Incorporating Text Embeddings in Deep Neural Networks for Click-Through Rate Prediction

Click-through rate (CTR) prediction plays a vital role in enhancing the effectiveness of online advertising and recommendation systems. Existing deep CTR models primarily focused on numerical and categorical features, emphasizing their significance in accurately forecasting user engagement. However, the textual information associated with the content, such as product descriptions, movie titles, and ad titles, can provide valuable insights into user preferences and interests. In this paper, we explore the effects of incorporating text features into deep neural network-based CTR model using text embeddings. We employ various pre-trained word embeddings, such as Word2Vec, Globe, and FastText, to convert the textual information into dense vector representations that capture semantic meaning and contextual relationships. Our findings reveal that the incorporation of text embeddings significantly boosts the model’s ability to capture fine-grained textual information, leading to more accurate CTR predictions.