Hybrid Recommendation System for Forum based Social Network Platforms

With the widespread use of recommendation systems, it has become a need to determine and present content specifically for each user on social networks. Within the scope of the study, a personalized feed system is being produced by looking at the past interactions of users for forum based social network platforms. A Collaborative Filtering Based Model (CFBM) and a Content Based Model (CBM) are used. In CBFM, item-item based approaches such as Item2Vec and TF-IDF weighted Item2Vec have been developed, taking into account the topics that the user enters in a particular session. In the CBM, BERT architecture and FastText model are used for vectorization of forum messages. In the evaluation, it was seen that CFBM is more successful than the CBM. When the interaction numbers of the suggested contents are compared, it is seen that CFBM recommends the contents that receive more interaction, while CBM is more prone to recommend newly produced contents.