tum-nlp/gpt-2-medium-target-aware-counterspeech-generation

---

Target-Aware Counter-Speech Generation

The target-aware counter-speech generation model is an autoregressive generative language model fine-tuned on hate- and counter-speech pairs from the CONAN datasets for generating more contextually relevant counter-speech, based on the gpt2-medium model. The model utilizes special tokens that embedded target demographic information to guide the generation towards more relevant responses, avoiding off-topic and generic responses. The model is trained on 8 target demographics, including Migrants, People of Color (POC), LGBT+, Muslims, Women, Jews, Disabled, and Other.

Uses

The model is intended for generating counter-speech responses for a given hate speech sequence, combined with special tokens for target-demographic embeddings.

Bias, Risks, and Limitations

We observed negative effects such as content hallucination and toxic response generation. Though the intended use is to generate counter-speech for combating online hatred, the usage is to be monitored carefully with human post-editing or approval system, ensuring safe and inclusive online environment.

How to Get Started with the Model

Use the code below to get started with the model.

types = ["MIGRANTS", "POC", "LGBT+", "MUSLIMS", "WOMEN", "JEWS", "other", "DISABLED"] # A list of all available target-demographic tokens
from transformers import AutoTokenizer, AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(tum-nlp/gpt-2-medium-target-aware-counterspeech-generation)
tokenizer = AutoTokenizer.from_pretrained(tum-nlp/gpt-2-medium-target-aware-counterspeech-generation)
tokenizer.padding_side = "left" 

prompt = "<|endoftext|> <other> Hate-speech: Human are not created equal, some are born lesser. Counter-speech: "
input = tokenizer(prompt, return_tensors="pt", padding=True)
output_sequences = model.generate(
        input_ids=inputs['input_ids'].to(model.device),
        attention_mask=inputs['attention_mask'].to(model.device),
        pad_token_id=tokenizer.eos_token_id,
        max_length=128,
        num_beams=3,
        no_repeat_ngram_size=3,
        num_return_sequences=1,
        early_stopping=True
    )
  result = tokenizer.decode(output_sequences, skip_special_tokens=True)

Training Hyperparameters

training_args = TrainingArguments(
  num_train_epochs=20,
  learning_rate=3.800568576836524e-05,
  weight_decay=0.050977894796868116,
  warmup_ratio=0.10816909354342182,
  optim="adamw_torch",
  lr_scheduler_type="cosine",
  evaluation_strategy="epoch",
  save_strategy="epoch",
  save_total_limit=3,
  load_best_model_at_end=True,
  auto_find_batch_size=True,
)

Evaluation

Testing Data, Factors & Metrics

Testing Data

The model's performance is tested on three test sets, from which two are subsets of the CONAN dataset and one is the sexist portion of the EDOS dataset

Metrics

The model's performance is tested on a custom evaluation pipeline for counter-speech generation. The pipeline includes CoLA, Toxicity, Hatefulness, Offensiveness, Label and Context Similarity, Validity as Counter-Speech, Repetition Rate, target-demographic F1 and the Arithmetic Mean

Results

CONAN

Model Name	CoLA	TOX	Hate	OFF	L Sim	C Sim	VaCS	RR	F1	AM
Human	0.937	0.955	1.000	0.997	-	0.751	0.980	0.861	0.885	0.929
target-aware gpt2-medium	0.958	0.946	1.000	0.996	0.706	0.784	0.946	0.419	0.880	0.848

CONAN SMALL

Model Name	CoLA	TOX	Hate	OFF	L Sim	C Sim	VaCS	RR	F1	AM
Human	0.963	0.956	1.000	1.000	1.000	0.768	0.988	0.995	0.868	0.949
target-aware gpt2-medium	0.975	0.931	1.000	1.000	0.728	0.783	0.888	0.911	0.792	0.890

EDOS

Model Name	CoLA	TOX	Hate	OFF	C Sim	VaCS	RR	F1	AM
target-aware gpt2-medium	0.930	0.815	0.999	0.975	0.689	0.857	0.518	0.747	0.816