Abstract

This paper considers the problem of semantically typing pet products using only independent and crowdsourced reviews provided for them on e-commerce websites by customers purchasing the product, rather than detailed product descriptions. Instead of proposing new methods, we consider the feasibility of established text classification algorithms in support of this goal. We conduct a detailed series of experiments, using three different methodologies and a two-level pet product taxonomy. Our results show that classic methods can serve as robust solutions to this problem, and that, while promising when more data is available, language models and word embeddings tend both to be more computationally intensive, as well as being susceptible to degraded performance in the long tail.

Date

October 31, 2023

Authors

Xinyu Liu, Tiancheng Sun, Diantian Fu, Zijue Li, Sheng Qian, Ruyue Meng, Mayank Kejriwal

Book

Iberoamerican Knowledge Graphs and Semantic Web Conference

Pages

151-167

Publisher

Springer Nature Switzerland