{"id":"W4416828803","doi":"10.1016/j.jss.2025.112714","title":"CREST: Improving interpretability and effectiveness of troubleshooting at ericsson through criterion-Specific trouble report retrieval","year":2025,"lang":"en","type":"article","venue":"Journal of Systems and Software","topic":"Software Engineering Research","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Carleton University; Ericsson (Canada)","funders":"Telefonaktiebolaget LM Ericsson; Carleton University","keywords":"Troubleshooting; Interpretability; Relevance (law); Software; Volume (thermodynamics); Data retrieval; Human–computer information retrieval","routes":{"ca_aff":true,"ca_fund":true,"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":[],"consensus_categories":[],"category_scores_codex":[0.002928906,0.0001542444,0.0005470549,0.0001830376,0.0001230742,0.0002071418,0.0003509334,0.0001013639,0.000001028903],"category_scores_gemma":[0.001840211,0.0001282602,0.00009581959,0.000382446,0.00007776912,0.0006136405,0.0003989096,0.0002897984,2.200075e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002225496,"about_ca_system_score_gemma":0.0001296031,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006988989,"about_ca_topic_score_gemma":9.077464e-7,"domain_scores_codex":[0.9980714,0.0002031774,0.0007621878,0.000314402,0.0004177952,0.0002309909],"domain_scores_gemma":[0.9966859,0.001913321,0.0004035258,0.0004021479,0.0005031959,0.00009195843],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006505401,0.000118664,0.9518338,0.006984026,0.0002013022,0.0005011064,0.002187916,0.0006692247,0.02626358,0.001551572,0.0001476916,0.008890538],"study_design_scores_gemma":[0.003540917,0.001229192,0.9547014,0.00744078,0.00006960498,0.005091733,0.0005952022,0.008583119,0.01493112,0.001204083,0.001962915,0.0006499945],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6196261,0.005710672,0.3737779,0.00002504587,0.0006732025,0.0001459537,0.000001839761,0.00003096502,0.000008312574],"genre_scores_gemma":[0.9905075,0.00007956544,0.009298838,0.000003836687,0.00005925065,0.000003104172,5.132018e-7,0.00001013148,0.00003726157],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3708814,"threshold_uncertainty_score":0.5230299,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01474976466855962,"score_gpt":0.2766366461209478,"score_spread":0.2618868814523882,"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."}}