{"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/another-diversity-promoting-objective","title":"Another Diversity-Promoting Objective Function for Neural Dialogue Generation","arxiv_id":"1811.08100","date":"2018-11-20","proceeding":null,"authors":["Ryo Nakamura","Katsuhito Sudoh","Koichiro Yoshino","Satoshi Nakamura"],"abstract":"Although generation-based dialogue systems have been widely researched, the\nresponse generations by most existing systems have very low diversities. The\nmost likely reason for this problem is Maximum Likelihood Estimation (MLE) with\nSoftmax Cross-Entropy (SCE) loss. MLE trains models to generate the most\nfrequent responses from enormous generation candidates, although in actual\ndialogues there are various responses based on the context. In this paper, we\npropose a new objective function called Inverse Token Frequency (ITF) loss,\nwhich individually scales smaller loss for frequent token classes and larger\nloss for rare token classes. This function encourages the model to generate\nrare tokens rather than frequent tokens. It does not complicate the model and\nits training is stable because we only replace the objective function. On the\nOpenSubtitles dialogue dataset, our loss model establishes a state-of-the-art\nDIST-1 of 7.56, which is the unigram diversity score, while maintaining a good\nBLEU-1 score. On a Japanese Twitter replies dataset, our loss model achieves a\nDIST-1 score comparable to the ground truth.","url_abs":"http://arxiv.org/abs/1811.08100v2","url_pdf":"http://arxiv.org/pdf/1811.08100v2.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":"another-diversity-promoting-objective","repo_url":"https://github.com/reppy4620/Dialog","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"dialogue-generation","task_name":"Dialogue Generation"},{"task_slug":"diversity","task_name":"Diversity"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}