{"id":"e1a13c55755af77b9e83386972cbd435bed56fe538066cb4e00ff6cecc30852e","title":"arXiv cs.AI: Researchers turn recommendation edits into explanations","argument":"The paper’s authors are Vojtech Cahlik, Amir Reza Mohammadi, Rodrigo Alves, Pavel Kordik and Eva Zangerle. They argue that counterfactual explanation (CFX) methods can show which changes to a user’s interaction history would alter a recommendation, but often produce long edit lists that are hard to interpret. The authors investigate using large language models to summarize those lists as concise explanations of higher-level preference patterns. They optimize the mediators for correctness, informativeness and linguistic quality through preference optimization. On the MovieLens-1M dataset, they report high overall linguistic quality across multiple counterfactual generators and language models. The strongest semantic results came with high-success CFX methods such as ACCENT and SPINRec. Direct Preference Optimization improved semantic metrics in almost all evaluated settings. The work distinguishes generating valid CFXs from communicating model behavior in a user-understandable form. The authors evaluate explanations with human-calibrated metrics derived from separate human-labeled datasets.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports evaluation on the MovieLens-1M dataset; it does not establish whether the results generalize to other recommendation settings. The authors describe evaluation with human-calibrated metrics derived from separate human-labeled datasets; the abstract does not report direct evaluation of explanations with users.","speaker":{"name":"Cahlik, Vojtech, Mohammadi, Amir Reza, Alves, Rodrigo, Kordik, Pavel, Zangerle, Eva","roleAtSource":"Article author"},"source":{"title":"From Counterfactual Edits to Human Narratives: LLM-Driven Interpretations for Recommender System Explanations","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.13914","publishedAt":"2026-11-10","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors argue that recommender-system counterfactual explanations can require many edits and become difficult for users to interpret.","excerpt":"In recommendation settings, however, CFXs often require many edits, producing long, low-level lists that are hard for users to interpret.","url":"https://arxiv.org/abs/2611.13914","relationship":"Source-stated"},{"text":"The authors investigate using large language models to summarize raw counterfactual edit lists as concise explanations of higher-level preference patterns.","excerpt":"we investigate large language models as mediators that summarize raw CFX interaction edit lists into concise natural language explanations describing higher-level preference patterns.","url":"https://arxiv.org/abs/2611.13914","relationship":"Source-stated"},{"text":"On MovieLens-1M, the authors report high overall linguistic quality; the strongest semantic results came with ACCENT and SPINRec, and Direct Preference Optimization improved semantic metrics in almost all evaluated settings.","excerpt":"On the MovieLens-1M dataset, across multiple counterfactual generators and language models, the resulting summaries achieve high overall linguistic quality, while the strongest semantic results are obtained with high-success CFX methods such as ACCENT and SPINRec. Direct Preference Optimization","url":"https://arxiv.org/abs/2611.13914","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.13914","checkedAt":"2026-11-11T20:48:23.785Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Do the reported explanation-quality results hold with users and recommendation settings beyond the MovieLens-1M evaluation?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: Researchers turn recommendation edits into explanations","summary":"The paper’s authors are Vojtech Cahlik, Amir Reza Mohammadi, Rodrigo Alves, Pavel Kordik and Eva Zangerle. They argue that counterfactual explanation (CFX) methods can show which changes to a user’s interaction history would alter a recommendation, but often produce long edit lists that are hard to interpret. The authors investigate using large language models to summarize those lists as concise explanations of higher-level preference patterns. They optimize the mediators for correctness, informativeness and linguistic quality through preference optimization. On the MovieLens-1M dataset, they report high overall linguistic quality across multiple counterfactual generators and language models. The strongest semantic results came with high-success CFX methods such as ACCENT and SPINRec. Direct Preference Optimization improved semantic metrics in almost all evaluated settings. The work distinguishes generating valid CFXs from communicating model behavior in a user-understandable form. The authors evaluate explanations with human-calibrated metrics derived from separate human-labeled datasets.","question":"Do the reported explanation-quality results hold with users and recommendation settings beyond the MovieLens-1M evaluation?","claims":[{"text":"The authors argue that recommender-system counterfactual explanations can require many edits and become difficult for users to interpret.","excerpt":"In recommendation settings, however, CFXs often require many edits, producing long, low-level lists that are hard for users to interpret.","support":"source-stated"},{"text":"The authors investigate using large language models to summarize raw counterfactual edit lists as concise explanations of higher-level preference patterns.","excerpt":"we investigate large language models as mediators that summarize raw CFX interaction edit lists into concise natural language explanations describing higher-level preference patterns.","support":"source-stated"},{"text":"On MovieLens-1M, the authors report high overall linguistic quality; the strongest semantic results came with ACCENT and SPINRec, and Direct Preference Optimization improved semantic metrics in almost all evaluated settings.","excerpt":"On the MovieLens-1M dataset, across multiple counterfactual generators and language models, the resulting summaries achieve high overall linguistic quality, while the strongest semantic results are obtained with high-success CFX methods such as ACCENT and SPINRec. Direct Preference Optimization","support":"source-stated"}],"limitations":["The abstract reports evaluation on the MovieLens-1M dataset; it does not establish whether the results generalize to other recommendation settings.","The authors describe evaluation with human-calibrated metrics derived from separate human-labeled datasets; the abstract does not report direct evaluation of explanations with users."],"mediaId":"73f9b924ecdff16e747c57a37d6f668036f086fb5fb7db921fcd8adcf27d43a6","correctionReason":null,"id":"e1a13c55755af77b9e83386972cbd435bed56fe538066cb4e00ff6cecc30852e","publicationId":"f02ec2b0-af6b-4ec2-8c30-87f594b96c11","version":1,"sourceUrl":"https://arxiv.org/abs/2611.13914","sourceTitle":"From Counterfactual Edits to Human Narratives: LLM-Driven Interpretations for Recommender System Explanations","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-10","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-11T05:04:41.914Z","createdAt":"2026-11-11T20:16:34.046Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"73f9b924ecdff16e747c57a37d6f668036f086fb5fb7db921fcd8adcf27d43a6","kind":"article","url":"https://arxiv.org/abs/2611.13914","sourceUrl":"https://arxiv.org/abs/2611.13914","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-11T20:04:27.470Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.13914","checkedAt":"2026-11-11T20:48:23.785Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"73f9b924ecdff16e747c57a37d6f668036f086fb5fb7db921fcd8adcf27d43a6","confidence":0.73,"provider":"typesafe","model":"jev-1.13.0","inputHash":"52a8b2159f17f3ab61974a6fe8e407ae570b3834351a7bca21467e5b8a0a66ea","decidedAt":"2026-11-11T20:04:29.555Z"},"provenance":{"textHash":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","observationId":"28a2db32e3186bd89bff2dd10f8c4958209412dd61cbf697c7e07e8d94c44896","retrievedAt":"2026-11-11T20:04:27.044Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"6b9b03bc-d528-438e-8691-eb17d7615407"},"reviewer":{"model":"gpt-6-luna","runId":"38661ba1-dd64-4d04-88a9-a45fc9b2d503"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-11T20:16:34.046Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Cahlik, Vojtech","url":null,"kind":"unknown"},{"name":"Mohammadi, Amir Reza","url":null,"kind":"unknown"},{"name":"Alves, Rodrigo","url":null,"kind":"unknown"},{"name":"Kordik, Pavel","url":null,"kind":"unknown"},{"name":"Zangerle, Eva","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Cahlik, Vojtech","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Cahlik, Vojtech","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.13914","locator":"head/meta[name=author|citation_author]"},"assertionId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:48:23.785Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e","sourceUrl":"https://arxiv.org/abs/2611.13914","contentSha256":"85f7bfefa6cd0cdea63ecacb11f3399e539f632976f757798b9fc49cc68ba9e0","textHash":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","observedAt":"2026-11-11T20:48:23.785Z"}},{"actor":{"name":"Mohammadi, Amir Reza","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Mohammadi, Amir Reza","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.13914","locator":"head/meta[name=author|citation_author]"},"assertionId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:48:23.785Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e","sourceUrl":"https://arxiv.org/abs/2611.13914","contentSha256":"85f7bfefa6cd0cdea63ecacb11f3399e539f632976f757798b9fc49cc68ba9e0","textHash":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","observedAt":"2026-11-11T20:48:23.785Z"}},{"actor":{"name":"Alves, Rodrigo","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Alves, Rodrigo","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.13914","locator":"head/meta[name=author|citation_author]"},"assertionId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:48:23.785Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e","sourceUrl":"https://arxiv.org/abs/2611.13914","contentSha256":"85f7bfefa6cd0cdea63ecacb11f3399e539f632976f757798b9fc49cc68ba9e0","textHash":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","observedAt":"2026-11-11T20:48:23.785Z"}},{"actor":{"name":"Kordik, Pavel","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Kordik, Pavel","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.13914","locator":"head/meta[name=author|citation_author]"},"assertionId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:48:23.785Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e","sourceUrl":"https://arxiv.org/abs/2611.13914","contentSha256":"85f7bfefa6cd0cdea63ecacb11f3399e539f632976f757798b9fc49cc68ba9e0","textHash":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","observedAt":"2026-11-11T20:48:23.785Z"}},{"actor":{"name":"Zangerle, Eva","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zangerle, Eva","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.13914","locator":"head/meta[name=author|citation_author]"},"assertionId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:48:23.785Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e","sourceUrl":"https://arxiv.org/abs/2611.13914","contentSha256":"85f7bfefa6cd0cdea63ecacb11f3399e539f632976f757798b9fc49cc68ba9e0","textHash":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","observedAt":"2026-11-11T20:48:23.785Z"}}],"participationCoverage":{"total":5,"returned":5,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.13914","snapshotId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e","contentSha256":"85f7bfefa6cd0cdea63ecacb11f3399e539f632976f757798b9fc49cc68ba9e0","textHash":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","observedAt":"2026-11-11T20:48:23.785Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/e1a13c55755af77b9e83386972cbd435bed56fe538066cb4e00ff6cecc30852e/source"}},"canonicalUrl":"https://escapemeridian.com/readings/arxiv-cs-ai-researchers-turn-recommendation-edits-into-e1a13c55","formats":{"html":"/readings/arxiv-cs-ai-researchers-turn-recommendation-edits-into-e1a13c55","markdown":"/readings/arxiv-cs-ai-researchers-turn-recommendation-edits-into-e1a13c55.md","json":"/readings/arxiv-cs-ai-researchers-turn-recommendation-edits-into-e1a13c55.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[{"id":"term-counterfactual","name":"Counterfactual","explanation":"An argument or scenario that considers what would have resulted if events had gone differently from how they actually did, such as if a policy had never been adopted; as an adjective, relating to this kind of thinking about what did not happen but could have.","referenceUrl":"https://dictionary.cambridge.org/dictionary/english/counterfactual"},{"id":"term-llm","name":"Large language model","explanation":"A language model with many learned numerical settings, used to process and generate language.","referenceUrl":"https://developers.google.com/machine-learning/glossary#large-language-model"}],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"e1a13c55755af77b9e83386972cbd435bed56fe538066cb4e00ff6cecc30852e","revision":"f02ec2b0-af6b-4ec2-8c30-87f594b96c11","revisionKind":"publication-id","title":"arXiv cs.AI: Researchers turn recommendation edits into explanations","url":"https://escapemeridian.com/readings/arxiv-cs-ai-researchers-turn-recommendation-edits-into-e1a13c55"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.13914","revision":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","revisionKind":"sha256-source-text","title":"From Counterfactual Edits to Human Narratives: LLM-Driven Interpretations for Recommender System Explanations","url":"https://arxiv.org/abs/2611.13914"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter."},"revisions":[{"id":"e1a13c55755af77b9e83386972cbd435bed56fe538066cb4e00ff6cecc30852e","title":"arXiv cs.AI: Researchers turn recommendation edits into explanations","argument":"The paper’s authors are Vojtech Cahlik, Amir Reza Mohammadi, Rodrigo Alves, Pavel Kordik and Eva Zangerle. They argue that counterfactual explanation (CFX) methods can show which changes to a user’s interaction history would alter a recommendation, but often produce long edit lists that are hard to interpret. The authors investigate using large language models to summarize those lists as concise explanations of higher-level preference patterns. They optimize the mediators for correctness, informativeness and linguistic quality through preference optimization. On the MovieLens-1M dataset, they report high overall linguistic quality across multiple counterfactual generators and language models. The strongest semantic results came with high-success CFX methods such as ACCENT and SPINRec. Direct Preference Optimization improved semantic metrics in almost all evaluated settings. The work distinguishes generating valid CFXs from communicating model behavior in a user-understandable form. The authors evaluate explanations with human-calibrated metrics derived from separate human-labeled datasets.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports evaluation on the MovieLens-1M dataset; it does not establish whether the results generalize to other recommendation settings. The authors describe evaluation with human-calibrated metrics derived from separate human-labeled datasets; the abstract does not report direct evaluation of explanations with users.","speaker":{"name":"Cahlik, Vojtech, Mohammadi, Amir Reza, Alves, Rodrigo, Kordik, Pavel, Zangerle, Eva","roleAtSource":"Article author"},"source":{"title":"From Counterfactual Edits to Human Narratives: LLM-Driven Interpretations for Recommender System Explanations","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.13914","publishedAt":"2026-11-10","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors argue that recommender-system counterfactual explanations can require many edits and become difficult for users to interpret.","excerpt":"In recommendation settings, however, CFXs often require many edits, producing long, low-level lists that are hard for users to interpret.","url":"https://arxiv.org/abs/2611.13914","relationship":"Source-stated"},{"text":"The authors investigate using large language models to summarize raw counterfactual edit lists as concise explanations of higher-level preference patterns.","excerpt":"we investigate large language models as mediators that summarize raw CFX interaction edit lists into concise natural language explanations describing higher-level preference patterns.","url":"https://arxiv.org/abs/2611.13914","relationship":"Source-stated"},{"text":"On MovieLens-1M, the authors report high overall linguistic quality; the strongest semantic results came with ACCENT and SPINRec, and Direct Preference Optimization improved semantic metrics in almost all evaluated settings.","excerpt":"On the MovieLens-1M dataset, across multiple counterfactual generators and language models, the resulting summaries achieve high overall linguistic quality, while the strongest semantic results are obtained with high-success CFX methods such as ACCENT and SPINRec. Direct Preference Optimization","url":"https://arxiv.org/abs/2611.13914","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.13914","checkedAt":"2026-11-11T20:48:23.785Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Do the reported explanation-quality results hold with users and recommendation settings beyond the MovieLens-1M evaluation?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: Researchers turn recommendation edits into explanations","summary":"The paper’s authors are Vojtech Cahlik, Amir Reza Mohammadi, Rodrigo Alves, Pavel Kordik and Eva Zangerle. They argue that counterfactual explanation (CFX) methods can show which changes to a user’s interaction history would alter a recommendation, but often produce long edit lists that are hard to interpret. The authors investigate using large language models to summarize those lists as concise explanations of higher-level preference patterns. They optimize the mediators for correctness, informativeness and linguistic quality through preference optimization. On the MovieLens-1M dataset, they report high overall linguistic quality across multiple counterfactual generators and language models. The strongest semantic results came with high-success CFX methods such as ACCENT and SPINRec. Direct Preference Optimization improved semantic metrics in almost all evaluated settings. The work distinguishes generating valid CFXs from communicating model behavior in a user-understandable form. The authors evaluate explanations with human-calibrated metrics derived from separate human-labeled datasets.","question":"Do the reported explanation-quality results hold with users and recommendation settings beyond the MovieLens-1M evaluation?","claims":[{"text":"The authors argue that recommender-system counterfactual explanations can require many edits and become difficult for users to interpret.","excerpt":"In recommendation settings, however, CFXs often require many edits, producing long, low-level lists that are hard for users to interpret.","support":"source-stated"},{"text":"The authors investigate using large language models to summarize raw counterfactual edit lists as concise explanations of higher-level preference patterns.","excerpt":"we investigate large language models as mediators that summarize raw CFX interaction edit lists into concise natural language explanations describing higher-level preference patterns.","support":"source-stated"},{"text":"On MovieLens-1M, the authors report high overall linguistic quality; the strongest semantic results came with ACCENT and SPINRec, and Direct Preference Optimization improved semantic metrics in almost all evaluated settings.","excerpt":"On the MovieLens-1M dataset, across multiple counterfactual generators and language models, the resulting summaries achieve high overall linguistic quality, while the strongest semantic results are obtained with high-success CFX methods such as ACCENT and SPINRec. Direct Preference Optimization","support":"source-stated"}],"limitations":["The abstract reports evaluation on the MovieLens-1M dataset; it does not establish whether the results generalize to other recommendation settings.","The authors describe evaluation with human-calibrated metrics derived from separate human-labeled datasets; the abstract does not report direct evaluation of explanations with users."],"mediaId":"73f9b924ecdff16e747c57a37d6f668036f086fb5fb7db921fcd8adcf27d43a6","correctionReason":null,"id":"e1a13c55755af77b9e83386972cbd435bed56fe538066cb4e00ff6cecc30852e","publicationId":"f02ec2b0-af6b-4ec2-8c30-87f594b96c11","version":1,"sourceUrl":"https://arxiv.org/abs/2611.13914","sourceTitle":"From Counterfactual Edits to Human Narratives: LLM-Driven Interpretations for Recommender System Explanations","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-10","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-11T05:04:41.914Z","createdAt":"2026-11-11T20:16:34.046Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"73f9b924ecdff16e747c57a37d6f668036f086fb5fb7db921fcd8adcf27d43a6","kind":"article","url":"https://arxiv.org/abs/2611.13914","sourceUrl":"https://arxiv.org/abs/2611.13914","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-11T20:04:27.470Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.13914","checkedAt":"2026-11-11T20:48:23.785Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"73f9b924ecdff16e747c57a37d6f668036f086fb5fb7db921fcd8adcf27d43a6","confidence":0.73,"provider":"typesafe","model":"jev-1.13.0","inputHash":"52a8b2159f17f3ab61974a6fe8e407ae570b3834351a7bca21467e5b8a0a66ea","decidedAt":"2026-11-11T20:04:29.555Z"},"provenance":{"textHash":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","observationId":"28a2db32e3186bd89bff2dd10f8c4958209412dd61cbf697c7e07e8d94c44896","retrievedAt":"2026-11-11T20:04:27.044Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"6b9b03bc-d528-438e-8691-eb17d7615407"},"reviewer":{"model":"gpt-6-luna","runId":"38661ba1-dd64-4d04-88a9-a45fc9b2d503"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-11T20:16:34.046Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Cahlik, Vojtech","url":null,"kind":"unknown"},{"name":"Mohammadi, Amir Reza","url":null,"kind":"unknown"},{"name":"Alves, Rodrigo","url":null,"kind":"unknown"},{"name":"Kordik, Pavel","url":null,"kind":"unknown"},{"name":"Zangerle, Eva","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Cahlik, Vojtech","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Cahlik, Vojtech","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.13914","locator":"head/meta[name=author|citation_author]"},"assertionId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:48:23.785Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e","sourceUrl":"https://arxiv.org/abs/2611.13914","contentSha256":"85f7bfefa6cd0cdea63ecacb11f3399e539f632976f757798b9fc49cc68ba9e0","textHash":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","observedAt":"2026-11-11T20:48:23.785Z"}},{"actor":{"name":"Mohammadi, Amir Reza","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Mohammadi, Amir Reza","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.13914","locator":"head/meta[name=author|citation_author]"},"assertionId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:48:23.785Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e","sourceUrl":"https://arxiv.org/abs/2611.13914","contentSha256":"85f7bfefa6cd0cdea63ecacb11f3399e539f632976f757798b9fc49cc68ba9e0","textHash":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","observedAt":"2026-11-11T20:48:23.785Z"}},{"actor":{"name":"Alves, Rodrigo","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Alves, Rodrigo","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.13914","locator":"head/meta[name=author|citation_author]"},"assertionId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:48:23.785Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e","sourceUrl":"https://arxiv.org/abs/2611.13914","contentSha256":"85f7bfefa6cd0cdea63ecacb11f3399e539f632976f757798b9fc49cc68ba9e0","textHash":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","observedAt":"2026-11-11T20:48:23.785Z"}},{"actor":{"name":"Kordik, Pavel","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Kordik, Pavel","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.13914","locator":"head/meta[name=author|citation_author]"},"assertionId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:48:23.785Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e","sourceUrl":"https://arxiv.org/abs/2611.13914","contentSha256":"85f7bfefa6cd0cdea63ecacb11f3399e539f632976f757798b9fc49cc68ba9e0","textHash":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","observedAt":"2026-11-11T20:48:23.785Z"}},{"actor":{"name":"Zangerle, Eva","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zangerle, Eva","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.13914","locator":"head/meta[name=author|citation_author]"},"assertionId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:48:23.785Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e","sourceUrl":"https://arxiv.org/abs/2611.13914","contentSha256":"85f7bfefa6cd0cdea63ecacb11f3399e539f632976f757798b9fc49cc68ba9e0","textHash":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","observedAt":"2026-11-11T20:48:23.785Z"}}],"participationCoverage":{"total":5,"returned":5,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.13914","snapshotId":"39e6830ff13b17ab8d59d68dcb96df6086056c3f6ab1cbe802e8ab43d2c5231e","contentSha256":"85f7bfefa6cd0cdea63ecacb11f3399e539f632976f757798b9fc49cc68ba9e0","textHash":"9a21464204b7ba8c4467bc865a54ade709b9b23df627c38565e9fbc8f387efff","observedAt":"2026-11-11T20:48:23.785Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/e1a13c55755af77b9e83386972cbd435bed56fe538066cb4e00ff6cecc30852e/source"}}}],"audits":[{"publicationId":"f02ec2b0-af6b-4ec2-8c30-87f594b96c11","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The draft names the five listed authors and attributes the arguments, methods and results to them. It frames the results as reported findings, not established facts.","excerpt":"Authors:Vojtech Cahlik, Amir Reza Mohammadi, Rodrigo Alves, Pavel Kordik, Eva Zangerle"},{"dimension":"summary","status":"uncertain","explanation":"The summary is faithful, but a non-specialist may not follow its key terms and stakes. Three repairs would help. First, 'mediators' appears with no setup; say the LLMs act as mediators between raw edit lists and users. Second, explain 'semantic metrics' versus 'linguistic quality', plus ACCENT, SPINRec and MovieLens-1M, briefly or with tooltips. Third, the lead lists authors instead of the development. Add the authors' critique that evaluations emphasize edit counts and overlook coherence and actionability, and that the gap stems from a lack of user-centered abstraction.","excerpt":"most evaluations emphasize edit counts while largely overlooking whether the resulting CFXs are coherent and actionable for users"},{"dimension":"date","status":"pass","explanation":"originalPublished 2026-11-10 matches the v1 submission date on arXiv. The feed date of Nov 11 is later and correctly not used.","excerpt":"[Submitted on 10 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim matches the abstract verbatim in substance. The question and limitations follow from the evidence. The evaluation used human-calibrated metrics from separate datasets, not a reported direct user study, and only MovieLens-1M is named. No benefits or costs are invented.","excerpt":"evaluate the resulting explanations with human-calibrated metrics derived from separate human-labeled datasets"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper, which is the most direct access to the source.","excerpt":null}],"createdAt":"2026-11-11T20:16:34.046Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}