Social media advertising is becoming increasingly important in promoting products and services. A typical social networking campaign contains multiple ad groups, and each ad group typically contains multiple ads. When managing advertising campaigns, budget allocation for selected ads and ad groups is particularly important. It would be very helpful to know in advance how well specific ads or groups of ads will perform in the near future.
The master thesis describes the process of building predictive models for advertisement clicks in social networks. We want to get closer to the perspective of a campaign manager who typically has ad statistics for the past days, but needs to decide how to distribute the advertising budget among ads in the future. By optimizing the allocation of advertising money, we can radically improve the performance of advertising campaign management, where the development of predictive models plays a key role. Success in this regard is measured by the number of clicks the ads receive within a limited budget.
The main result of the master thesis is development and description of a process for building prediction models for predicting clicks on ads in social networks. We compared different algorithms for time series forecasting on the Facebook and Twitter social networks. The subject of the comparison was their ability to predict one or more time steps in advance. We focused primarily on predicting the number of clicks on ads. The results of the experiments showed that the methods used, especially the long short-term memory neural networks and the regression models obtained with the XGBoost algorithm, provide meaningful predictions for up to 24 hours in advance.
|
