Digital Era Filters to Personalise User Experience & Satisfaction ...

21st century Era of Information Age where number of new technologies is popping each day rapidly. In this era of Internet the central focus of any business is seamless communication and commerce through one and only Internet. It is like convergence of computer towards information and its associated products/services. As today there is nothing which computers can't suggest or sell. But when number of choices and available information online is overwhelming there is a need to filter and prioritise to deliver relevant choice or information to users. This potential problem of highly available information and choices on internet can be alleviated by applying adequate filters for many Internet end-users.

A filtering system that predicts user preferences from list of available choices based on their interests, usages, or feedbacks. This kind of personalised, enhanced user experience increases business sales, retain customers, provides end-user satisfaction and the level of trust it builds by perfectly relating to users choice is commendable with these Recommendation System. There are many big market players who use recommendation system to engage their customers like Amazon, Netflix, Youtube, IBM, Microsoft, Google and many more.

Recommendation System Types:-

1. Content based filtering - Based on past user content or interactions history. Hypothesis for this is if a user was interested in an item 1 in past, then they will be once again be interested in it in future as well.

2. Collaborative filtering - Based on user behaviour. Eg. If a user likes item 1 and another user also likes the same item 1 along with another item 2, then the first user could also be interested in the second item. Two methods to achieve this are:

  a. Memory based - Uses clusters of similar user to predict interactions of similar users or use clusters of similar items to predict interaction of same user with similar items.     
  b. Model based - Uses ML and data mining techniques. 
  
3. Hybrid filtering - Combine various aspects of above filterings types to make recommendations.
Building a Simple Recommendation System with following steps:

Building a Recommendation System with following steps:

1. Load Input Dataset.
2. Data Profiling.
3. Limit interactions to relevant fields.
4. Clean the skewness and check for unique values in data.
5. Transform to feature vectors.
6. Similarity calculation among feature vectors.
7. Split the dataset in train and test
8. Fit model.
9. Calculate evaluation metrics for model.
10. Get Recommendations.

Lot of open source frameworks & libraries can contribute well to building a recommendation system like LightFM is one of them. Also there are lot of training dataset that is available for models learning of your system like Yelp Dataset.

So if you want to drive the information traffic in this digital information era. Engage more customers for your business. Want to reduce workload and overhead. Deliver relevant content. Gain perfection in understanding your customers better. Increase business sales and revenue. Recommendation system is an apt. choice for most of these.