{"id":"W7160410202","doi":"10.1145/3786995.3786997","title":"From Transformers Come Themes: Evaluating BERTopic for Qualitative Analysis of Social Media Data","year":2025,"lang":"","type":"article","venue":"","topic":"Computational and Text Analysis Methods","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Latent Dirichlet allocation; Qualitative analysis; Social media; Qualitative research; Qualitative property; Sentence; Relevance (law); Topic model","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.006436863,0.0002351906,0.001181969,0.0006381969,0.0007625308,0.00009059722,0.001044319,0.0001766856,0.003373408],"category_scores_gemma":[0.002670404,0.0002122389,0.000845116,0.005208271,0.0005865606,0.0003195188,0.0001615072,0.000133863,0.000003013439],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001018753,"about_ca_system_score_gemma":0.0006912504,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006043944,"about_ca_topic_score_gemma":0.01723385,"domain_scores_codex":[0.9946439,0.002109677,0.001195021,0.0007361801,0.0009596392,0.0003556151],"domain_scores_gemma":[0.9759141,0.02258187,0.000447706,0.0003172811,0.0006596252,0.00007936997],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001093278,0.0002033255,0.0002930132,0.00005336732,0.01767197,1.287042e-7,0.6156952,0.0002162744,0.0001134682,0.1026257,0.0006231951,0.2623951],"study_design_scores_gemma":[0.0008802569,0.00003884464,0.004463613,0.00003143695,0.0210226,3.674832e-9,0.5329993,0.2837687,0.00005412459,0.1539901,0.002464449,0.000286483],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1623438,0.00106139,0.7948055,0.01094189,0.0009594085,0.0008241554,0.003697614,0.00004119679,0.02532513],"genre_scores_gemma":[0.9612657,0.00005723891,0.03444305,0.0001888904,0.0003155043,0.00002473684,0.00177486,0.000008341884,0.001921628],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.798922,"threshold_uncertainty_score":0.9975377,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3151434078999855,"score_gpt":0.5883443346130658,"score_spread":0.2732009267130803,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}