Stage-oe-small.jpg

Inproceedings3891

Aus Aifbportal
Wechseln zu:Navigation, Suche


Quantifying Explanations of Neural Networks in E-Commerce Based on LRP


Quantifying Explanations of Neural Networks in E-Commerce Based on LRP



Published: 2021 Juli

Buchtitel: Proceedings of Machine Learning and Knowledge Discovery in Databases: Applied Data Science Track - European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD'21)
Verlag: Springer

Referierte Veröffentlichung

BibTeX

Kurzfassung
Neural networks are a popular tool in e-commerce, in partic- ular for product recommendations. To build reliable recommender sys- tems, it is crucial to understand how exactly recommendations come about. Unfortunately, neural networks work as black boxes that do not provide explanations of how the recommendations are made. In this paper, we present TransPer, an explanation framework for neu- ral networks. It uses novel, explanation measures based on Layer-Wise Relevance Propagation and can handle heterogeneous data and complex neural network architectures, such as combinations of multiple neural networks into one larger architecture. We apply and evaluate our frame- work on two real-world online shops. We show that the explanations provided by TransPer help (i) understand prediction quality, (ii) find new ideas on how to improve the neural network, (iii) help the online shops understand their customers, and (iv) meet legal requirements such as the ones mandated by GDPR.

Download: Media:sub_183.pdf

Projekt

TransPer



Forschungsgruppe

Web Science


Forschungsgebiet

Maschinelles Lernen, Künstliche Intelligenz