Publication Type Download Page

Transformer-Based Multi-Label Publication Type Prediction Download

TM (Transformer-Based Multi-Label Publication Type Prediction Model) is a single multi-label model that fine-tunes SPECTER2-base, a BERT-based encoder pre-trained on biomedical literature using citation-informed contrastive learning (see Publication Type Tagging using Transformer Models and Multi-Label Classification) . TM uses title, abstract, and metadata (e.g., numbers of references), with asymmetric loss, contrastive learning, and label smoothing applied during fine-tuning to improve robustness and calibration. TM gives probabilistic scores between 0 and 1, indicating the predicted probability that a given article is of a particular study design. Probabilistic scores can be converted to binary values using any chosen threshold, where scores above the threshold are classified as true and those below are false. There is no single threshold that is appropriate for all literature uses as users may prioritize specificity or sensitivity in retrieving literature given the different goals of different users, such as finding or excluding all items of a given study design.

File Downloads

All articles indexed by PubMed in English up through 12/31/2025: TM_scores_through2025.tsv.gz
# of rows (including header row): 34,955,046
Compressed size: 3.2 GB
Uncompressed size: 15.4 GB

All articles indexed by PubMed in English currently available for 2026.: TM_scores_2026.tsv.gz

File Format

Files are tab-delimited with a header row. There are 50 columns, listed below.

The PROBLEMATIC_INDICATORS column may contain any of the following character indicators:
C - the article is a target of a Comment
E - the article is a target of an Erratum
X - the article is a target of an Expression of Concern
P - the article was Republished
R - the article is a Reprint
T - the article was Retracted

File Columns

#	Column	Description	Broad Category
1	PMID	article id
2	PROBLEMATIC_INDICATORS	indicators of a problematic type
3	multicenter_study	probability 0.0000 to 1.0000	Interventional Trial Research
4	cross-over_studies	probability 0.0000 to 1.0000	Interventional Trial Research
5	double-blind_method	probability 0.0000 to 1.0000	Interventional Trial Research
6	random_allocation	probability 0.0000 to 1.0000	Interventional Trial Research
7	veterinary_randomized_controlled_trial	probability 0.0000 to 1.0000	Interventional Trial Research
8	randomized_controlled_trial_humans	probability 0.0000 to 1.0000	Interventional Trial Research
9	clinical_trial	probability 0.0000 to 1.0000	Interventional Trial Research
10	clinical_trial_phase_i	probability 0.0000 to 1.0000	Interventional Trial Research
11	clinical_trial_phase_ii	probability 0.0000 to 1.0000	Interventional Trial Research
12	clinical_trial_phase_iii	probability 0.0000 to 1.0000	Interventional Trial Research
13	clinical_trial_phase_iv	probability 0.0000 to 1.0000	Interventional Trial Research
14	clinical_trial_protocol	probability 0.0000 to 1.0000	Interventional Trial Research
15	controlled_clinical_trial	probability 0.0000 to 1.0000	Interventional Trial Research
16	equivalence_trial	probability 0.0000 to 1.0000	Interventional Trial Research
17	pragmatic_clinical_trial	probability 0.0000 to 1.0000	Interventional Trial Research
18	clinical_study	probability 0.0000 to 1.0000	Interventional Trial Research
19	follow-up_studies	probability 0.0000 to 1.0000	Observational Clinical Research
20	prospective_studies	probability 0.0000 to 1.0000	Observational Clinical Research
21	case-control_studies	probability 0.0000 to 1.0000	Observational Clinical Research
22	cohort_studies	probability 0.0000 to 1.0000	Observational Clinical Research
23	cross-sectional_studies	probability 0.0000 to 1.0000	Observational Clinical Research
24	longitudinal_studies	probability 0.0000 to 1.0000	Observational Clinical Research
25	retrospective_studies	probability 0.0000 to 1.0000	Observational Clinical Research
26	observational_study	probability 0.0000 to 1.0000	Observational Clinical Research
27	twin_study	probability 0.0000 to 1.0000	Qualitative & Genetic Methods
28	feasibility_studies	probability 0.0000 to 1.0000	Clinical Evaluation & Validation
29	case_reports	probability 0.0000 to 1.0000	Clinical Evaluation & Validation
30	case_series	probability 0.0000 to 1.0000	Clinical Evaluation & Validation
31	diagnostic_test_accuracy	probability 0.0000 to 1.0000	Clinical Evaluation & Validation
32	predictive_value_of_tests	probability 0.0000 to 1.0000	Clinical Evaluation & Validation
33	reproducibility_of_results	probability 0.0000 to 1.0000	Clinical Evaluation & Validation
34	genome-wide_association_study	probability 0.0000 to 1.0000	Qualitative & Genetic Methods
35	cross-cultural_comparison	probability 0.0000 to 1.0000	Qualitative & Genetic Methods
36	focus_groups	probability 0.0000 to 1.0000	Qualitative & Genetic Methods
37	interviews_as_topic	probability 0.0000 to 1.0000	Qualitative & Genetic Methods
38	historical_article	probability 0.0000 to 1.0000	Scholarly Discourse and Evidence Synthesis
39	interview	probability 0.0000 to 1.0000	Scholarly Discourse and Evidence Synthesis
40	systematic_review	probability 0.0000 to 1.0000	Scholarly Discourse and Evidence Synthesis
41	meta-analysis	probability 0.0000 to 1.0000	Scholarly Discourse and Evidence Synthesis
42	practice_guideline	probability 0.0000 to 1.0000	Scholarly Discourse and Evidence Synthesis
43	review	probability 0.0000 to 1.0000	Scholarly Discourse and Evidence Synthesis
44	systematic_reviews_as_topic	probability 0.0000 to 1.0000	Scholarly Discourse and Evidence Synthesis
45	meta-analysis_as_topic	probability 0.0000 to 1.0000	Scholarly Discourse and Evidence Synthesis
46	clinical_trials_as_topic	probability 0.0000 to 1.0000	Scholarly Discourse and Evidence Synthesis
47	expression_of_concern	probability 0.0000 to 1.0000	Scholarly Discourse and Evidence Synthesis
48	published_erratum	probability 0.0000 to 1.0000	Scholarly Discourse and Evidence Synthesis
49	retraction_of_publication	probability 0.0000 to 1.0000	Scholarly Discourse and Evidence Synthesis
50	editorial	probability 0.0000 to 1.0000	Scholarly Discourse and Evidence Synthesis

References:

Menke JD, Kilicoglu H, Smalheiser NR. Publication Type Tagging using Transformer Models and Multi-Label Classification. AMIA Annu Symp Proc. 2025 May 22;2024:818-827.

Menke JD, Ming S, Radhakrishna S, Kilicoglu H, Smalheiser NR. Enhancing automated indexing of publication types and study designs in biomedical literature using full-text features. medRxiv [Preprint]. 2025 Apr 28:2025.04.23.25326300. doi: 10.1101/2025.04.23.25326300.