Recommendation systems have quickly become the solution for the information glut that the Internet has brought us. Search engines and content aggregators function as machine-learning enabled gatekeepers to keep users from drowning in a sea of unsorted data.
Nowhere is this truer and crucial than in food delivery applications. This vertical in particular has seen phenomenal growth over the past few years, mainly due to the higher adoption of smartphones among the Indian population and the rise of affordable internet. As the user-base of these applications grow, so does their need to add more restaurants to their ever-growing menu.
In order to service an increasingly large and diverse user base, these apps have been heavily incentivised to innovate in the recommendation system field. With every passing day, the glut of restaurants increases, pushing food services applications to create systems that deliver relevant restaurants to users.
Join us as we take a deeper look into how recommendation systems work, and how they are being used by food delivery giants to increase orders.
How Do Recommendation Systems Work
Recommendation systems are generally run by machine learning algorithms, with the filtering method determining the type of recommendation systems. There are two main types of recommendation systems. The first is the content-based system, which filters results based on similarities of item attributes, and the second is collaborative, which provides results based on similarity from interactions.
In content-based recommendation systems, the first step is to give attributes to the items in the list. This will allow for the creation of metadata that will be used to sort the items. Post this, existing data can be used to create a user profile, such as previous orders.
Collaborative recommendation systems function instead of interaction matrices, with the machine learning algorithms learning how to predict items that incentivise engagement. This can be done through data is provided by the users, which reinforces certain recommendations of the system.
Either through a click or rating, the recommendations of the system and their effectiveness are measured. This creates a loop where the recommendation engine functions on user actions, analyses them and proceeds to provide more recommendations. This will then be pushed to the front-end of the website to garner more user actions.
Systems such as those used by Swiggy and Zomato function at the intersection of these two systems, as they use both approaches in order to show the customer the relevant result first. This drives engagement on the home page as well as user satisfaction at easier user experience.
The Search For A Relevant Restaurant
Everyone who has attempted to use these applications has known the trouble of finding the perfect restaurant to eat from. These are governed by a variety of factors. However, the apps have 200-odd choices to pick from at almost any given location, considering their penetration into local markets. This has created the need for a personalised picking.
Restaurants are sorted according to "relevance" on both apps, which is a fancy way of saying that machine learning algorithms are learning from user actions. For example, if a user is known to order regularly from restaurant 'A', it will slowly move the recommendation for that specific location towards the top.
Soon, the restaurant will be the first one to be displayed, which makes more sense for the user as well. This not only saves time but also creates a degree of personalisation for the users. This increases the satisfaction of using the application, as customers also feel valued when recommendations are tailored to them.
Dynamic sorting is also a huge part of the experience for mobile applications, as heavy users tend to order from different locations every day. A typical user might place an order at his or her workplace, and go home and place another order for dinner. The sorting for both of these locations are held to the same level of personalisation, and more relevant results are moved to the top as user actions are recorded.
A Data-Driven Approach To A Satisfying Meal
Foodtech companies have become more relevant in the emerging tech landscape with their use of machine learning and data science to recommend perfect meals at the right time of day. For example, Zomato uses user data such as search history, browsing history and order history to create a strong user profile. According to this profile, they ascribe characteristics to the user such as cuisine preferences, price range, and time of order. It also takes into consideration the bookmarks that the user has made when recommending places to order from.
This also extends towards other parts of the application, such as when the user decides which restaurant to order from and opens the menu. This is also driven by machine learning algorithms, which push previously ordered or recommended items based on the user profile to the top of the menu. This creates a "recommended" tab, which Zomato says has decreased order time by 21%. The degree of personalisation also offers a specialised user experience.
On Swiggy's side, they utilise machine learning heavily to recommend the best restaurants to users. They also create a personalised coupon and discount packages which take into consideration a lot of factors. For example, if a user has not logged into the app for a long time, he/she is given a bigger discount on a relevant restaurant to incentivise them to use the app. Moreover, these coupons are also personalised according to the customer, and also serve to bring the price down to their range.
They also have a system that recommends specific types of food from various restaurants around the user's locations. A collection of the same dish, for example, pizza, pasta or burgers, is shown in an easy-to-pick menu that prompts customers to pick their favourite dish from a curated selection of restaurants.
Foodtech seems to be taking applying data towards recommending food to the next level. It is left to see how future advancements will change the state of the market.
Register for our upcoming events:
- Meetup: NVIDIA RAPIDS GPU-Accelerated Data Analytics & Machine Learning Workshop, 18th Oct, Bangalore
- Join the Grand Finale of Intel Python HackFury2: 21st Oct, Bangalore
- Machine Learning Developers Summit 2020: 22-23rd Jan, Bangalore | 30-31st Jan, Hyderabad