No abstract

As the online advertisement industry continues to grow, it is predicted that online advertisement will account for about 45% of global advertisement spending by 2020.a Thus, predicting the click-through rates (CTRs) of advertisements is increasingly crucial for the advertisement industry. Many studies have already addressed the CTR prediction. However, most studies tried to solve the problem using only metadata such as user id, URL of the landing page, business category, device type, etc., and did not include multimedia contents such as images or texts. Using these multimedia features with deep learning techniques, we propose a method to effectively predict CTRs for online banners, a popular form of online advertisements. We show that multimedia features of advertisements are useful for the task at hand. In our previous work [ 1 ], we proposed a CTR prediction model, which outperformed the state-of-the-art method that uses the three features mentioned above, and also we introduced an attention network for visualizing how much each feature affected the prediction result. In this work, we introduce another text analysis technique and more detailed metadata. As a result, we have achieved much better performance as compared to our previous work. Besides, for better analyzing of our model, we introduce another visualization technique to show regions in an image that make its CTR better or worse. Our prediction model gives us useful suggestions for improving design of advertisements to acquire higher CTRs.

Packaging design has a pronounced effect on consumer purchase behavior and can be a critical factor in marketing. Despite the importance, there are very few studies that have investigated optimal designs. In this work, in order to analyze packaging designs and support designing processes, we propose a deep learning based method with ensemble learning to predict user preference for packaging design. For qualitative analysis, we visualize the feature maps from the prediction model. Moreover, we predict the mentioned frequencies of semantic attributes in the questionnaires, which represent impressions that users have. The experimental results suggest that in terms of user preference prediction, the proposed model achieves a correlation coefficient of 0.652 between the ground truth user preference scores and the predicted values. As for the semantic attributes prediction, the highest correlation coefficient reaches 0.75. Also the visualization successfully indicates key elements in designs.

Given the promising results obtained by deep-learning techniques in multimedia analysis, the explainability of predictions made by networks has become important in practical applications. We present a method to generate semantic and quantitative explanations that are easily interpretable by humans. The previous work to obtain such explanations has focused on the contributions of each feature, taking their sum to be the prediction result for a target variable; the lack of discriminative power due to this simple additive formulation led to low explanatory performance. Our method considers not only individual features but also their interactions, for a more detailed interpretation of the decisions made by networks. The algorithm is based on the factorization machine, a prediction method that calculates factor vectors for each feature. We conducted experiments on multiple datasets with different models to validate our method, achieving higher performance than the previous work. We show that including interactions not only generates explanations but also makes them richer and is able to convey more information. We show examples of produced explanations in a simple visual format and verify that they are easily interpretable and plausible.