{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/dont-settle-for-average-go-for-the-max-fuzzy-1","title":"Don't Settle for Average, Go for the Max: Fuzzy Sets and Max-Pooled Word Vectors","arxiv_id":"1904.13264","date":"2019-04-30","proceeding":"ICLR 2019 5","authors":["Vitalii Zhelezniak","Aleksandar Savkov","April Shen","Francesco Moramarco","Jack Flann","Nils Y. Hammerla"],"abstract":"Recent literature suggests that averaged word vectors followed by simple\npost-processing outperform many deep learning methods on semantic textual\nsimilarity tasks. Furthermore, when averaged word vectors are trained\nsupervised on large corpora of paraphrases, they achieve state-of-the-art\nresults on standard STS benchmarks. Inspired by these insights, we push the\nlimits of word embeddings even further. We propose a novel fuzzy bag-of-words\n(FBoW) representation for text that contains all the words in the vocabulary\nsimultaneously but with different degrees of membership, which are derived from\nsimilarities between word vectors. We show that max-pooled word vectors are\nonly a special case of fuzzy BoW and should be compared via fuzzy Jaccard index\nrather than cosine similarity. Finally, we propose DynaMax, a completely\nunsupervised and non-parametric similarity measure that dynamically extracts\nand max-pools good features depending on the sentence pair. This method is both\nefficient and easy to implement, yet outperforms current baselines on STS tasks\nby a large margin and is even competitive with supervised word vectors trained\nto directly optimise cosine similarity.","url_abs":"http://arxiv.org/abs/1904.13264v1","url_pdf":"http://arxiv.org/pdf/1904.13264v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"dont-settle-for-average-go-for-the-max-fuzzy-1","repo_url":"https://github.com/Babylonpartners/fuzzymax","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null},{"paper_slug":"dont-settle-for-average-go-for-the-max-fuzzy-1","repo_url":"https://github.com/vackosar/fasttext-vector-norms-and-oov-words","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}}],"tasks":[{"task_slug":"sts","task_name":"STS"},{"task_slug":"semantic-textual-similarity","task_name":"Semantic Textual Similarity"},{"task_slug":"sentence","task_name":"Sentence"},{"task_slug":"word-embeddings","task_name":"Word Embeddings"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1904.13264","atlas_url":"https://app.syntology.ai/?focus=1904.13264","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}