Elamur Research⌕
☰
home / papers / 2026 / Grounded Multimodal In-Context Learning for Product Weight Estimation at Scale in E-commerce

Grounded Multimodal In-Context Learning for Product Weight Estimation at Scale in E-commerce

Paper · 2026 · Bhavuk Singhal, Arsh Keshari and Ravindra Kumar Yadav

Summary

E-commerce logistics needs product weight to set shipping charges, but sellers rarely state it reliably. Items in the same fine-grained category can also look alike while differing by kilograms. The authors, from Meesho, treat weight estimation as grounded multimodal reasoning. A large vision-language model, Gemini 2.5 Flash, predicts one of 50 discrete 500 g buckets (500 g–25 kg) from a product's image and text, using few-shot in-context examples.

The examples are retrieved automatically from historical listings. Only products with many delivered orders are used, since their declared weights are probably correct. The examples are then sampled so that their bucket frequencies match the sub-category's real weight distribution. The approach was tested on an 18-month corpus spanning 14 sub-categories, with weights verified by experts. It beat multimodal KNN baselines and random exemplar selection.

In a production pilot covering millions of listings, exact-bucket accuracy roughly doubled and shipping revenue leakage fell by about 22%. Because the method needs no retraining, new categories can be added quickly. This makes it a cheaper alternative to manual audits or weighing hardware.

Key points

  • Data and baseline:

    • The evaluation uses a chronological split: 12 months form the retrieval pool and the next 6 months are the test set.
    • The baseline embeds image and text with FLAVA, searches neighbours with FAISS, and predicts using either the Top-1 neighbour or a Top-5 average.

  • Number of exemplars: Accuracy grows roughly logarithmically with N_shot. Beyond N_shot = 50, gains fall below 0.5% while cost and latency keep rising non-linearly, so 50 shots was chosen.

  • Backbone comparison (accuracy / normalized cost):

    ModelAccuracyNormalized cost
    Gemini 2.5 Flash73.81%1.0x
    Gemini 1.5 Pro74.23%2.57x
    GPT-4o73.89%5.15x
    GPT-4o mini66.45%0.81x
    LLaVA-v1.6-34B56.24%0.15x
    Qwen2-VL-72B59.13%0.13x

  • Production pilot: The system ran on four sub-categories (Compost & Fertilizers, Kitchen Storage, Mops with Bucket, Spice Racks), with these changes:

    • accuracy rose from ~37% to ~71%;
    • the undercharging rate dropped from ~70% to ~34%;
    • underestimation by two or more buckets fell from ~40% to ~5%;
    • revenue leakage decreased by ~22%.

  • Where the gains come from: The biggest improvements were in heavy buckets, where underestimation costs the most.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 3The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,320 worksVision & language · 657 works

Cite

Cite as Bhavuk Singhal, Arsh Keshari and Ravindra Kumar Yadav. 2026. Grounded Multimodal In-Context Learning for Product Weight Estimation at Scale in E-commerce. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-industry.41

@inproceedings{singhal2026grounded,
  title = {{Grounded Multimodal In-Context Learning for Product Weight Estimation at Scale in E-commerce}},
  author = {Bhavuk Singhal, Arsh Keshari and Ravindra Kumar Yadav},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-industry.41},
  url = {https://doi.org/10.18653/v1/2026.acl-industry.41}
}

References

27 references, 14 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Bain & Company. 2025. Ai’s trillion-dollar opportunity. https://www.bain.com/insights/ais-trillio n-dollar-opportunity-tech-report-2024/. ↗
  2. Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. 2023a. Visual instruction tuning. Advances in neural information processing systems, 36:34892– 34916.
  3. Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. 2023b. Visual instruction tuning. Preprint, arXiv:2304.08485.
  4. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, and 1 others. 2020. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901.
  5. Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, and 1 others. 2025. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. arX…
  6. Wenliang Dai, Junnan Li, Dongxu Li, Anthony Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale N Fung, and Steven Hoi. 2023. Instructblip: Towards general-purpose vision-language models with instruction tuning. Advances in neural information processing systems, 36:49250–49267.
  7. Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia, Jingjing Xu, Zhiyong Wu, Baobao Chang, and 1 others. 2024. A survey on in-context learning. In Proceedings of the 2024 conference on empirical methods in natural language processing, pages 1107–1128.
  8. Matthijs Douze, Alexandr Guzhva, Chengqi Deng, Jeff Johnson, Gergely Szilvasy, Pierre-Emmanuel Mazaré, Maria Lomeli, Lucas Hosseini, and Hervé Jégou. 2025. The faiss library. IEEE Transactions on Big Data.
  9. Eshopbox. 2025. Minimizing shipping overcharges: A guide to tackling weight discrepancies. https://www.eshopbox.com/blog/minimizing-shipp ing-overcharges. ↗
  10. Jiaying Gong and Hoda Eldardiry. 2024. Multi-label zero-shot product attribute-value extraction. In Proceedings of the ACM Web Conference 2024, pages 2259–2270. find
  11. Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, and 1 others. 2024. Gpt-4o system card. arXiv preprint arXiv:2410.21276.
  12. Rattapol Kasemrat and Tanpat Kraiwanit. 2024. Benchmarking machine learning models for predictive analytics in e-commerce. Available at SSRN 4832967. find
  13. Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. 2023. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. In International conference on machine learning, pages 19730–19742. PMLR.
  14. Jiachang Liu, Dinghan Shen, Yizhe Zhang, William B Dolan, Lawrence Carin, and Weizhu Chen. 2022. What makes good in-context examples for gpt-3? In Proceedings of Deep Learning Inside Out (DeeLIO 2022): The 3rd workshop on knowledge extraction and integration for deep learning architectures, pages 10…
  15. Yao Lu, Max Bartolo, Alastair Moore, Sebastian Riedel, and Pontus Stenetorp. 2022. Fantastically ordered prompts and where to find them: Overcoming few-shot prompt order sensitivity. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), p…
  16. Aristides Milios, Siva Reddy, and Dzmitry Bahdanau. 2023. In-context learning for text classification with many labels. arXiv preprint arXiv:2309.10954. find
  17. Ori Ram, Yoav Levine, Itay Dalmedigos, Dor Muhlgay, Amnon Shashua, Kevin Leyton-Brown, and Yoav Shoham. 2023. In-context retrieval-augmented language models. Transactions of the Association for Computational Linguistics, 11:1316–1331.
  18. Redseer. 2024. India’s festive e-commerce market 2024: Unveiling growth trends and emerging opportunities. https://redseer.com/reports/indias-festi ve-e-commerce-market-2024/. ↗
  19. Ohad Rubin, Jonathan Herzig, and Jonathan Berant. 2022. Learning to retrieve prompts for in-context learning. In Proceedings of the 2022 conference of the North American chapter of the association for computational linguistics: human language technologies, pages 2655–2671.
  20. Shiprocket. 2025. How to reduce weight discrepancies – hacks for 2026. https://www.shiprocket.in/ blog/reduce-weight-discrepancies/. ↗
  21. Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela. 2021. Flava: A foundational language and vision alignment model. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 15617–15629. find
  22. Ankush Verma, Chetan Nagar, Sharda Haryani, and Sumit Jain. 2025. Prediction of e-commerce shoppers’ purchasing intention using knn algorithm. In International Conference on Recent Advancements and Modernisations in Sustainable Intelligent Technologies and Applications (RAMSITA 2025), pages 63–72. A… find
  23. SSCat Dumbbells Mops With Bucket Spice Racks Collapsible Wardrobe Collapsible Shoe Racks Storage Drawer Units Study Table Kitchen Storage Racks & Holders Detergents Plant Containers Drying Racks Compost & Fertilizers Jars & Containers Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze B… — Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution
  24. Qifan Wang, Li Yang, Bhargav Kanagal, Sumit Sanghai, D Sivakumar, Bin Shu, Zac Yu, and Jon Elsas. 2020. Learning to extract attribute value from product via question answering: A multi-task approach. In Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining,… find
  25. Huimin Xu, Wenting Wang, Xinnian Mao, Xinyu Jiang, and Man Lan. 2019. Scaling up open tagging from tens to thousands: Comprehension empowered attribute value extraction from product title. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 5214– 5223. find
  26. Guineng Zheng, Subhabrata Mukherjee, Xin Luna Dong, and Feifei Li. 2018. Opentag: Open attribute value extraction from product profiles. In Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining, pages 1049–1058. find
  27. Henry Peng Zou, Gavin Heqing Yu, Ziwei Fan, Dan Bu, Han Liu, Peng Dai, Dongmei Jia, and Cornelia Caragea. 2024. Eiven: Efficient implicit attribute value extraction using multimodal llm. arXiv preprint arXiv:2404.08886. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/singhal-2026-grounded-multimodal-context-learning-product/