Articles from Source: Code-as-Craft

Etsy is enhancing product recommendations using Contrastive Reinforcement Learning. 🌟 The goal is to better understand unique product details beyond basic descriptions. This approach helps surface items that match buyer preferences, even when those preferences aren’t explicitly stated. The challenge lies in transforming raw seller data into structured insights while capturing the creativity behind each listing. 🛍️ By leveraging buyer engagement signals, Etsy fine-tunes language models to...

🚀 Etsy is enhancing the buyer-seller connection through its Ads Search ranking model. By introducing the Multigate Mixture of Experts (MMoE) and using add-to-cart as an auxiliary signal, they aim to better predict purchase intent. This model improvement helps surface more relevant listings for buyers while ensuring sellers reach interested customers. Stay tuned for more updates on how these changes benefit the marketplace! 🛍️✨ #Etsy #AdsInnovation #MachineLearning #Ecommerce...

Etsy has transitioned its database management from a sharded MySQL architecture to Vitess, enhancing scalability and resilience. Originally, Etsy's system relied on an index database, which posed risks of outages and complex manual scaling. By adopting Vitess in 2018, queries are now managed more efficiently, reducing reliance on a single point of failure. This migration, completed over five years, introduces vindexes for improved sharding strategies and allows for cross-shard queries,...

Etsy's Search Relevance team is enhancing the shopping experience by tackling search accuracy. They've developed a Semantic Relevance Evaluation and Enhancement Framework using large language models (LLMs). This approach captures how well listings align with buyer intent, complementing traditional engagement metrics. Key components include high-quality human-curated data, semantic relevance models, and real-time applications to improve search outcomes for both buyers and sellers. Stay tuned...

Etsy has significantly enhanced its experimentation process with the evolution of CUPED, reducing average experiment duration by 3 days! 🕒 This variance reduction technique uses pre-experiment data to improve accuracy and speed in measuring outcomes. The introduction of CUPAC further cuts variance, leading to quicker decision-making. These advancements allow teams to conduct more experiments, ultimately enhancing the buyer and seller experience on the platform. #Etsy #CUPED #Experimentation...

Etsy has made significant strides in reducing experiment durations with the evolution of its CUPED technique. 🛍️ CUPED, or Controlled-Experiment Using Pre-Experiment Data, has decreased average experiment time by 3 days by leveraging historical data to enhance accuracy and speed. 📊 The introduction of CUPAC further refines this process, yielding an additional 10% variance reduction, allowing more experiments to be conducted yearly. 🧪 Etsy continues to explore new techniques to enhance its...

🚀 Etsy's search team has implemented the Speculation Rules API (SRA) to enhance product page performance significantly. By prefetching pages, they've achieved improvements of 20-24% across various metrics. Key highlights include two methods of prefetching: traditional and speculative, with SRA offering advantages in caching and performance. The team also learned valuable lessons about user interaction and analytics during the implementation process. #Etsy #WebPerformance #SRA #TechInnovation...

Etsy has over 100 million unique items from 5 million sellers, showcasing human creativity. However, this vast inventory presents challenges in accurately capturing product attributes, from standard to niche details. The platform collects both structured data, which is easy to process, and unstructured data, which is harder to interpret. To address this, Etsy is leveraging LLMs to transform unstructured data into structured attributes, enhancing search and buyer experiences. This innovative...

Etsy is enhancing its experimentation capabilities by empowering teams to define their own segmentations. 📊 Previously limited to 13 hard-coded segmentations, the new Segmentation Library allows for 3x more custom options, leading to faster insights. Experimenters can quickly analyze data tailored to user groups, improving decision-making. This innovation has already saved teams 2-5 hours monthly. ⏳ #Etsy #Experimentation #DataInsights #Segmentation #Innovation

Etsy is enhancing buyer experiences by using large language models (LLMs) to create detailed buyer profiles based on shopping behaviors. 🛍️ These profiles capture individual interests, helping to tailor search results for nearly 90 million users while maintaining privacy compliance. Users have the option to opt-out of profile generation. 🔍 Technical improvements have reduced the time for profile generation from 21 days to just 3 days, making personalization more efficient and cost-effective....

Exploring prompt engineering in AI-assisted onboarding at Etsy reveals both benefits and limitations. The study focused on how well LLMs provide reliable answers to Etsy-specific questions, particularly in the Travel & Entertainment domain. Initial findings indicate that concise prompts improve answer accuracy. However, some instances showcased LLM "hallucinations," emphasizing the need for careful prompt design. For more insights, check out the full article! 📝🤖 #AIEducation #Etsy...