{"id":"https://openalex.org/W4417299467","doi":"https://doi.org/10.48550/arxiv.2505.07610","title":"Concept-Level Explainability for Auditing & Steering LLM Responses","display_name":"Concept-Level Explainability for Auditing & Steering LLM Responses","publication_year":2025,"publication_date":"2025-05-12","ids":{"openalex":"https://openalex.org/W4417299467","doi":"https://doi.org/10.48550/arxiv.2505.07610"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2505.07610","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.07610","pdf_url":"https://arxiv.org/pdf/2505.07610","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-sa","license_id":"https://openalex.org/licenses/cc-by-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2505.07610","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5034958835","display_name":"Kenza Amara","orcid":"https://orcid.org/0000-0001-7139-5562"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Amara, Kenza","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5014635198","display_name":"Rita Sevastjanova","orcid":"https://orcid.org/0000-0002-2629-9579"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sevastjanova, Rita","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5020415668","display_name":"Mennatallah El\u2010Assady","orcid":"https://orcid.org/0000-0001-8526-2613"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"El-Assady, Mennatallah","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":3,"corresponding_author_ids":["https://openalex.org/A5034958835"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.3646000027656555,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.3646000027656555,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.1639000028371811,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.06849999725818634,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.745199978351593},{"id":"https://openalex.org/keywords/safer","display_name":"SAFER","score":0.7386000156402588},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.6531999707221985},{"id":"https://openalex.org/keywords/semantics","display_name":"Semantics (computer science)","score":0.6306999921798706},{"id":"https://openalex.org/keywords/audit","display_name":"Audit","score":0.5825999975204468},{"id":"https://openalex.org/keywords/attribution","display_name":"Attribution","score":0.4375999867916107}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7633000016212463},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.745199978351593},{"id":"https://openalex.org/C2776654903","wikidata":"https://www.wikidata.org/wiki/Q2601463","display_name":"SAFER","level":2,"score":0.7386000156402588},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.6531999707221985},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.6306999921798706},{"id":"https://openalex.org/C199521495","wikidata":"https://www.wikidata.org/wiki/Q181487","display_name":"Audit","level":2,"score":0.5825999975204468},{"id":"https://openalex.org/C143299363","wikidata":"https://www.wikidata.org/wiki/Q900584","display_name":"Attribution","level":2,"score":0.4375999867916107},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.37290000915527344},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3587000072002411},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.30219998955726624},{"id":"https://openalex.org/C2776291640","wikidata":"https://www.wikidata.org/wiki/Q2912517","display_name":"Value (mathematics)","level":2,"score":0.2919999957084656},{"id":"https://openalex.org/C183322885","wikidata":"https://www.wikidata.org/wiki/Q17007702","display_name":"Context model","level":3,"score":0.2833000123500824},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.2689000070095062},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.2687000036239624},{"id":"https://openalex.org/C90312973","wikidata":"https://www.wikidata.org/wiki/Q7449052","display_name":"Semantic data model","level":2,"score":0.2533000111579895}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2505.07610","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.07610","pdf_url":"https://arxiv.org/pdf/2505.07610","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-sa","license_id":"https://openalex.org/licenses/cc-by-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2505.07610","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2505.07610","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2505.07610","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.07610","pdf_url":"https://arxiv.org/pdf/2505.07610","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-sa","license_id":"https://openalex.org/licenses/cc-by-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4417299467.pdf","grobid_xml":"https://content.openalex.org/works/W4417299467.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"As":[0],"large":[1],"language":[2],"models":[3],"(LLMs)":[4],"become":[5],"widely":[6],"deployed,":[7],"concerns":[8],"about":[9],"their":[10],"safety":[11,218],"and":[12,96,121,138,169,184,195,201,212,219],"alignment":[13],"grow.":[14],"An":[15],"approach":[16],"to":[17,30,113,143,191],"steer":[18],"LLM":[19,75,152,217,230],"behavior,":[20],"such":[21],"as":[22],"mitigating":[23],"biases":[24],"or":[25,147],"defending":[26],"against":[27],"jailbreaks,":[28],"is":[29],"identify":[31],"which":[32],"parts":[33],"of":[34,40,60,72,136,151,225],"a":[35,48,80,210],"prompt":[36,199],"influence":[37],"specific":[38],"aspects":[39],"the":[41,58,64,69,73,87,94,102,145,149,222],"model's":[42],"output.":[43],"Token-level":[44],"attribution":[45,194],"methods":[46,163,203],"offer":[47],"promising":[49],"solution,":[50],"but":[51],"still":[52],"struggle":[53],"in":[54,63,93,166,228],"text":[55],"generation,":[56],"explaining":[57],"presence":[59],"each":[61],"token":[62,119],"output":[65],"separately,":[66],"rather":[67],"than":[68],"underlying":[70],"semantics":[71],"entire":[74],"response.":[76],"We":[77],"introduce":[78],"ConceptX,":[79],"model-agnostic,":[81],"concept-level":[82],"explainability":[83,227],"method":[84],"that":[85],"identifies":[86],"concepts,":[88],"i.e.,":[89],"semantically":[90],"rich":[91],"tokens":[92],"prompt,":[95],"assigns":[97],"them":[98],"importance":[99],"based":[100],"on":[101],"outputs'":[103],"semantic":[104],"similarity.":[105],"Unlike":[106],"current":[107],"token-level":[108,162],"methods,":[109],"ConceptX":[110,129,160,208],"also":[111],"offers":[112,209],"preserve":[114],"context":[115],"integrity":[116],"through":[117],"in-place":[118],"replacements":[120],"supports":[122],"flexible":[123],"explanation":[124],"goals,":[125],"e.g.,":[126],"gender":[127],"bias.":[128],"enables":[130],"both":[131,167],"auditing,":[132],"by":[133,140,177],"uncovering":[134],"sources":[135],"bias,":[137],"steering,":[139],"modifying":[141],"prompts":[142],"shift":[144,176],"sentiment":[146,175],"reduce":[148],"harmfulness":[150],"responses,":[153,207],"without":[154],"requiring":[155],"retraining.":[156],"Across":[157],"three":[158],"LLMs,":[159],"outperforms":[161],"like":[164],"TokenSHAP":[165],"faithfulness":[168],"human":[170],"alignment.":[171],"Steering":[172],"tasks":[173],"boost":[174],"0.252":[178],"versus":[179],"0.131":[180],"for":[181,215],"random":[182],"edits":[183],"lower":[185],"attack":[186],"success":[187],"rates":[188],"from":[189],"0.463":[190],"0.242,":[192],"outperforming":[193],"paraphrasing":[196],"baselines.":[197],"While":[198],"engineering":[200],"self-explaining":[202],"sometimes":[204],"yield":[205],"safer":[206],"transparent":[211],"faithful":[213],"alternative":[214],"improving":[216],"alignment,":[220],"demonstrating":[221],"practical":[223],"value":[224],"attribution-based":[226],"guiding":[229],"behavior.":[231]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-04-11T08:14:18.477133","created_date":"2025-10-10T00:00:00"}