Automatic text summarization has been a well-researched NLP topic in recent years. It is possible to build machine learning models that are capable of distilling crucial information from a larger piece of text and condensing it to a smaller one. Text summarization using deep neural networks has become an effective approach and there are many use cases for that technique. One possible use case is text ads generation in online search advertising. Many advertisers are publishing text ads with ad titles that are not effective. Less effective ad titles will result in a lower chance of user conversion (clicks), which is harmful to the advertisers, and also a waste of hosting resources to the ads marketplace. It is meaningful to build a model that could rewrite the adtitle by summarizing the ad content. In this paper, we will study and leverage several state-of-the-art text summarization models, compare their performance and limitations, and finally, propose our own solution that could outperform the existing ones.
