies. This helps in optimising media buying efforts and increasing the likelihood of achieving desired campaign outcomes.

Overall, recommendation engines provide media buyers with valuable insights, personalization capabilities, and optimization opportunities that enhance the effectiveness and efficiency of their media buying strategies. By leveraging these tools, media buyers can deliver more targeted and relevant advertisements, optimise ad placements, and drive better campaign performance.

Similar Posts

Leave a Reply