{"id":688,"date":"2021-10-14T11:51:51","date_gmt":"2021-10-14T09:51:51","guid":{"rendered":"https:\/\/www.mymenu.pl\/blog\/?p=688"},"modified":"2022-02-11T15:28:39","modified_gmt":"2022-02-11T14:28:39","slug":"personal-recommendation-system-for-recipes","status":"publish","type":"post","link":"https:\/\/www.mymenu.pl\/blog\/personal-recommendation-system-for-recipes\/","title":{"rendered":"Personal recommendation system for recipes"},"content":{"rendered":"

Personalization is a key feature for any modern website that offers some kind of product for their clients. Recipe blogs and platforms can also benefit from this new technology. In mymenu we developed a personal recommendation system for recipes.\u00a0 This means we generate a unique set of recipes for every returning user, based on their previous recipe engagement. In other words, our algorithm can transform your recipe platform into a Spotify or a Netflix for recipes!<\/strong><\/p>\n

 <\/p>\n

How does the recommendation system for recipes work?<\/span><\/h2>\n

Everyday housands of users search for meal inspo on the Internet. In order to make it easier for them we created a personal recommendation system for recipes.\u00a0<\/span><\/p>\n

\"Personalne<\/span><\/p>\n

 <\/p>\n

 <\/p>\n

Firstly, we track user events based on their activity on the website. For each User ID we gather info on the recipes they interact with based on selected recipe engagement events. To each event we applied a weighted scale. As a result we can assign a score for every recipe.<\/span><\/p>\n

 <\/p>\n

 <\/p>\n

 <\/p>\n

\"recommendation<\/p>\n

 <\/p>\n

 <\/p>\n

Secondly, our machine learning hybrid algorithms calculate similar recipes depending on the collected data on user activity. We choose the closest similar recipe to the ones the user has interacted with. The simillarity is determined based on the biggest shared score the recipe obtains from user interactions. For better accuracy we also consider the time that has passed since the last interaction.<\/span><\/p>\n

Simillar solutions for content-based algorithms are used in Spotify and Netflix.<\/strong><\/p>\n

 <\/p>\n

\"\"<\/p>\n

 <\/p>\n

With just 1 user interaction we are able to cover 99% of our user database in order to calculate a personalised recipe selection and update the recommendations daily per each user. Because we gather thousands of event data every day, our personal recommendation system for recipes is more efficient and requires less data in order to generate results, as opposed to models based on simple user comparison.<\/span><\/p>\n

 <\/p>\n

 <\/p>\n

 <\/p>\n

 <\/p>\n

On their next visit, the users see a personalised recipe selection that matches their preferences. The content we recommend responds exactly to their needs, but it\u2019s also something they would probably never be able to discover by themselves.<\/span><\/p>\n

 <\/p>\n

Personalization for recipes performs better<\/span><\/h2>\n

The results? An A\/B test of our system proved that in the user group for which we were showing the personal recommendations:\u00a0<\/span><\/p>\n