Session-Based Recommender System for Social Networks' Forum Platform

Recommendation systems are important technology, especially for e-commerce and social media platforms. The study has been focused on forum based data. In the forum system, data is producing by users and This production makes data more sparse and pattern free form. This study proposes session-based recommendation system for forum type data. System has been built with recurrent neural network architecture. Each topic is considered as a item and entries in the topics are contracting sequence system. With this our problem space becomes proper for RNN(Recurrent Neural Network), GRU(Gated recurrent Unit) and LSTM(Long Short Term Memory) based architectures. For testing phase, HIT@K and MAP@K error metrics has been used. According error metrics results, Stacked LSTM architecture has been more successful then the others and Study shows that, Stacked based architectures are more precise then single-layered recurrent models.