题  目：Heterogeneous Preferences toward Video Ads: An Interpretable Deep Learning Approach with Multi-level Attentions（对视频广告的异质性偏好：一种具有多层注意力机制的可解释深度学习方法）

时  间：2024年10月16日（星期三），13:30-15:00

地  点：后主楼1722会议室

主讲人：张  铄 副教授（上海交通大学）

主持人：龚诗阳 教  授（766全讯白菜网999策略白菜手机论坛）

摘要：

Given the current popularity of user generated content, many e-commerce platforms have also incorporated the feature of recommending content to their users similar to how users normally browse content on social media. In particular, users will be exposed to a series of ads and they may decide on whether to continue scrolling down to see more or click through the current ad to visit the store and buy the product advertised in the ad. We develop an interpretable deep learning approach with a multi-level attention mechanism. To train the model, we collect a detailed dataset from a major e-commerce platform in China with a complete set of users, video and their interaction data. Overall, we find that demographics are less important in determining users’ preferences toward video ads compared to their past purchase history, contradicting to firms’ common practices of tagging customers based on these descriptors. Text and audio in a video will only affect users’ browsing preferences at the beginning and the end (probably when they need to first learn and recall the product), while the visuals contribute the most to their preferences in the middle. The results will help firms understand users’ heterogeneous content preferences better but more importantly improve their future creation of video ads for their target audience.

演讲嘉宾简介：

张铄，上海交通大学营销系副教授，于2014年在上海交通大学获得经济学学士学位，2020年在美国圣路易斯华盛顿大学获得经济学博士学位。张铄主要的研究兴趣是应用包括回归分析，结构模型，机器学习在内的量化方法构建的实证模型，研究个体在数字经济平台的决策，公司的定价策略，以及研发数据驱动的营销策略。张铄的研究成果发表/接收于包括Marketing Science，Management Science在内的国际学术期刊。张铄入选上海海外高层次人才引进计划，浦江人才计划，主持国家自然科学基金青年项目，参与自科创新群体项目，国际重点项目及重点项目等。