{"id":"https://openalex.org/W7168388724","doi":"https://doi.org/10.1145/3805712.3809891","title":"Reward Shaping for Robust Refusal in Small Language Models for Retrieval-Augmented Question Answering","display_name":"Reward Shaping for Robust Refusal in Small Language Models for Retrieval-Augmented Question Answering","publication_year":2026,"publication_date":"2026-07-15","ids":{"openalex":"https://openalex.org/W7168388724","doi":"https://doi.org/10.1145/3805712.3809891"},"language":null,"primary_location":{"id":"doi:10.1145/3805712.3809891","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3805712.3809891","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 49th International ACM SIGIR Conference on Research and Development in Information Retrieval","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3805712.3809891","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5140910593","display_name":"Thilina C. Rajapakse","orcid":"https://orcid.org/0000-0001-5482-664X"},"institutions":[{"id":"https://openalex.org/I83019370","display_name":"Eindhoven University of Technology","ror":"https://ror.org/02c2kyt77","country_code":"NL","type":"education","lineage":["https://openalex.org/I83019370"]}],"countries":["NL"],"is_corresponding":false,"raw_author_name":"Thilina C. Rajapakse","raw_affiliation_strings":["Eindhoven University of Technology, Eindhoven, Netherlands"],"raw_orcid":"https://orcid.org/0000-0001-5482-664X","affiliations":[{"raw_affiliation_string":"Eindhoven University of Technology, Eindhoven, Netherlands","institution_ids":["https://openalex.org/I83019370"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5140875532","display_name":"Maarten de Rijke","orcid":"https://orcid.org/0000-0002-1086-0202"},"institutions":[{"id":"https://openalex.org/I4210135670","display_name":"Amsterdam University of the Arts","ror":"https://ror.org/04dde1554","country_code":"NL","type":"education","lineage":["https://openalex.org/I4210135670"]},{"id":"https://openalex.org/I887064364","display_name":"University of Amsterdam","ror":"https://ror.org/04dkp9463","country_code":"NL","type":"education","lineage":["https://openalex.org/I887064364"]}],"countries":["NL"],"is_corresponding":false,"raw_author_name":"Maarten de Rijke","raw_affiliation_strings":["University of Amsterdam, Amsterdam, Netherlands"],"raw_orcid":"https://orcid.org/0000-0002-1086-0202","affiliations":[{"raw_affiliation_string":"University of Amsterdam, Amsterdam, Netherlands","institution_ids":["https://openalex.org/I4210135670","https://openalex.org/I887064364"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"4064","last_page":"4068"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":null,"topics":[],"keywords":[{"id":"https://openalex.org/keywords/question-answering","display_name":"Question answering","score":0.6958000063896179},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4041999876499176},{"id":"https://openalex.org/keywords/action","display_name":"Action (physics)","score":0.29679998755455017},{"id":"https://openalex.org/keywords/variety","display_name":"Variety (cybernetics)","score":0.2759000062942505}],"concepts":[{"id":"https://openalex.org/C44291984","wikidata":"https://www.wikidata.org/wiki/Q1074173","display_name":"Question answering","level":2,"score":0.6958000063896179},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4885999858379364},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4041999876499176},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.38850000500679016},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.38100001215934753},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.36730000376701355},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.29679998755455017},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.2872999906539917},{"id":"https://openalex.org/C136197465","wikidata":"https://www.wikidata.org/wiki/Q1729295","display_name":"Variety (cybernetics)","level":2,"score":0.2759000062942505},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.26249998807907104},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.2531999945640564}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3805712.3809891","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3805712.3809891","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 49th International ACM SIGIR Conference on Research and Development in Information Retrieval","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3805712.3809891","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3805712.3809891","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 49th International ACM SIGIR Conference on Research and Development in Information Retrieval","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":8,"referenced_works":["https://openalex.org/W2970482702","https://openalex.org/W3159959439","https://openalex.org/W4312536613","https://openalex.org/W4361001766","https://openalex.org/W4400526199","https://openalex.org/W4404534210","https://openalex.org/W4404782827","https://openalex.org/W4404927183"],"related_works":[],"abstract_inverted_index":{"We":[0,26,126],"focus":[1],"on":[2],"smaller":[3],"open-source":[4],"LMs":[5,86],"(2\u20137B":[6],"parameters),":[7],"which":[8],"are":[9],"attractive":[10],"for":[11,75],"practical":[12],"deployment":[13],"due":[14],"to":[15,37,87,94,96,111,131],"their":[16],"lower":[17],"computational":[18],"cost":[19],"and":[20,77,93,114,123,129],"greater":[21],"accessibility":[22],"than":[23],"frontier-scale":[24],"models.":[25],"show":[27],"that":[28,84],"instruction-tuned":[29,54],"models":[30,55],"generate":[31],"answers":[32],"even":[33],"when":[34,39,98],"explicitly":[35],"prompted":[36],"refuse":[38,95],"the":[40,46,49],"answer":[41,60,97],"is":[42,100],"not":[43],"supported":[44],"by":[45],"documents.":[47],"In":[48],"presence":[50],"of":[51,121],"distractor":[52,112],"documents,":[53],"demonstrate":[56],"inconsistent":[57],"performance,":[58],"with":[59,104,119],"accuracy":[61],"metrics":[62],"deteriorating":[63],"in":[64,109,115],"most":[65],"cases.":[66],"To":[67],"mitigate":[68],"this":[69],"behavior,":[70],"we":[71],"introduce":[72],"Reward":[73],"Shaping":[74],"Refusal":[76],"Reasoning":[78],"(RSRR),":[79],"a":[80],"reinforcement":[81],"learning":[82],"framework":[83],"teaches":[85],"reason":[88],"step-by-step":[89],"over":[90],"multiple":[91],"documents":[92,113],"evidence":[99],"insufficient.":[101],"Models":[102],"trained":[103],"RSRR":[105],"achieve":[106],"substantial":[107],"improvements":[108],"robustness":[110],"correct":[116],"refusal":[117],"accuracy,":[118],"gains":[120],"39.8%":[122],"43.3%,":[124],"respectively.":[125],"release":[127],"code":[128],"data":[130],"reproduce":[132],"all":[133],"results.":[134],"https://github.com/ThilinaRajapakse/rsrr":[135]},"counts_by_year":[],"updated_date":"2026-07-17T05:52:16.776730","created_date":"2026-07-16T00:00:00"}