Multimodal Representation Learning for Retail Product Ontology

点击翻译按钮获取中文摘要

In the past decade, retailers have experienced a massive surge of information ranging from how their products**sell and which products their customers prefer, to how their promotional campaigns perform. This dramatic**change in data availability was mainly due to advances in the Internet-of-Things and data warehousing,**allowing billions of customer transactions to be logged in big data centers. This abundance of information,**however, cannot easily translate into benefits for the retailer and customer unless it is rigorously analyzed to**guide business decisions. It comes as no surprise that Big Data analytics are now becoming a business**imperative in the realm of retail, with more and more retailers applying data mining technologies to transform**their processes, from product recommendations to sales forecasting. Still, a large amount of the raw data**available to retailers is incomplete and too generic to justify meaningful analysis. For example, many products**are characterized only by name, type, and a brief textual description, whereas attributes such as typical**purchase patterns, the demographics to which the product appeals, and whether it is a luxury brand are usually**absent. It is entirely up to the manual effort of human experts to identify these "hidden" attributes, but this**incurs significant costs and delays to the retailer or any third party that offers services on big data analytics. The**goal of this work is to develop novel methodologies based on machine learning technologies to automatically**discover these hidden attributes with minimal human intervention. These methods will learn non-obvious**information by mining text and images associated with each product, which will involve a combination of**techniques from Natural Language Processing and Computer Vision. Through this process one can produce**rich product ontologies that allow for effective market-driven analytics.