Matching Products with Deep NLP Models
Leonidas Akritidis, Panayiotis Bozanis · 2023
Following the explosive data growth that is presently taking place on the Web, the eCommerce industry has evolved towards enterprises and services that collect product-oriented data from multiple external sources. Since the majority of these sources are usually uncontrolled and independent of each other, they provide their information in a diverse manner, rendering the identification of products a difficult task. The problem becomes particularly challenging by considering the native sparseness and high dimensionality of text, the specific peculiarities of product data, the large data volume and the dynamic nature of the involved applications. Despite the uncontested significance of the problem, there is a lack of works in the relevant literature that confront it by taking into account all the aforementioned challenges. In this paper, we summarize these challenges and provide some useful insights on how they can be effectively tackled. We also present several components of a preliminary system that is being developed to accurately and efficiently identify product entities in diverse data originating from multiple external sources.