Разработка рекомендательной системы для покупателей интернет-магазина на основе методов машинного обучения
Аннотация
Nowadays, websites and online resources everywhere, there is a hug amount of information that available to everyone, it makes finding the right information much harder, in this case the recommender system come to the area. Machine learning helped a lot with recommender system, such as similarity and collaborative filtering or item-based filtering, but still we need better recommender system, so there is another technology that is called deep learning or deep neural network.
The big challenge in the area is that for each dataset or case we need a different model or recommender system, so it’s not easy to find the model that will be fit for your data, in this case we need more and more of different system to make performance as better as possible.
Different types of neural networks help developers to make better recommender system, in this thesis we are going to use recurrent neural network (RNN) with back propagation to make games recommender system to Mozgo company. By using such activation function in hidden layers such as sigmoid or Relu, we can make better recommender system with less error or loss.
At the last chapter of this thesis you will see the result of the model that we created with neural network and we tested network with different bias and batch sized also learning rate, and we got good result.
The big challenge in the area is that for each dataset or case we need a different model or recommender system, so it’s not easy to find the model that will be fit for your data, in this case we need more and more of different system to make performance as better as possible.
Different types of neural networks help developers to make better recommender system, in this thesis we are going to use recurrent neural network (RNN) with back propagation to make games recommender system to Mozgo company. By using such activation function in hidden layers such as sigmoid or Relu, we can make better recommender system with less error or loss.
At the last chapter of this thesis you will see the result of the model that we created with neural network and we tested network with different bias and batch sized also learning rate, and we got good result.