Recommender System Based on Prediction of Distribution in Stream Data and Presenting Unexpected Suggestion

Due to the huge volume of data today, difficulties arise in research and computation. As the volume of the data increases, it is not possible to process it effectively by passing over the same data many times. The fact that a data item can be processed only once imposes restrictions on the implementation of the underlying algorithms. For this reason, algorithms to be used in stream data mining should be designed to work with a single pass over the data. In most cases, there is an inherent temporal component to the flow mining process, as data can evolve/change/increase/decrease over time. This behavior of flow data is called temporal territoriality. Therefore, a simple adaptation of single-pass stream mining algorithms may not be an effective solution for the task. Stream mining algorithms need to be carefully designed with a focus on the development of background data. 

Within the scope of this project, a dynamic window structure to be created for flow data will be used to predict future behavior and time of occurrence based on the frequency of the window and the accuracy of the previous estimates. In the system to be developed, a new model will be created based on online learning and updating the decision function for each new observation. By determining the next time of an event in the flow data, it will be possible to create timely and useful suggestions for users with the model to be developed. By constantly updating the models to be created according to user behavior analysis with flow data, changes in user interests will be detected and useful predictions will be offered.