# Intelligent Online Selling Point Extraction for E-Commerce Recommendation

**Authors:** Xiaojie Guo, Shugen Wang, Hanqing Zhao, Shiliang Diao, Jiajia Chen, Zhuoye Ding, Zhen He, Yun Xiao, Bo Long, Han Yu, Lingfei Wu  
**arXiv:** [2112.10613](https://arxiv.org/abs/2112.10613)  
**Date:** 2021-12-16  
**Summary page:** https://www.summarizepaper.com/en/paper/2112.10613v1/

## Key points

- Significant advancements in automatic product description generation for e-commerce in the past decade
- Importance of adapting generated descriptions dynamically to suit e-commerce platforms' diverse services
- Human experts currently responsible for writing selling points, resulting in high costs
- Potential cost reduction by using machines to automatically generate product selling points
- Introduction of the Intelligent Online Selling Point Extraction (IOSPE) system in JD.com, a large-scale e-commerce platform
- IOSPE has become a core service for 62 key categories of products, covering over 4 million products since July 2020
- IOSPE has generated over 0.1 billion selling points, saving human labor and scaling up the selling point creation operation
- IOSPE-generated selling points have increased click-through rate (CTR) by 1.89% and extended average customer engagement by more than 2.03%
- Informative product descriptions are crucial for user experience and decision-making on e-commerce platforms
- Challenges include adapting description writing patterns according to different services and limited display space on mobile devices' homepages
- The Intelligent Online Product Selling Point Extraction (IOPSE) system addresses these challenges with a hybrid framework of natural language mining and generation
- IOPSE consists of selling point extraction and personalized assignment modules
- Aims to generate concise yet informative selling points while supporting personalization
- Development and deployment of IOSPE and IOPSE have significantly improved efficiency and effectiveness in creating selling points for JD.com's e-commerce platform

## Summary

In the past decade, there have been significant advancements in automatic product description generation for e-commerce. As e-commerce platforms offer a diverse range of services, it is crucial to adapt the patterns of generated descriptions dynamically. One important type of product description is the selling point, which should be concise yet convey key information and be eye-catching to readers. Currently, human experts are responsible for writing these selling points, resulting in high costs for their creation and maintenance. However, these costs can be significantly reduced if machines can automatically generate product selling points. This paper presents the Intelligent Online Selling Point Extraction (IOSPE) system that has been developed and deployed to serve the recommendation system in JD.com, a large-scale e-commerce platform. Since July 2020, IOSPE has become a core service for 62 key categories of products, covering more than 4 million products. The system has generated over 0.1 billion selling points, effectively scaling up the selling point creation operation and saving human labor. The IOSPE-generated selling points have shown significant improvements for JD.com. They have increased the click-through rate (CTR) by 1.89% and extended the average duration customers spend on products by more than 2.03% compared to previous practices. These improvements are particularly noteworthy considering the scale of JD.com's operations. Informative product descriptions play a critical role in providing a desirable user experience on e-commerce platforms as they help customers make informed decisions and assist sellers in promoting their products unlike physical stores where salespersons interact directly with customers who rely heavily on textual and pictorial product descriptions to provide information and persuade users to make purchases . Due to the diversity of services offered by e-commerce platforms, it is essential to adapt description writing patterns accordingly; for instance, on mobile devices' homepages with limited display space, product descriptions need to be concise while existing approaches often generate comprehensive product information resulting in lengthy descriptions without supporting personalization . To address these challenges ,the Intelligent Online Product Selling Point Extraction (IOPSE) system has been developed and deployed utilizing a hybrid framework of natural language mining and generation consisting of two main components: selling point extraction and personalized assignment . The selling point extraction module comprises three main steps:selling point coarse screening ,selling point generation ,and selling point fine screening . This system aims to overcome the limitations of existing approaches by generating concise yet informative selling points as well as supporting personalization . Overall ,the development and deployment of IOSPE and IOPSE have significantly improved the efficiency and effectiveness of creating selling points for JD . com's e - commerce platform leading to increased CTRs (click - through rates )and longer customer engagement with products demonstrating the value automated product description generation brings about in large - scale e - commerce platforms .