The problem Recommendation System to understand users better solves

Companies are having hard time understanding their customers and if they master it, their revenue starts growing. So, I am trying to solve it by using ML and understanding user's preferences. The project can be directly used for E-commerce or related applications, and can be slightly modified for other purposes. Based on user's input or history, their taste or interest is determined and ads or recommendations are done. This inreases their attachment to the company. The ML code runs over the dataset and builds a model file. So the next time, it does'nt need to train model all over again. This saves a lot of time and resources. The code needs to be run again only when new datasets are added.

Challenges I ran into

The algorithm being used is Cosine similarity that works on vectors. So, i have to transform the input into a vector which was tedious. I cleaned the dataset using nltk removing stop words and prioritising the weight of words for a given input (eg. Alisha Soft bed - here bed has highest priority, followed by soft then alisha. So recommendations are similar for bed). So after making it as vector, cosine is used to find distance of each and transforming it into readable text was difficlut. After reffering lot of resources, i finally got the output. Since it is a heavy weight model, training it each time would require lot of RAM memory. So, I saved them as a pickle file, and loaded them into project. Finally used flask framework to deliver it as a dashboard.