BayanDuygu commited on
Commit
60d89a9
·
verified ·
1 Parent(s): cc8657e

edited benchmarking info

Browse files
Files changed (1) hide show
  1. README.md +20 -3
README.md CHANGED
@@ -432,8 +432,6 @@ political animosity, animal aversion, vegan antipathy, ethnic group hostility, a
432
 
433
  The dataset includes 4 labels, **offensive**, **hate**, **neutral** and **civilized**.
434
 
435
- Hate speech detection task is also a part of [SentiTurca](https://huggingface.co/datasets/turkish-nlp-suite/SentiTurca) benchmark. Details of benchmarking and benchmark code can be found under [SentiTurca repo](https://huggingface.co/datasets/turkish-nlp-suite/SentiTurca) and [SentiTurca Github repo](https://github.com/turkish-nlp-suite/SentiTurca).
436
-
437
 
438
  Here is the distribution of target groups:
439
 
@@ -453,7 +451,7 @@ Here is the distribution of target groups:
453
  | Veganism | 1.3K |
454
  | Total | 52K |
455
 
456
- All text is scraped from Eksisozluk.com in a targeted manner and sampled. The annotations are done by the data company [Co-one](https://www.co-one.co/).
457
 
458
 
459
  ## Dataset Instances
@@ -476,5 +474,24 @@ An instance looks like:
476
  |---------|----:|---:|---:|
477
  |Turkish Hate Map|42175|5000|5000|
478
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
479
  ## Citation
480
  Coming soon!!
 
432
 
433
  The dataset includes 4 labels, **offensive**, **hate**, **neutral** and **civilized**.
434
 
 
 
435
 
436
  Here is the distribution of target groups:
437
 
 
451
  | Veganism | 1.3K |
452
  | Total | 52K |
453
 
454
+ All text is scraped from Eksisozluk.com in a targeted manner and sampled. The annotations are done by the data company [Co-one](https://www.co-one.co/). For more details please refer to the [research paper]()
455
 
456
 
457
  ## Dataset Instances
 
474
  |---------|----:|---:|---:|
475
  |Turkish Hate Map|42175|5000|5000|
476
 
477
+
478
+
479
+ ## Benchmarking
480
+ This dataset is a part of [SentiTurca](https://huggingface.co/datasets/turkish-nlp-suite/SentiTurca) benchmark, in the benchmark the subset name is **hate**, named according to the GLUE tasks.
481
+ Model benchmarking information can be found under SentiTurca HF repo and benchmarking scripts can be found under [SentiTurca Github repo](https://github.com/turkish-nlp-suite/SentiTurca).
482
+
483
+ For this dataset we benchmarked a transformer based model BERTurk and a handful of LLMs. Success of each model is follows:
484
+
485
+ | Model | acc./F1 |
486
+ |---|---|
487
+ | Gemini 1.0 Pro | 0.33/0.29 |
488
+ | GPT-4 Turbo | 0.38/0.32 |
489
+ | Claude 3 Sonnet | 0.16/0.29 |
490
+ | Llama 3 70B | 0.55/0.35 |
491
+ | Qwen2-72B | 0.70/0.35 |
492
+ | BERTurk | 0.61/0.58 |
493
+
494
+ For a critique of the results, misclassified instances and more please consult to the [research paper]().
495
+
496
  ## Citation
497
  Coming soon!!