Marketplace Algorithm Insights

Revolutionizing E-commerce Search: When LLMs Meet Hierarchical Taxonomies Researchers at eBay have developed a fascinating Chain-of-Thought approach that's transforming how e-commerce platforms categorize user queries. Instead of relying solely on traditional behavioral signals like click-through rates, this method leverages the semantic understanding of Large Language Models. How it works under the hood: The system treats query categorization as a tree traversal problem. Starting from the root of a product taxonomy, it uses an LLM to score the semantic relevance of each category branch on a 1-10 scale. The algorithm then applies dynamic thresholding- mapping scores to a standard normal distribution and pruning categories that fall below configurable selection and minimum thresholds. What makes this approach particularly clever is its breadth-first search strategy. At each taxonomy level, it only explores the most semantically promising branches, visiting just 1.7% to 24.8% of total category nodes while maintaining high accuracy. Key technical innovations: - Context-aware scoring that incorporates user intent (buying vs. browsing vs. seeking accessories) - Relative thresholding that adapts to score distributions at each node - Scalable variants using k-NN embedding filters for high-volume deployment - Built-in taxonomy diagnostics that identify structural gaps The results are impressive: Testing with models like Mixtral-8x7B showed 89.8% improvement in F1 scores over embedding-based baselines, with significant gains in both precision and recall across human judgment datasets. Beyond performance gains, this approach offers valuable insights for taxonomy optimization- automatically flagging categories like "Designer Sunglasses" that may be buried too deep in existing hierarchies. This research demonstrates how modern NLP can enhance traditional e-commerce systems while providing interpretable, actionable insights for platform optimization.

Your catalog cannot be the same across marketplaces — especially in fashion. On Myntra → presentation is everything. Model imagery, styling angles, fabric detail shots, size & fit clarity, and trend relevance directly impact conversion. It’s fashion-first and discovery-driven. On Amazon search intent dominates. Keyword-rich titles, structured attributes, and reviews drive ranking. On Flipkart strong USPs, price perception, and clear bullets influence buying decisions. Same product. Different algorithm. Different customer mindset. Brands must focus on optimizing their catalog based on each marketplace’s algorithm to drive better visibility, higher conversion, and stronger results.

Amazon is finally acknowledging something the industry has been feeling for years: visibility isn’t just driven by keywords, it’s driven by context. With the rollout of Prompt-Based Advertising, Amazon is opening the door to a new way of understanding how products surface inside AI-generated conversations. This isn’t about replacing PPC. It’s about revealing how the algorithm interprets intent, relevance and product data beyond traditional keyword matching. Here’s why this matters if you care about performance, ROI and long-term visibility: → Prompts let us track a new layer of discoverability. We can now see how a model reads a listing, not just how it matches a keyword. That means new insights into relevance, value signals, attributes and category fit. → This shifts the conversation from “what people search” to “what the model understands.” If your product data isn’t structured or consistent, the prompt won’t surface you, even if your bids are strong. → Keyword-only PPC strategies will start missing context-based opportunities. The model weighs factors like attributes, benefits, safety, availability, reviews, and consumer constraints. And that impacts ad placement long before the auction. → It’s a major step forward for optimizing listings. Prompt behavior will expose data gaps, inconsistencies, and relevance issues you can fix to increase organic and paid visibility. Now, an important nuance: This is early. Early means unstable performance, shifting logic, and a learning curve for advertisers. But early also means insight, real insight, into how Amazon’s models interpret categories in 2026 and beyond. Zooming out, this isn’t just an ad update. It’s Amazon giving us visibility into the future mechanics of retail media: → AI will define relevance → Prompts will influence category dynamics → And listings with strong data structures will outperform those relying only on bid strategy My take? Track it. Test it. Pay attention to how prompts interpret your catalog. Because your future visibility won’t be built on keywords alone, it’ll be built on how well your data speaks the language of AI. — G #amazonads #ppc #retailmedia #ecommerce #aivisibility #amazonppc #roi #growthstrategy #productdata #listingoptimization #retailinnovation

What creating winning products in Amazon's Stores really means. For years, the hardest part of selling on Amazon wasn't launching a product, it was choosing what to launch in the first place. You had to intuit demand. You had to guess at customer needs. And most sellers made decisions with incomplete data, hoping their instincts were right. In our interview at Amazon Accelerate 2025, Zhen Li, the product manager behind Product Opportunity Explorer, shared how Amazon is reinventing the pre-listing journey so sellers can identify real market demand before committing inventory, capital, or time. Here are the things every seller should understand: 1️⃣ Customer-first research is the new foundation. The biggest challenge sellers face is understanding what customers actually want before they build anything. They are moving the starting point from "What product should I launch?" to "What problems are customers experiencing right now?"  ✅ This flips the entire process, because instead of discovering niches by accident, sellers can now identify demand signals directly from Amazon's space. 2️⃣ Product Opportunity Explorer is expanding into new marketplaces and categories. Sellers will now get deep, multi-market insights across the US and newer marketplaces, from Europe to India to emerging regions.  ✅ That means you can: - Spot demand globally - Validate opportunities before expanding - Compare customer behavior across regions Amazon is giving sellers the ability to test their ideas against real customer signals. 3️⃣ "Identity Market Demand" is coming, and it's a huge unlock. Li introduced a new feature Amazon is rolling out: Identity Market Demand, which gives sellers access to the why behind shopper behavior.  ✅ Think of it as Amazon telling you: "This is the customer segment behind the demand, here's what they care about, and here's what they can't find today." This is the deepest, most specific customer insight Amazon has ever surfaced to sellers. 4️⃣ Sellers can now access expert-level insights before they launch anything. One thing was clear: Amazon wants sellers to build better products before they exist. ✅ With the new tools inside Product Opportunity Explorer and AI, sellers accelerate every step of pre-listing: - Understand unmet customer needs - Spot gaps in the market - Listing strategy (before a listing exists)  - Market-fit validation - Idea refinement - Design decisions based on data This change is the beginning of a product development cycle built with customer insights from day one. So what does this mean for sellers? Product creation on Amazon is becoming less about guesswork and more about guided discovery. Less about chasing trends and more about understanding needs. Less about intuition and more about intelligence. 👀 Watch the full interview on Carbon6 's channel: https://lnkd.in/gnsVck42   ♻️ Repost so more sellers have these changes on their radar. #AmazonAccelerate #Amazon #ProductDevelopment #AmazonFBA

Ranking might sound technical, but they sit at the core of how marketplaces like OLX deliver relevant experiences to millions of users every day. In a recent tech blog, the OLX Engineering team shared their journey in advancing their ranking models, with a focus on two key learning-to-rank algorithms: RankNet and LambdaRank. - RankNet represents a meaningful step beyond traditional classification models. Instead of predicting whether an item is relevant in isolation, RankNet learns to compare pairs of items and predict which should be ranked higher. This pairwise approach allows the model to directly optimize for ordering, making it far better suited for ranking problems than standard classifiers. - LambdaRank builds on this idea and pushes it further. While RankNet optimizes pairwise comparisons, LambdaRank directly incorporates ranking quality metrics—such as NDCG—into the learning process. By adjusting gradients based on how much a ranking swap would affect the final metric, LambdaRank aligns model training more closely with real business goals, leading to stronger ranking performance in practice. By moving from simple classification to learning-to-rank approaches, the team can surface the most relevant content more effectively and increase overall marketplace effectiveness. RankNet and LambdaRank offer a clear and practical lens into how modern ranking systems evolve—and why learning to rank has become such a powerful tool for real-world platforms. #DataScience #MachineLearning #Ranking #LearningtoRank #Recommendation #Relevance #SnacksWeeklyonDataScience – – –  Check out the "Snacks Weekly on Data Science" podcast and subscribe, where I explain in more detail the concepts discussed in this and future posts:    -- Spotify: https://lnkd.in/gKgaMvbh   -- Apple Podcast: https://lnkd.in/gFYvfB8V    -- Youtube: https://lnkd.in/gcwPeBmR https://lnkd.in/gVt9D2rW