EDBT 2026 Demo / reviewers in the wild / expert
Sharadha Ramanan
dblp:270/0277
· DBLP profile ↗
2ranked-venue papers in the field
0as first author
2since 2021 · last 2023
0009-0009-5416-2869ORCID · corroborated
Domains — venue-derived; a paper can count in several
Big Data, Cloud & Distributed Data Systems · 2
| Year | Publication | Venue | Position |
|---|---|---|---|
| 2023 | ENCODE: Ensemble Contextual Bandits in Big Data Settings - A Case Study in E-Commerce Dynamic PricingabstractWe present ENCODE, an innovative ensemble-based Contextual Bandit (CB) model, engineered explicitly for dynamic pricing in large-scale e-commerce for kid’s clothing. ENCODE uniquely addresses the retailer’s multifaceted need for optimal pricing – both immediate and cumulative – subject to market-driven price triggers like competitor fluctuations and seasonal trends. The model integrates four cornerstone CB algorithms: LinUCB, Vowpal Wabbit, Contextual Thompson Sampling, and BayesUCB, delivering a unified solution for maximizing yield in alignment with business constraints.Our model’s originality manifests in its comprehensive treatment of complex, real-world pricing issues: from optimizing long-tail products with limited data to seamlessly managing an extensive array of products and catering to diverse customer segments. ENCODE also uniquely accommodates pricing strategies within product families, is highly responsive to shifts in competitor pricing, and accounts for the intricate interplay between related products in a non-stationary environment.Demonstrably effective, the model yielded a 13.8% improvement in margin through Contextual Thompson Sampling alone, with an additional 6% gain from considering inter-related products. In comparative analysis, ENCODE surpassed standalone CB algorithms by 19% in cumulative margins. It also boasts computational efficiency, optimized through dimensionality reduction and hyperparameter fine-tuning. Scalability is ensured through advanced cloud-based implementations. Hence, ENCODE stands as a comprehensive, highly scalable, and markedly effective solution to contemporary e-commerce pricing complexities. Srividhya Sethuraman, Uma Maheswari G, Siddhesh Thombre, Vikash Patel, Sharadha Ramanan |
IEEE Big Data | 6 |
| 2021 | Markdown Pricing For a Large Scale RetailerabstractA large European-based international discount store chain retailer, operating in more than 1000 stores across countries, wanted to make a transition from traditional cost-based markdown pricing to an optimal markdown pricing that maximizes the overall revenue as well as clears the inventory. Their regular pricing strategy was to price lowest among all their competitors since they are a discount store chain. Their markdown pricing problem is both critical and complex because there are many requirements to be satisfied, albeit simultaneously. The overall revenue had to be maximized as well as the given percentage of inventory was to be cleared by the end of the markdown period. Also, inventory had to remain in the stores till the end of the season so that the footfalls do not decrease and the markdown pricing was to be at the item group level for interrelated items. We developed a novel markdown solution by utilizing the retailers’ historical markdown performance data to come up with the markdown price estimates. Our approach was to obtain initial markdown price estimates for each item, apply ML/DL algorithms to forecast sales, compute price elasticities and build a nonlinear markdown price optimization system to recommend optimal prices. In this paper, we give the details and results for one category - the Clothing category. We did back testing for multiple periods to compare our sales forecasting model outputs using initial price, with their actuals. We have employed a distributed computing and parallel execution framework in cloud to obtain optimal markdown prices for products in the clearance season. Our final recommended price was greater by around 20% than the actual markdown prices for most items and this led to around 6% decrease in sales units while satisfying their inventory constraints. Our optimal pricing solution resulted in 10% increase in revenue for Clothing as compared to their actuals. Our markdown solution was later scaled to other categories as well as to stores in other countries. Vibhati Burman, Rajesh Kumar Vashishtha, Srividhya Sethuraman, Ganesh Radhakrishnan, Prashanth Ganesan, Suresh Kumar V, Sharadha Ramanan |
IEEE BigData | 7 |