Yet, building recommendation
systems is a challenging task

Data Sparsity

Users tend to rate a very limited number of items; leading to a sparse User-Item rating matrix, with insufficient data.

Cold Start

In case of new users, recommender has no preference-related data points. Similarly, for newly on boarded products, lack of rating inputs limits recommendation quality.

Accuracy

Correctly predicting a user’s item preference has always been a challenge especially in data sparsity situations. As product range widens, tracking change in user’s interest increases the complexity of recommendation.

Speed

The most relevant recommendations must be served in less than a second, to offer a smooth user experience.

Scalability

Traffic-triggering events (discounts, festive sales etc) could cause a system designed to recommend 100's of items to 1000's of users; to fail to recommend 1000's of items to 10,000's of users.

Data Sparsity

Users tend to
rate a very limited
number of items;
leading to a sparse User-Item rating matrix, with insufficient data.

Cold Start

In case of new users, recommender has no
preference-related
data points. Similarly, for newly on boarded products,
lack of rating inputs limits recommendation quality.

Accuracy

Correctly predicting a
user’s item preference has
always been a challenge
especially in data sparsity situations. As product range
widens, tracking change
in user’s interest increases
the complexity of
recommendation.

Speed

The most relevant
recommendations
must be served in
less than a second,
to offer a smooth
user experience

Scalability

Traffic-triggering events
(discounts, festive sales etc)
could cause a system designed
to recommend hundreds of items
to thousands of users; to fail to
recommend thousands of items
to millions of users.

A hybrid model addresses
some of these challenges

Users’ data is viewed as a sequence of macro interactions between users and product-items. Each macro interaction, in turn, includes a pattern of micro behaviours performed by the user during the shopping experience.

Combining browsing and transaction behaviour with an underlying ML layer, helps build a factory of high-accuracy, personalised recommendations.

Hybrid recommendation models use context-aware collaborative filtering; to optimally serve recommendations.