Building and applying a machine-learning model at Google comes with interesting challenges. For example, even though it may be contrary to intuition, we don’t always have sufficient amount of labeled data to train strong supervised learning models. Or, even when the model quality is good enough for a product requirement, it may not meet our serving requirements. In this talk I will share some of these challenges & experiences from quality improvements in Search Ads products.

In the second part of the talk, we will discuss our upcoming paper (“Interpretable Neural Architectures for Attributing an Ad’s Performance to its Writing Style”) to be presented at an EMNLP workshop (Analyzing and Interpreting Neural Networks for NLP) about how we can use neural networks to understand ad performance and attribute it to advertising text. It is an interesting open NLP research problem to understand what words influence our decision making, and we hope that this line of research will lead to better ad copywriting in the future.

This colloquium is sponsored by Google.