frett commited on
Commit
c46f95d
·
verified ·
1 Parent(s): ec83577

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +6 -15
README.md CHANGED
@@ -304,27 +304,18 @@ dataset_info:
304
  dataset_size: 1114583
305
  ---
306
 
307
- **Text2DistBench** is a benchmark for evaluating whether large language models can infer distributional knowledge from natural language evidence.
308
- Given metadata and a set of user comments about an entity (e.g., a movie or song), models must estimate statistics such as:
309
- stance distribution, topic distribution, and most/second-most frequent labels.
310
 
311
  The dataset is constructed from movie and music entities released between 2025-12-01 and 2026-03-01.
312
-
313
  It includes two types of configurations:
314
 
315
  **Posterior configurations** include both metadata and user comments as evidence.
316
- These configurations differ by sample size (50 or 100 comments) and task type:
317
- `posterior_sampled_50_estimation`,
318
- `posterior_sampled_50_most_freq`,
319
- `posterior_sampled_50_second_freq`,
320
- `posterior_sampled_100_estimation`,
321
- `posterior_sampled_100_most_freq`,
322
- `posterior_sampled_100_second_freq`
323
 
324
- **Prior configurations** contain only metadata without comments:
325
- `prior_estimation`,
326
- `prior_most_freq`,
327
- `prior_second_freq`
328
 
329
 
330
  ---
 
304
  dataset_size: 1114583
305
  ---
306
 
307
+ **Text2DistBench** is a reading comprehension benchmark for evaluating whether large language models can infer distributional knowledge from natural language evidence.
308
+ Given metadata and a set of user comments about an entity (e.g., a movie or song), models must understand and response statistics such as:
309
+ stance/topic distribution, and most/second- frequent labels.
310
 
311
  The dataset is constructed from movie and music entities released between 2025-12-01 and 2026-03-01.
 
312
  It includes two types of configurations:
313
 
314
  **Posterior configurations** include both metadata and user comments as evidence.
315
+ These configurations differ by sample size (50 or 100 comments) and task type (estimation, most_freq, second_freq):
316
+ `posterior_sampled_<comment_num>_<task>`
 
 
 
 
 
317
 
318
+ **Prior configurations** contain only metadata without comments: `prior_<task>`
 
 
 
319
 
320
 
321
  ---