When the competition in e-commerce rises, it is essential to ensure content is creative as well as effective in driving conversions. Using language generation from LLMs, we suggest a process that includes prompt engineering, multi-objective training and post-processing to make marketing texts engaging and effective for sales. We use a two-step process that applies sentiment modification, boosts diversity and adds calls to action. The way we test offline and online helps achieve a 12.5% increase in click through rate (CTR) and 8.3% increase in conversion rate (CVR). With this solution, creating texts can be automated and it will guide future improvements in live, customized user experiences.
Paper
References (57)
Scroll for more · 38 remaining