{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2025,8,23]],"date-time":"2025-08-23T05:26:16Z","timestamp":1755926776152,"version":"3.40.3"},"publisher-location":"Cham","reference-count":22,"publisher":"Springer Nature Switzerland","isbn-type":[{"type":"print","value":"9783031333798"},{"type":"electronic","value":"9783031333804"}],"license":[{"start":{"date-parts":[[2023,1,1]],"date-time":"2023-01-01T00:00:00Z","timestamp":1672531200000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"},{"start":{"date-parts":[[2023,5,27]],"date-time":"2023-05-27T00:00:00Z","timestamp":1685145600000},"content-version":"vor","delay-in-days":146,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"}],"content-domain":{"domain":["link.springer.com"],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2023]]},"abstract":"<jats:title>Abstract<\/jats:title><jats:p>Dense embedding-based semantic matching is widely used in e-commerce product search to address the shortcomings of lexical matching such as sensitivity to spelling variants. The recent advances in BERT-like language model encoders, have however, not found their way to realtime search due to the strict inference latency requirement imposed on e-commerce websites. While bi-encoder BERT architectures enable fast approximate nearest neighbor search, training them effectively on query-product data remains a challenge due to training instabilities and the persistent generalization gap with cross-encoders. In this work, we propose a four-stage training procedure to leverage large BERT-like models for product search while preserving low inference latency. We introduce query-product interaction pre-finetuning to effectively pretrain BERT bi-encoders for matching and improve generalization. Through offline experiments on an e-commerce product dataset, we show that a distilled small BERT-based model (75M params) trained using our approach improves the search relevance metric by up to 23% over a baseline DSSM-based model with similar inference latency. The small model only suffers a 3% drop in relevance metric compared to the 20x larger teacher. We also show using online A\/B tests at scale, that our approach improves over the production model in exact and substitute products retrieved.<\/jats:p>","DOI":"10.1007\/978-3-031-33380-4_6","type":"book-chapter","created":{"date-parts":[[2023,5,26]],"date-time":"2023-05-26T08:05:58Z","timestamp":1685088358000},"page":"73-85","update-policy":"https:\/\/doi.org\/10.1007\/springer_crossmark_policy","source":"Crossref","is-referenced-by-count":6,"title":["Web-Scale Semantic Product Search with\u00a0Large Language Models"],"prefix":"10.1007","author":[{"given":"Aashiq","family":"Muhamed","sequence":"first","affiliation":[]},{"given":"Sriram","family":"Srinivasan","sequence":"additional","affiliation":[]},{"given":"Choon-Hui","family":"Teo","sequence":"additional","affiliation":[]},{"given":"Qingjun","family":"Cui","sequence":"additional","affiliation":[]},{"given":"Belinda","family":"Zeng","sequence":"additional","affiliation":[]},{"given":"Trishul","family":"Chilimbi","sequence":"additional","affiliation":[]},{"given":"S. V. N.","family":"Vishwanathan","sequence":"additional","affiliation":[]}],"member":"297","published-online":{"date-parts":[[2023,5,27]]},"reference":[{"key":"6_CR1","doi-asserted-by":"crossref","unstructured":"Sch\u00fctze, H., Manning, C.D., Raghavan, P.: Introduction to Information Retrieval. Cambridge University Press, Cambridge (2008)","DOI":"10.1017\/CBO9780511809071"},{"key":"6_CR2","doi-asserted-by":"crossref","unstructured":"Huang, P.-S., He, X., Gao, J., Deng, L., Acero, A., Heck, L.P.: Learning deep structured semantic models for web search using clickthrough data. In: CIKM (2013)","DOI":"10.1145\/2505515.2505665"},{"key":"6_CR3","doi-asserted-by":"crossref","unstructured":"Nigam, P., et al.: Semantic product search. In: SIGKDD (2019)","DOI":"10.1145\/3292500.3330759"},{"key":"6_CR4","doi-asserted-by":"crossref","unstructured":"Malkov Y.A., Yashunin, D.A.: Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs. IEEE Trans. Pattern Anal. Mach. Intell. 42(4), 824\u2013836 (2018)","DOI":"10.1109\/TPAMI.2018.2889473"},{"key":"6_CR5","unstructured":"Devlin, J., Chang, M.W., Lee, K., Toutanova, K.: BERT: pre-training of deep bidirectional transformers for language understanding. In: NACCL (2019)"},{"key":"6_CR6","doi-asserted-by":"crossref","unstructured":"Reimers, N., Gurevych, I.: Sentence-BERT: sentence embeddings using Siamese BERT-networks. In: EMNLP-IJCNLP (2019)","DOI":"10.18653\/v1\/D19-1410"},{"key":"6_CR7","doi-asserted-by":"crossref","unstructured":"Khattab, O., Zaharia, M.: Colbert: efficient and effective passage search via contextualized late interaction over BERT. In: SIGIR (2020)","DOI":"10.1145\/3397271.3401075"},{"key":"6_CR8","doi-asserted-by":"crossref","unstructured":"Lu, W., Jiao, J., Zhang, R.: TwinBERT: distilling knowledge to twin-structured BERT models for efficient retrieval. ArXiv, vol. abs\/2002.06275 (2020)","DOI":"10.1145\/3340531.3412747"},{"key":"6_CR9","doi-asserted-by":"crossref","unstructured":"Huang, J.-T., et al.: Embedding-based retrieval in Facebook search. In: SIGKDD (2020)","DOI":"10.1145\/3394486.3403305"},{"key":"6_CR10","doi-asserted-by":"crossref","unstructured":"Li, S., et al.: Embedding-based product retrieval in Taobao search. In: SIGKDD (2021)","DOI":"10.1145\/3447548.3467101"},{"key":"6_CR11","doi-asserted-by":"crossref","unstructured":"Karpukhin, V., et al.: Dense passage retrieval for open-domain question answering. In: EMNLP (2020)","DOI":"10.18653\/v1\/2020.emnlp-main.550"},{"key":"6_CR12","unstructured":"Xiong, L., et al.: Approximate nearest neighbor negative contrastive learning for dense text retrieval. In: ICLR (2021)"},{"key":"6_CR13","doi-asserted-by":"crossref","unstructured":"Liu, Y., et al.: Pre-trained language model for web-scale retrieval in baidu search. In: SIGKDD (2021)","DOI":"10.1145\/3447548.3467149"},{"key":"6_CR14","doi-asserted-by":"crossref","unstructured":"Sennrich, R., Haddow, B., Birch, A.: Neural machine translation of rare words with subword units. In :Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Berlin, Germany, pp. 1715\u20131725, Association for Computational Linguistics, Aug. 2016","DOI":"10.18653\/v1\/P16-1162"},{"key":"6_CR15","unstructured":"Liu, Y., et al.: RoBERTa: a robustly optimized BERT pretraining approach, ArXiv, vol. abs\/1907.11692 (2019)"},{"key":"6_CR16","unstructured":"Lan, Z., Chen, M., Goodman, S., Gimpel, K., Sharma, P., Soricut, R.: AlBERT: a lite BERT for self-supervised learning of language representations. In: ICLR (2020)"},{"key":"6_CR17","unstructured":"Ji, S., Vishwanathan, S.V.N., Satish, N., Anderson, M.J., Dubey, P.: BlackOut: speeding up recurrent neural network language models with very large vocabularies. In: ICLR (2016)"},{"key":"6_CR18","unstructured":"Hofst\u00e4tter, S., Althammer, S., Schr\u00f6der, M., Sertkan, M., Hanbury, A.: Improving efficient neural ranking models with cross-architecture knowledge distillation. ArXiv, vol. abs\/2010.02666 (2020)"},{"key":"6_CR19","unstructured":"Chen, T., Kornblith, S., Norouzi, M., Hinton, G.: A simple framework for contrastive learning of visual representations. In: ICML (2020)"},{"key":"6_CR20","doi-asserted-by":"crossref","unstructured":"Kudo, T., Richardson, J.: SentencePiece: a simple and language independent subword tokenizer and detokenizer for neural text processing. In: EMNLP (2018)","DOI":"10.18653\/v1\/D18-2012"},{"key":"6_CR21","doi-asserted-by":"crossref","unstructured":"Conneau, A.: Unsupervised cross-lingual representation learning at scale. In: ACL (2020)","DOI":"10.18653\/v1\/2020.acl-main.747"},{"key":"6_CR22","unstructured":"Zheng, S., Lin, H., Zha, S., Li, M.: Accelerated large batch optimization of BERT pretraining in 54 minutes. ArXiv, vol. abs\/2006.13484 (2020)"}],"container-title":["Lecture Notes in Computer Science","Advances in Knowledge Discovery and Data Mining"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/link.springer.com\/content\/pdf\/10.1007\/978-3-031-33380-4_6","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2024,3,13]],"date-time":"2024-03-13T19:58:36Z","timestamp":1710359916000},"score":1,"resource":{"primary":{"URL":"https:\/\/link.springer.com\/10.1007\/978-3-031-33380-4_6"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2023]]},"ISBN":["9783031333798","9783031333804"],"references-count":22,"URL":"https:\/\/doi.org\/10.1007\/978-3-031-33380-4_6","relation":{},"ISSN":["0302-9743","1611-3349"],"issn-type":[{"type":"print","value":"0302-9743"},{"type":"electronic","value":"1611-3349"}],"subject":[],"published":{"date-parts":[[2023]]},"assertion":[{"value":"27 May 2023","order":1,"name":"first_online","label":"First Online","group":{"name":"ChapterHistory","label":"Chapter History"}},{"value":"PAKDD","order":1,"name":"conference_acronym","label":"Conference Acronym","group":{"name":"ConferenceInfo","label":"Conference Information"}},{"value":"Pacific-Asia Conference on Knowledge Discovery and Data Mining","order":2,"name":"conference_name","label":"Conference Name","group":{"name":"ConferenceInfo","label":"Conference Information"}},{"value":"Osaka","order":3,"name":"conference_city","label":"Conference City","group":{"name":"ConferenceInfo","label":"Conference Information"}},{"value":"Japan","order":4,"name":"conference_country","label":"Conference Country","group":{"name":"ConferenceInfo","label":"Conference Information"}},{"value":"2023","order":5,"name":"conference_year","label":"Conference Year","group":{"name":"ConferenceInfo","label":"Conference Information"}},{"value":"25 May 2023","order":7,"name":"conference_start_date","label":"Conference Start Date","group":{"name":"ConferenceInfo","label":"Conference Information"}},{"value":"28 May 2023","order":8,"name":"conference_end_date","label":"Conference End Date","group":{"name":"ConferenceInfo","label":"Conference Information"}},{"value":"27","order":9,"name":"conference_number","label":"Conference Number","group":{"name":"ConferenceInfo","label":"Conference Information"}},{"value":"pakdd2023","order":10,"name":"conference_id","label":"Conference ID","group":{"name":"ConferenceInfo","label":"Conference Information"}},{"value":"https:\/\/pakdd2023.org\/","order":11,"name":"conference_url","label":"Conference URL","group":{"name":"ConferenceInfo","label":"Conference Information"}},{"value":"Double-blind","order":1,"name":"type","label":"Type","group":{"name":"ConfEventPeerReviewInformation","label":"Peer Review Information (provided by the conference organizers)"}},{"value":"Microsoft CMT","order":2,"name":"conference_management_system","label":"Conference Management System","group":{"name":"ConfEventPeerReviewInformation","label":"Peer Review Information (provided by the conference organizers)"}},{"value":"813","order":3,"name":"number_of_submissions_sent_for_review","label":"Number of Submissions Sent for Review","group":{"name":"ConfEventPeerReviewInformation","label":"Peer Review Information (provided by the conference organizers)"}},{"value":"143","order":4,"name":"number_of_full_papers_accepted","label":"Number of Full Papers Accepted","group":{"name":"ConfEventPeerReviewInformation","label":"Peer Review Information (provided by the conference organizers)"}},{"value":"0","order":5,"name":"number_of_short_papers_accepted","label":"Number of Short Papers Accepted","group":{"name":"ConfEventPeerReviewInformation","label":"Peer Review Information (provided by the conference organizers)"}},{"value":"18% - The value is computed by the equation \"Number of Full Papers Accepted \/ Number of Submissions Sent for Review * 100\" and then rounded to a whole number.","order":6,"name":"acceptance_rate_of_full_papers","label":"Acceptance Rate of Full Papers","group":{"name":"ConfEventPeerReviewInformation","label":"Peer Review Information (provided by the conference organizers)"}},{"value":"3.5","order":7,"name":"average_number_of_reviews_per_paper","label":"Average Number of Reviews per Paper","group":{"name":"ConfEventPeerReviewInformation","label":"Peer Review Information (provided by the conference organizers)"}},{"value":"10","order":8,"name":"average_number_of_papers_per_reviewer","label":"Average Number of Papers per Reviewer","group":{"name":"ConfEventPeerReviewInformation","label":"Peer Review Information (provided by the conference organizers)"}},{"value":"Yes","order":9,"name":"external_reviewers_involved","label":"External Reviewers Involved","group":{"name":"ConfEventPeerReviewInformation","label":"Peer Review Information (provided by the conference organizers)"}}]}}