Online-to-offline (O2O) omnichannel retail has grown very fast, but in the case of food, differences between product illustrations posted online and the actual goods that consumers receive have been a persistent problem. Hence, this pioneering study proposes a two-stage AI deep-learning method for mitigation of the quality risk posed to O2O omnichannel operations by O2O food dissimilarity, and investigates what information should be disclosed online to minimize the gap between O2O consumer expectations and perceptions, based on information theory and expectation-confirmation theory. Its empirical results reveal, first, that the proposed method can successfully assess the similarity between offline versions of products and the online images utilized for marketing them; and second, that restaurants can use food-similarity records to determine which online food-marketing images are most likely to achieve positive expectation confirmation. Lastly, it recommends that online food platforms include multiple, varied images of most heterogeneous foods, as a further means of avoiding O2O dissimilarity problems.

线上到线下（O2O）的全渠道零售增长非常快，但就食品而言，在线发布的产品插图与消费者收到的实际商品之间的差异一直是一个问题。因此，这项开创性的研究提出了一种两阶段的 AI 深度学习方法，以减轻 O2O 食品差异对 O2O 全渠道运营带来的质量风险，并调查应在线披露哪些信息以最大程度地减少 O2O 消费者期望和感知之间的差距，基于信息论和期望确认理论。其实证结果表明，首先，所提出的方法可以成功评估产品的离线版本与用于营销它们的在线图像之间的相似性；其次，餐厅可以使用食物相似性记录来确定哪些在线食品营销图像最有可能获得积极的预期确认。最后，它建议在线食品平台包含大多数异类食品的多种不同图像，作为避免 O2O 差异问题的进一步手段。