Recommender Systems: Content Based and Collaborative Filtering

Introduction

Recommender systems are the most powerful machine learning algorithms that aim to predict about users interests and recommend products to them. These systems use data that originates from explicit user ratings after watching a movie or listening to a song, from implicit search engine queries and purchase histories. Brands like youtube, netflix, and spotify use such kinds of data to suggest content to a user.

Why do we need a recommender system ?

By using a data model to filter through your users favourite products it is easier to make recommendations to them for what they would like to watch, buy or use.

By ensuring that your users receive time to time recommendations that would suit their area of interest they would keep coming back to you. Companies use a recommender system in order to enhance their user experience and increase their sales such that the user is more likely to buy additional recommended products and consume more content.

Types of recommender systems 

1. 
   

   Collaborative filtering 

   
   

This type of recommender system is based on the approach that people will like the things in the future that they used to like in the past. The system will generate recommendations based on the rating profiles of multiple users or items. It checks for similar interests in users and provides recommendations.

This system will generate more accurate and efficient results if we have a large quantity of information about users and their liked items. Collaborative filtering models can recommend an item to user A based on the interests of a similar user B.

User-based nearest-neighbour collaborative filtering : there are three users A, B and C respectively and their interest in candies. The system finds out the users who have the same sort of taste of purchasing products and similarity between users is computed based upon the purchase behaviour. User A and User C are similar because they have purchased similar products.

Item-based nearest-neighbour collaborative filtering :The system checks the items that are similar to the items the user bought. The similarity between different items is computed based on the items and not the users for the prediction. Users X and Y both purchased items A and B so they are found to have similar tastes. Singular value decomposition (SVD algorithm) is used as a method of collaborative filtering in recommendation systems. It is a matrix factorisation method that is used to reduce the features in the data by reducing the dimensions from N to K where (K>N). Matrix factorisation is done by using a user-item rating matrix. Matrix factorization is about taking 2 matrices whose product is the original matrix. Vectors are used to represent item ‘qi’ and user ‘pu’ such that their dot product is the expected rating  

2. 
   

   Content based filtering

   
   

This type of recommendation system works on the principle of similar content suggestion. For example, if you are searching for a mobile phone, then the system will check for the same category. It works by the data that we take from the user, either explicitly (rating) or implicitly (clicking on a link). By the data we create a user profile, which is then used to suggest to the user, as the user provides more input or takes more actions on the recommendation, the engine becomes more accurate.In the user profile, vectors that describe the user's preference are createdAtAt. In the creation of a user's profile, we use the utility matrix which describes the relationship between user and item with this information. The best estimate we can make regarding which item the user likes is some aggregation of the profiles of those items. To check the similarity between the products or mobile phone for example the system computes distances between them and calculates euclidean distance between their camera and ram. We can compute distance calculation for any of the features of the product. If the euclidean distance comes out to be 0 between both they are more likely to have similar properties. Different metrics are used for different scenarios. For computing the similarity between numeric data, Euclidean distance is used, for textual data, cosine similarity is calculated and for categorical data, Jaccard similarity is computed.

Cosine Similarity: Cosine of the angle between the two vectors of the item, vectors of A and B is calculated for imputing similarity. If the vectors are closer, then small will be the angle and large will be the cosine. 

Jaccard Similarity: Users who have rated item A and B divided by the total number of users who have rated either A or B gives us the similarity. It is used for comparing the similarity.

Netflix Reccomender system : How it works 

About 80% of what people watch on netflix comes from their recommendation algorithm. And since their recommendation system is their secret weapon it is worth taking a closer look at. Netflix has set up 13,000 recommendation clusters based on users viewing preferences. The personalised recommendations are based on several factors that include a user's previous interactions like viewing history, search and ratings ,information about the specific title, the device used to watch videos on Netflix and the watch time. This complex system uses various algorithms including reinforcement learning, neural networks, casual modelling, probabilistic graphical models, matrix factorisation, and ensemble learning. Netflix takes customer feedback and re-trains their models frequently to improve accuracy of their suggestions.

Conclusion 

The recommendation system is a powerful system that can add value to the company or business. In the future, it will continue to be researched and developed to bring a better experience to users.Recommendation system has changed the whole scenario by making it easy for the user to choose their desired choices. It recommends user personalised content and makes it easier for them to browse through the internet and to choose a movie to watch on a weekend. Even though the pre existing techniques are good enough, there is still scope for improvement.