I am implementing a module which finds based on user interactions on an online portal to find which attributes of a certain product and what values for those products influence the buyer to choose the said Item.

My employers have suggested me to use Genetic Algorithm to achieve this. I have started implementing and found even the initial population generation to be a bit ineffective. Before I go further with this I would like your thoughts whether 'Genetic Algorithm' is the right way or is there a better way to go about this.

For me each genome is a list of attributes and their corresponding values.

Eg. genome 1: Color: Black, Material: wool, thread count:5 genome 2: Color: white, Handwash: yes, Dry clean only: No

Its a mixture of different attributes and their values. Am I going about this the wrong way?

  • $\begingroup$ Are you working from a history of sales for the customers? Or are you using the genetic algorithm to choose "live" what to present to customers and seeing what works best at the end of a real sales day? Do you have any customer attributes to use to personalise the suggestions, or are you treating all customers the same (just with random variations)? $\endgroup$ – Neil Slater Jul 18 '17 at 7:10
  • $\begingroup$ Currently im using history data. though the intent is to find patterns which can be used to improve real time user experiences $\endgroup$ – Sujit.Warrier Jul 18 '17 at 7:30
  • $\begingroup$ The plan outlined in the question seems quite vague, so it is difficult to assess it. For instance you say "based on user interactions", then give an example genome which does not use any interaction data, it is just a product description. It is perfectly valid to present recommendations/adverts based purely on product description (and measure user interactions in bulk results, not per user), but that doesn't appear to be your goal. I think you need to give more details about your goal, and be more precise in the problem description, before anyone could help you. $\endgroup$ – Neil Slater Jul 18 '17 at 8:54
  • $\begingroup$ you are right. first I would like to know what attributes and what value for those attributes for a product or a set product, compels a user to purchase an item. (disregarding demographic, social or seasonal influences for now) $\endgroup$ – Sujit.Warrier Jul 18 '17 at 9:01
  • $\begingroup$ I still think you need to explain more in your question. Use the edit link to give more details. Add some concrete goals and your plans to achieve them, give one or two examples of the user interaction data you have (simplified, and with anything identifying removed) - for instance is the interaction data what the user has looked at, whether they have clicked an existing advert, what they bought in the past? $\endgroup$ – Neil Slater Jul 18 '17 at 9:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.