Descriptions from the customers: Comparative analysis of review-based product description generation methods

Slava Novgorodov, Ido Guy, Guy Elad, Kira Radinsky

Research output: Contribution to journalArticlepeer-review

1 Scopus citations

Abstract

Product descriptions play an important role in the e-commerce ecosystem. Yet, on leading e-commerce websites product descriptions are often lacking or missing. In this work, we suggest to overcome these issues by generating product descriptions from user reviews. We identify the set of candidates using a supervised approach that extracts review sentences in their original form, diversifies them, and selects the top candidates. We present extensive analyses of the generated descriptions, including a comparison to the original descriptions and examination of review coverage. We also perform an A/B test that demonstrates the impact of presenting our descriptions on user traffic.

Original languageEnglish
Article number44
JournalACM Transactions on Internet Technology
Volume20
Issue number4
DOIs
StatePublished - 1 Nov 2020
Externally publishedYes

Keywords

  • Deep multi-task leaning
  • electronic commerce
  • language generation
  • user-generated content

ASJC Scopus subject areas

  • Computer Networks and Communications

Fingerprint

Dive into the research topics of 'Descriptions from the customers: Comparative analysis of review-based product description generation methods'. Together they form a unique fingerprint.

Cite this