DevMesh

Psychology based recommendation

Pranjall Kumar

Pranjall Kumar

Bengaluru, Karnataka

1 0
  • 0 Collaborators

Preference to products like clothing, furniture etc can be very subjective. Best sellers are not always the best recommendations to receive. This Project aims to provide reasonable recommendations to consumers with very limited data about the user (a new user) by learning their psychology. ...learn more

Project status: Under Development

Artificial Intelligence

Code Samples [1]

Overview / Usage

Unlike products like Mobile, Oven etc. products like clothing, furniture, shoes, food etc tend to have a subjective inclination. This results in having different personal choices for such products. And best seller recommendations don't work. Here we need to understand that certain groups of people will prefer certain kinds of products.

Here we need recommendations of the form "People who saw this also saw" or "People who bought this also bought". However finding such correlations from large data-sets is cumbersome and for new platforms with considerably less data, finding such correlations accurately won't be possible.

This project tries to learn user possible preferences beforehand by segregating users into different classes and recommend products accordingly. This helps in a personalised feed of recommendation of products which improves user comfort, saves user's time and also helps build a bond with the user. In all it will help build a user bias towards the platform.

Methodology / Approach

To help categorise a user into one of the classes. A few multiple choice questions are asked to get insights about the person (possibly during signup). These questions are supposed to be very few to avoid user frustration. The exact content of the questions depends on the platform and will be experimental e.g. a furniture platform can have questions like preferred colour schemes with 5 to 6 choices and so on.

User provides responses to those questions. Then based on these responses, a Neural Network classifies the users into a few different classes. Again, the exact number of classes will be experimental and depends on the platform. These different classes get a personalised feed of recommendations which is highly likely to cater their preferences.

The performance of the Neural Network can be made to improve as the platform gradually gathers more and more data by using additional features via insights available from the data.

Technologies Used

Python.

  • Pandas.
  • Numpy.
  • Matplotlib.

Repository

https://github.com/pranjallk1995/Smart-Recommendation-System

Comments (0)