Automatic Product Copywriting for E-commerce

Abstract

Product copywriting is a critical component of e-commerce recommendation platforms. It aims to attract users' interest and improve user experience by highlighting product characteristics with textual descriptions. In this paper, we report our experience deploying the proposed Automatic Product Copywriting Generation (APCG) system into the JD.com e-commerce product recommendation platform. It consists of two main components: 1) natural language generation, which is built from a transformer-pointer network and a pre-trained sequence-to-sequence model based on millions of training data from our in-house platform; and 2) copywriting quality control, which is based on both automatic evaluation and human screening. For selected domains, the models are trained and updated daily with the updated training data. In addition, the model is also used as a real-time writing assistant tool on our live broadcast platform. The APCG system has been deployed in JD.com since Feb 2021. By Sep 2021, it has generated 2.53 million product descriptions, and improved the overall averaged click-through rate (CTR) and the Conversion Rate (CVR) by 4.22% and 3.61%, compared to baselines, respectively on a year-on-year basis. The accumulated Gross Merchandise Volume (GMV) made by our system is improved by 213.42%, compared to the number in Feb 2021.

The potential of generative AI for personalized persuasion at scale

“…This was done to statistically account for variance in our outcome measures attributable to individual differences unrelated to people’s ad preferences (e.g., extraverts potentially giving higher scores on rating scales, or individual variation in the amount of money they can afford to spend when purchasing a smartphone). Specifically, we calculated the residuals for each outcome measure by regressing the targeted outcome (e.g., WTP for the phone advertised with the Openness ad) on the equivalent outcomes for the other traits (e.g., WTP for the phone advertised with the Conscientiousness, Extraversion and Agreeableness ads; see 43 for a similar approach). This allows us to isolate the unique variance in a participant’s preference that is unique to each specific ad (as opposed to the variance that is shared among all of them).…”

“…For example, ad agencies have started to employ LLMs to create generic “ad copy” that can be published quickly 42 . Similarly, recent research suggests that automatically generated product descriptions in combination with human screening can result in improved click-through and conversion rates in e-commerce sites 43 . While these developments speak to the ability of LLMs to generate generic persuasive content, they do not offer any insights into (1) whether LLMs can create persuasive messages that are personalized to the needs and motivations of an individual and (2) whether doing so indeed makes these persuasive attempts more influential.…”

The Potential of Generative AI for Personalized Persuasion at Scale

“…For example, ad agencies have started to employ LLMs to help create generic "ad copy" that can be published quickly (36). Similarly, recent research suggests that automatically generated product descriptions result in improved click-through rates and conversion rates on a large e-commerce website (37). While these developments speak to the ability of LLMs to generate simple yet relevant content, they do not offer any insights into whether LLMs can create messages that are personalized to and therefore more influential for specific psychological characteristics.…”

Paraphrase Generation Model Integrating Transformer Architecture, Part-of-Speech Features, and Pointer Generator Network

“…Ravuru et al proposed combining VAE [12] and the PGN [8] and applying them to the task of paraphrasing [16]. Zhang et al combined the Transformer architecture with the PGN [17]. Zhang et al suggested that writing product copy is similar to abstract extraction.…”