These are the parameters used for the experiment in the paper:
Ismail Harrando, Pasquale Lisena and Raphaël Troncy. Apples to Apples: A Systematic Evaluation of Topic Models. In Recent Advances in Natural Language Processing (RANLP), September 2021.
| 20NG | AFP | Yahoo-Balanced | Yahoo-Unbalanced | ||
|---|---|---|---|---|---|
| All | num topics | 20 | 17 | 26 | 26 |
| LDA | alpha | 0.1 | 0.1 | 10 | 10 |
| beta | null | null | null | null | |
| random_seed | 5 | 5 | 5 | 5 | |
| iterations | 800 | 1000 | 1500 | 1400 | |
| optimize_interval | 10 | 10 | 10 | 20 | |
| topic_threshold | 0 | 0 | 0 | 0 | |
| LFTM | alpha | 1 | 0.1 | 1.0 | 1.0 |
| beta | 1 | 0.1 | 0.1 | 0.1 | |
| lambda | 1 | 1 | 1 | 1 | |
| initer | 1000 | 800 | 500 | 1000 | |
| niter | 200 | 200 | 100 | 100 | |
| topn | 10 | 10 | 10 | 10 | |
| D2T | batch_size | 6144 | 6144 | 6144 | 3072 |
| n_epochs | 80 | 15 | 20 | 40 | |
| lr | 0.05 | 0.04 | 0.05 | 0.1 | |
| l1_doc | 0.000002 | 0.000002 | 0.000002 | 0.000002 | |
| l1_word | 0.000000015 | 0.000000015 | 0.000000015 | 0.000000015 | |
| word_dim | 0 | 0 | 0 | 0 | |
| GSDMM | alpha | 0.1 | 0.1 | 1.0 | 1.0 |
| beta | 0.1 | 0.1 | 1.0 | 1.0 | |
| n_iter | 15 | 10 | 10 | 10 | |
| CTM | bert_input_size | 768 | 768 | 768 | 768 |
| num_epochs | 200 | 100 | 25 | 25 | |
| hidden_sizes | (100,) | (100,) | (200,) | (200,) | |
| batch_size | 200 | 200 | 200 | 10 | |
| dropout | 0.4 | 0.4 | 0.4 | 10 | |
| inference_type | 'combined' | 'combined' | 'combined' | 'combined' | |
| PVTM | vector_size | 50 | 50 | 50 | 50 |
| hs | 0 | 0 | 0 | 0.1 | |
| dbow_words | 1 | 1 | 1 | 10 | |
| dm | 0 | 0 | 0 | 10 | |
| epochs | 30 | 100 | 100 | 30 | |
| window | 20 | 20 | 20 | 10 | |
| seed | 123 | 123 | 123 | 10 | |
| min_count | 5 | 5 | 5 | 10 | |
| workers | 5 | 5 | 5 | 10 | |
| alpha | 0.1 | 0.05 | 0.05 | 0.01 | |
| min_alpha | 0.025 | 0.25 | 0.25 | 10 | |
| random_state | 123 | 123 | 123 | 10 | |
| covariance_type | 'diag' | 'diag' | 'diag' | 10 | |
| HDP | max_chunks | None | None | None | 0.1 |
| max_time | None | None | None | 0.1 | |
| chunksize | 256 | 256 | 256 | 10 | |
| kappa | 1.0 | 1.0 | 1.0 | 10 | |
| tau | 64.0 | 64.0 | 64.0 | 10 | |
| K | 15 | 15 | 15 | 10 | |
| T | 150 | 150 | 150 | 10 | |
| alpha | 1 | 1 | 1 | 10 | |
| gamma | 1 | 1 | 1 | 10 | |
| eta | 0.1 | 0.1 | 0.1 | 0.1 | |
| scale | 2.0 | 2.0 | 2.0 | 2.0 | |
| var_converge | None | None | None | 10 | |
| random_state | None | None | None | 10 | |
| LSI | use_tfidf | False | True | True | True |
| chunksize | 20000 | 20000 | 20000 | 10 | |
| decay | 1.0 | 2.0 | 2.0 | 1.0 | |
| distributed | False | False | False | 10 | |
| onepass | True | True | True | 10 | |
| power_iters | 2 | 2 | 2 | 10 | |
| extra_samples | 100 | 100 | 3000 | 3000 | |
| NMF | passes | 1 | 1 | 1 | 0.1 |
| kappa | 2.0 | 2.0 | 2.0 | 10 | |
| minimum_probability | 0.001 | 0.01 | 0.01 | 10 | |
| w_max_iter | 200 | 200 | 200 | 10 | |
| w_stop_condition | 0.0001 | 0.0001 | 0.0001 | 10 | |
| h_max_iter | 50 | 50 | 50 | 10 | |
| h_stop_condition | 0.01 | 0.01 | 0.01 | 10 | |
| eval_every | 10 | 10 | 10 | 10 | |
| normalize | True | True | True | 10 | |
| random_state | None | None | None | 10 |