{"id":482721,"date":"2018-04-25T12:08:26","date_gmt":"2018-04-25T19:08:26","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-research-item&#038;p=482721"},"modified":"2018-10-16T22:24:09","modified_gmt":"2018-10-17T05:24:09","slug":"subgoal-discovery-hierarchical-dialogue-policy-learning","status":"publish","type":"msr-research-item","link":"https:\/\/www.microsoft.com\/en-us\/research\/publication\/subgoal-discovery-hierarchical-dialogue-policy-learning\/","title":{"rendered":"Subgoal Discovery for Hierarchical Dialogue Policy Learning"},"content":{"rendered":"<p><span style=\"color: #000000;text-indent: 0px;letter-spacing: normal;font-family: 'Lucida Grande', helvetica, arial, verdana, sans-serif;font-size: 14.4px;font-style: normal;font-weight: 400;float: none;background-color: #ffffff\">Developing conversational agents to engage in complex dialogues is challenging partly because the dialogue policy needs to explore a large state-action space. In this paper, we propose a divide-and-conquer approach that discovers and exploits the hidden structure of the task to enable efficient policy learning. First, given a set of successful dialogue sessions, we present a Subgoal Discovery Network (SDN) to divide a complex goal-oriented task into a set of simpler subgoals in an unsupervised fashion. We then use these subgoals to learn a hierarchical policy which consists of 1) a top-level policy that selects among subgoals, and 2) a low-level policy that selects primitive actions to accomplish the subgoal. We exemplify our method by building a dialogue agent for the composite task of travel planning. Experiments with simulated and real users show that an agent trained with automatically discovered subgoals performs competitively against an agent with human-defined subgoals, and significantly outperforms an agent without subgoals. Moreover, we show that learned subgoals are human comprehensible.<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Developing conversational agents to engage in complex dialogues is challenging partly because the dialogue policy needs to explore a large state-action space. In this paper, we propose a divide-and-conquer approach that discovers and exploits the hidden structure of the task to enable efficient policy learning. First, given a set of successful dialogue sessions, we present [&hellip;]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","footnotes":""},"msr-content-type":[3],"msr-research-highlight":[],"research-area":[13556],"msr-publication-type":[193716],"msr-product-type":[],"msr-focus-area":[],"msr-platform":[],"msr-download-source":[],"msr-locale":[268875],"msr-post-option":[],"msr-field-of-study":[],"msr-conference":[],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-482721","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-artificial-intelligence","msr-locale-en_us"],"msr_publishername":"arXiv","msr_edition":"EMNLP 2018","msr_affiliation":"","msr_published_date":"2018-10-31","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"","msr_chapter":"","msr_isbn":"","msr_journal":"","msr_volume":"","msr_number":"MSR-TR-2018-12","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"","msr_how_published":"","msr_notes":"","msr_highlight_text":"","msr_release_tracker_id":"","msr_original_fields_of_study":"","msr_download_urls":"","msr_external_url":"","msr_secondary_video_url":"","msr_longbiography":"","msr_microsoftintellectualproperty":1,"msr_main_download":"","msr_publicationurl":"https:\/\/arxiv.org\/abs\/1804.07855","msr_doi":"","msr_publication_uploader":[{"type":"url","title":"https:\/\/arxiv.org\/abs\/1804.07855","viewUrl":false,"id":false,"label_id":0}],"msr_related_uploader":"","msr_attachments":[{"id":0,"url":"https:\/\/arxiv.org\/abs\/1804.07855"}],"msr-author-ordering":[{"type":"text","value":"Da Tang","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Xiujun Li","user_id":36287,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Xiujun Li"},{"type":"user_nicename","value":"Jianfeng Gao","user_id":32246,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Jianfeng Gao"},{"type":"text","value":"Chong Wang","user_id":0,"rest_url":false},{"type":"text","value":"Lihong Li","user_id":0,"rest_url":false},{"type":"text","value":"Tony Jebara","user_id":0,"rest_url":false}],"msr_impact_theme":[],"msr_research_lab":[199565],"msr_event":[],"msr_group":[144931,395930],"msr_project":[393245,377990,171313],"publication":[],"video":[],"download":[],"msr_publication_type":"inproceedings","related_content":{"projects":[{"ID":393245,"post_title":"Conversational Intelligence","post_name":"conversational-intelligence","post_type":"msr-project","post_date":"2017-07-05 10:01:45","post_modified":"2017-11-15 13:39:25","post_status":"publish","permalink":"https:\/\/www.microsoft.com\/en-us\/research\/project\/conversational-intelligence\/","post_excerpt":"Intelligent agents that can handle human language play a growing role in personalized, ubiquitous computing and the everyday use of devices. Agents need to be able to communicate and collaborate with humans in ways that are seamless and natural, and to be able to learn new behaviors, concepts, and relationships as first-class operations. In other words, our devices need to be able to converse with us. In this project, Microsoft Research AI teams are interested&hellip;","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/393245"}]}},{"ID":377990,"post_title":"Deep Reinforcement Learning for Goal-Oriented Dialogues","post_name":"deep-reinforcement-learning-goal-oriented-dialogue","post_type":"msr-project","post_date":"2017-04-18 11:51:36","post_modified":"2019-08-19 10:03:33","post_status":"publish","permalink":"https:\/\/www.microsoft.com\/en-us\/research\/project\/deep-reinforcement-learning-goal-oriented-dialogue\/","post_excerpt":"Microsoft Dialogue Challenge: Building End-to-End Task-Completion Dialogue Systems, at SLT 2018. [Proposal] All the data, source code and schedule information will be updated here. This project aims to develop intelligent dialogue agents to help users effectively accomplish tasks via natural language conversation. A typical goal-oriented dialogue system contains three major components: natural language understanding (NLU), natural language generation (NLG), and dialogue management (DM) that consists of state tracking and policy learning. Our research focus is&hellip;","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/377990"}]}},{"ID":171313,"post_title":"Dialog and Conversational Systems Research","post_name":"dialog-and-conversational-systems-research","post_type":"msr-project","post_date":"2014-03-14 09:46:35","post_modified":"2017-07-11 15:34:26","post_status":"publish","permalink":"https:\/\/www.microsoft.com\/en-us\/research\/project\/dialog-and-conversational-systems-research\/","post_excerpt":"Conversational systems interact with people through language to assist, enable, or entertain. Research at Microsoft spans dialogs that use language exclusively, or in conjunctions with additional modalities like gesture; where language is spoken or in text; and in a variety of settings, such as conversational systems in apps or devices, and situated interactions in the real world. Projects Spoken Language Understanding","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/171313"}]}}]},"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/482721","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":1,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/482721\/revisions"}],"predecessor-version":[{"id":482724,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/482721\/revisions\/482724"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=482721"}],"wp:term":[{"taxonomy":"msr-content-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-content-type?post=482721"},{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=482721"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=482721"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=482721"},{"taxonomy":"msr-product-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-product-type?post=482721"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=482721"},{"taxonomy":"msr-platform","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-platform?post=482721"},{"taxonomy":"msr-download-source","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-download-source?post=482721"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=482721"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=482721"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=482721"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=482721"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=482721"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=482721"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=482721"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}