Combining Text and Image data for Product Recommendability Modeling
Mark Capelo, Karan Aggarwal, Pranjul Yadav · 2019
Digital advertising aims to display relevant product based advertisements which matches user's intent. User's intent is usually captured via various metrics like clicks and sale. Research in the recent past have demonstrated that image of the product plays an important role in order to achieve the outcome of interest. However, product advertisements with nonrecommendable images, i.e., product depicting adult, racial or political content often leads to sub-optimal performances, along with bad reputation for the partner and the advertiser. To overcome this challenge, we propose a model for product recommendability, i.e., to identify whether the product is recommendable or not. Our proposed approach is a hybrid model build using image and text data. We demonstrate the superior performance of our proposed approach on a dataset, obtained from a major e-commerce advertiser.