How Alibaba’s AI Writer is Starting the Content Revolution

Alibaba Tech
3 min readFeb 22, 2018

E-commerce platforms are utilizing AI to generate engaging content for shoppers

Taobao, Alibaba’s huge e-commerce platform, dominates online sales in China and represents a significant and growing proportion of global online sales. To make the most of its long-term accumulation of content it has developed an AI writer that is revolutionizing online sales.

For Taobao’s mobile app, the move towards a content-based platform has become increasingly important within recent years. As opposed to only displaying products, a content-based platform has more ways at its disposal to describe products and enrich user experience. Content personalization and generation is vital to a content-based platform. Taobao’s AI Writer uses vast data resources and combines them with operational knowledge to create graphical and text content automatically.

During Alibaba’s 2017 Double 11 Global Shopping Festival, the AI Writer created personalized content displayed in real time on the Taobao and Tmall (Alibaba’s verified seller retailer) platforms. This type of personalized content provided smart selling points to really engage shoppers.

Content generated by AI writer for Taobao Mobile pages

Smart selling points

A great selling point picks out the core highlight of a product or a promotional site. Selling points can increase click rates for sellers and businesses. Traditional methods of producing selling points, though, have some drawbacks. These drawbacks include:

1. Due to issues such as time constraints and the vast quantity of selling points being generated, individual products or sites often have less than three selling points. This can sometimes lead to incompatibility between selling points and actual merchandise.

2. Users also have varying concerns and interests when looking at a product. Some users care about value for money, while some are focused on product quality. When a product only has a few selling points, it is unable to appeal to the concerns and interests of the target market.

For 2017’s Double 11, the AI Writer tried to generate personalized selling points in real time on the main site and some Taobao/Tmall landing pages. Online experiment results show that smart selling point results showed significant improvement in Click Through Rate (CTR).

What are the benefits?

No matter what the product, smart selling points can solve problems by discovering hidden selling points relating to that commodity, laying these selling points out into content that meets a set word count limit, and recommending the product to users. Unique features of smart selling point generation include:

1. Discovers user preferences: These preferences are based on user behavioral data. This data includes product history and search query log.

2. Generates selling points: Selling points are derived from sources including information from sellers, product titles, and descriptions.

3. Personalizes selling point sequence: The PairXNN model is proposed for this process.

How well does it work?

For 2017’s Double 11, smart selling points were shown in online scenarios including the Double 11 site landing page, the Double 11 category landing page, the mobile landing page, the Tmall landing page, and the Tmall customer category landing page.

The data shows that smart selling points really do have a clear effect. UCTR and PCTR both rose by between 6% to 14%. How much results had improved varied depending on the application scenario. Analysis suggests this is because of the different requirements different scenarios have for online promotion.

(Original article by Hu Jinxin胡金鑫 & Lin Guli林古立)

Alibaba Tech

First hand, detailed, and in-depth information about Alibaba’s latest technology → Search “Alibaba Tech” on Facebook



Alibaba Tech

First-hand & in-depth information about Alibaba's tech innovation in Artificial Intelligence, Big Data & Computer Engineering. Follow us on Facebook!