{"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/disan-directional-self-attention-network-for","title":"DiSAN: Directional Self-Attention Network for RNN/CNN-Free Language Understanding","arxiv_id":"1709.04696","date":"2017-09-14","proceeding":null,"authors":["Tao Shen","Tianyi Zhou","Guodong Long","Jing Jiang","Shirui Pan","Chengqi Zhang"],"abstract":"Recurrent neural nets (RNN) and convolutional neural nets (CNN) are widely\nused on NLP tasks to capture the long-term and local dependencies,\nrespectively. Attention mechanisms have recently attracted enormous interest\ndue to their highly parallelizable computation, significantly less training\ntime, and flexibility in modeling dependencies. We propose a novel attention\nmechanism in which the attention between elements from input sequence(s) is\ndirectional and multi-dimensional (i.e., feature-wise). A light-weight neural\nnet, \"Directional Self-Attention Network (DiSAN)\", is then proposed to learn\nsentence embedding, based solely on the proposed attention without any RNN/CNN\nstructure. DiSAN is only composed of a directional self-attention with temporal\norder encoded, followed by a multi-dimensional attention that compresses the\nsequence into a vector representation. Despite its simple form, DiSAN\noutperforms complicated RNN models on both prediction quality and time\nefficiency. It achieves the best test accuracy among all sentence encoding\nmethods and improves the most recent best result by 1.02% on the Stanford\nNatural Language Inference (SNLI) dataset, and shows state-of-the-art test\naccuracy on the Stanford Sentiment Treebank (SST), Multi-Genre natural language\ninference (MultiNLI), Sentences Involving Compositional Knowledge (SICK),\nCustomer Review, MPQA, TREC question-type classification and Subjectivity\n(SUBJ) datasets.","url_abs":"http://arxiv.org/abs/1709.04696v3","url_pdf":"http://arxiv.org/pdf/1709.04696v3.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":"disan-directional-self-attention-network-for","repo_url":"https://github.com/taoshen58/DiSAN","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"disan-directional-self-attention-network-for","repo_url":"https://github.com/2023-MindSpore-4/Code8/tree/main/cnn_direction_model","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null},{"paper_slug":"disan-directional-self-attention-network-for","repo_url":"https://github.com/MindSpore-paper-code-3/code6/tree/main/cnn_direction_model","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null}],"tasks":[{"task_slug":"natural-language-inference","task_name":"Natural Language Inference"},{"task_slug":"sentence","task_name":"Sentence"},{"task_slug":"sentence-embedding","task_name":"Sentence Embedding"},{"task_slug":"sentence-embedding-1","task_name":"Sentence-Embedding"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/natural-language-inference-on-snli","task":"Natural Language Inference","dataset":"SNLI","model":"300D Directional self-attention network encoders","rank_in_archive_order":70,"of":98,"metrics":{"% Test Accuracy":"85.6","% Train Accuracy":"91.1","Parameters":"2.4m"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1709.04696","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}