Wise hate speech (#377)

assets/ar/factuality_disinformation_harmful_content/hate_speech/OSACT4SubtaskB_GPT4_FewShot_Arabic.py

@@ -0,0 +1,91 @@
+from llmebench.datasets import OSACT4SubtaskBDataset
+from llmebench.models import OpenAIModel
+from llmebench.tasks import HateSpeechTask
+
+
+def metadata():
+    return {
+        "author": "Mohamed Bayan Kmainasi, Rakif Khan, Ali Ezzat Shahroor, Boushra Bendou, Maram Hasanain, and Firoj Alam",
+        "affiliation": "Arabic Language Technologies, Qatar Computing Research Institute (QCRI), Hamad Bin Khalifa University (HBKU)",
+        "model": "GPT-4o-2024-05-22",
+        "description": "For a comprehensive analysis and results, refer to our peer-reviewed publication available at [Springer](https://doi.org/10.1007/978-981-96-0576-7_30) or explore the preprint version on [arXiv](https://arxiv.org/abs/2409.07054).",
+    }
+
+
+def config():
+    return {
+        "dataset": OSACT4SubtaskBDataset,
+        "task": HateSpeechTask,
+        "model": OpenAIModel,
+        "model_args": {
+            "max_tries": 3,
+        },
+    }
+
+
+def prompt(input_sample, examples):
+    base_prompt = (
+        'هل تحتوي التغريدة التالية على خطاب كراهية؟ أجب بـ "كراهية" إن احتوت على خطاب كراهية، و أجب بـ "ليس كراهية" إن لم تكن كذلك.'
+        " هذه بعض الأمثلة:\n"
+    )
+
+    return [
+        {
+            "role": "system",
+            "content": "أنت خبير في تحليل و تصنيف التغريدات.",
+        },
+        {
+            "role": "user",
+            "content": few_shot_prompt(input_sample, base_prompt, examples),
+        },
+    ]
+
+
+def few_shot_prompt(input_sample, base_prompt, examples):
+    out_prompt = base_prompt + "\n"
+    for example in examples:
+        # Found chatgpt confused when using 0 and 1 in the prompt
+        # label = "no" if example["label"] == "0" else "yes"
+        label = "ليس كراهية" if example["label"] == "NOT_HS" else "كراهية"
+        out_prompt = (
+            out_prompt + "التغريدة: " + example["input"] + "التصنيف: " + label + "\n\n"
+        )
+
+    # Append the sentence we want the model to predict for but leave the Label blank
+    out_prompt = out_prompt + "التغريدة: " + input_sample + "التصنيف:\n"
+
+    # print("=========== FS Prompt =============\n")
+    # print(out_prompt)
+
+    return out_prompt
+
+
+def post_process(response):
+    out = response["choices"][0]["message"]["content"]
+    label = out.lower().strip()
+
+    if (
+        "ليس" in label
+        or "ليس كراهية" in label
+        or "لا" in label
+        or "no" in label
+        or "not" in label
+        or "don't" in label
+        or "not_hs" in label
+        or "not_hatespeech" in label
+        or "not_hate_speech" in label
+    ):
+        return "NOT_HS"
+    elif (
+        "كراهية" in label
+        or "نعم" in label
+        or "أجل" in label
+        or "yes" in label
+        or "contins" in label
+        or "hs" in label
+        or "hatespeech" in label
+        or "hate speech" in label
+    ):
+        return "HS"
+    else:
+        return None

assets/ar/factuality_disinformation_harmful_content/hate_speech/OSACT4SubtaskB_GPT4_FewShot_English.py

@@ -0,0 +1,91 @@
+from llmebench.datasets import OSACT4SubtaskBDataset
+from llmebench.models import OpenAIModel
+from llmebench.tasks import HateSpeechTask
+
+
+def metadata():
+    return {
+        "author": "Mohamed Bayan Kmainasi, Rakif Khan, Ali Ezzat Shahroor, Boushra Bendou, Maram Hasanain, and Firoj Alam",
+        "affiliation": "Arabic Language Technologies, Qatar Computing Research Institute (QCRI), Hamad Bin Khalifa University (HBKU)",
+        "model": "GPT-4o-2024-05-22",
+        "description": "For a comprehensive analysis and results, refer to our peer-reviewed publication available at [Springer](https://doi.org/10.1007/978-981-96-0576-7_30) or explore the preprint version on [arXiv](https://arxiv.org/abs/2409.07054).",
+    }
+
+
+def config():
+    return {
+        "dataset": OSACT4SubtaskBDataset,
+        "task": HateSpeechTask,
+        "model": OpenAIModel,
+        "model_args": {
+            "max_tries": 3,
+        },
+    }
+
+
+def prompt(input_sample, examples):
+    base_prompt = (
+        "Does the tweet contain hate speech? Answer by "
+        " 'hate speech' if it does, and 'not hate speech' otherwise. Here are some examples:\n"
+    )
+
+    return [
+        {
+            "role": "system",
+            "content": "You are an expert in tweets analysis and classification.",
+        },
+        {
+            "role": "user",
+            "content": few_shot_prompt(input_sample, base_prompt, examples),
+        },
+    ]
+
+
+def few_shot_prompt(input_sample, base_prompt, examples):
+    out_prompt = base_prompt + "\n"
+    for example in examples:
+        # Found chatgpt confused when using 0 and 1 in the prompt
+        # label = "no" if example["label"] == "0" else "yes"
+        label = "not hate speech" if example["label"] == "NOT_HS" else "hate speech"
+        out_prompt = (
+            out_prompt + "tweet: " + example["input"] + "\nlabel: " + label + "\n\n"
+        )
+
+    # Append the sentence we want the model to predict for but leave the Label blank
+    out_prompt = out_prompt + "tweet: " + input_sample + "\nlabel:\n"
+
+    # print("=========== FS Prompt =============\n")
+    # print(out_prompt)
+
+    return out_prompt
+
+
+def post_process(response):
+    out = response["choices"][0]["message"]["content"]
+    label = out.lower().strip()
+
+    if (
+        "ليس" in label
+        or "ليس كراهية" in label
+        or "لا" in label
+        or "no" in label
+        or "not" in label
+        or "don't" in label
+        or "not_hs" in label
+        or "not_hatespeech" in label
+        or "not_hate_speech" in label
+    ):
+        return "NOT_HS"
+    elif (
+        "كراهية" in label
+        or "نعم" in label
+        or "أجل" in label
+        or "yes" in label
+        or "contins" in label
+        or "hs" in label
+        or "hatespeech" in label
+        or "hate speech" in label
+    ):
+        return "HS"
+    else:
+        return None

assets/ar/factuality_disinformation_harmful_content/hate_speech/OSACT4SubtaskB_GPT4_FewShot_Mixed.py

@@ -0,0 +1,85 @@
+from llmebench.datasets import OSACT4SubtaskBDataset
+from llmebench.models import OpenAIModel
+from llmebench.tasks import HateSpeechTask
+
+
+def metadata():
+    return {
+        "author": "Mohamed Bayan Kmainasi, Rakif Khan, Ali Ezzat Shahroor, Boushra Bendou, Maram Hasanain, and Firoj Alam",
+        "affiliation": "Arabic Language Technologies, Qatar Computing Research Institute (QCRI), Hamad Bin Khalifa University (HBKU)",
+        "model": "GPT-4o-2024-05-22",
+        "description": "For a comprehensive analysis and results, refer to our peer-reviewed publication available at [Springer](https://doi.org/10.1007/978-981-96-0576-7_30) or explore the preprint version on [arXiv](https://arxiv.org/abs/2409.07054).",
+    }
+
+
+def config():
+    return {
+        "dataset": OSACT4SubtaskBDataset,
+        "task": HateSpeechTask,
+        "model": OpenAIModel,
+        "model_args": {
+            "max_tries": 3,
+        },
+    }
+
+
+def prompt(input_sample, examples):
+    base_prompt = 'هل تحتوي التغريدة التالية على خطاب كراهية؟ أجب بـ "hate speech" إن احتوت على خطاب كراهية، و أجب بـ "not hate speech" إن لم تكن كذلك.\n'
+
+    return [
+        {
+            "role": "system",
+            "content": "أنت خبير في تحليل و تصنيف التغريدات.",
+        },
+        {
+            "role": "user",
+            "content": few_shot_prompt(input_sample, base_prompt, examples),
+        },
+    ]
+
+
+def few_shot_prompt(input_sample, base_prompt, examples):
+    out_prompt = base_prompt + "\n"
+    for example in examples:
+        # Found chatgpt confused when using 0 and 1 in the prompt
+        # label = "no" if example["label"] == "0" else "yes"
+        label = "not hate speech" if example["label"] == "NOT_HS" else "hate speech"
+        out_prompt = (
+            out_prompt + "التغريدة: " + example["input"] + "التصنيف: " + label + "\n\n"
+        )
+
+    # Append the sentence we want the model to predict for but leave the Label blank
+    out_prompt = out_prompt + "التغريدة: " + input_sample + "التصنيف:\n"
+
+    return out_prompt
+
+
+def post_process(response):
+    out = response["choices"][0]["message"]["content"]
+    label = out.lower().strip()
+
+    if (
+        "ليس" in label
+        or "ليس كراهية" in label
+        or "لا" in label
+        or "no" in label
+        or "not" in label
+        or "don't" in label
+        or "not_hs" in label
+        or "not_hatespeech" in label
+        or "not_hate_speech" in label
+    ):
+        return "NOT_HS"
+    elif (
+        "كراهية" in label
+        or "نعم" in label
+        or "أجل" in label
+        or "yes" in label
+        or "contins" in label
+        or "hs" in label
+        or "hatespeech" in label
+        or "hate speech" in label
+    ):
+        return "HS"
+    else:
+        return None

assets/ar/factuality_disinformation_harmful_content/hate_speech/OSACT4SubtaskB_GPT4_ZeroShot_Arabic.py

@@ -0,0 +1,72 @@
+from llmebench.datasets import OSACT4SubtaskBDataset
+from llmebench.models import OpenAIModel
+from llmebench.tasks import HateSpeechTask
+
+
+def metadata():
+    return {
+        "author": "Mohamed Bayan Kmainasi, Rakif Khan, Ali Ezzat Shahroor, Boushra Bendou, Maram Hasanain, and Firoj Alam",
+        "affiliation": "Arabic Language Technologies, Qatar Computing Research Institute (QCRI), Hamad Bin Khalifa University (HBKU)",
+        "model": "GPT-4o-2024-05-22",
+        "description": "For a comprehensive analysis and results, refer to our peer-reviewed publication available at [Springer](https://doi.org/10.1007/978-981-96-0576-7_30) or explore the preprint version on [arXiv](https://arxiv.org/abs/2409.07054).",
+    }
+
+
+def config():
+    return {
+        "dataset": OSACT4SubtaskBDataset,
+        "task": HateSpeechTask,
+        "model": OpenAIModel,
+        "model_args": {
+            "class_labels": ["HS", "NOT_HS"],
+            "max_tries": 3,
+        },
+    }
+
+
+def prompt(input_sample):
+    return [
+        {
+            "role": "system",
+            "content": "أنت خبير في تحليل و تصنيف التغريدات.",
+        },
+        {
+            "role": "user",
+            "content": f' هل تحتوي التغريدة التالية على خطاب كراهية؟ أجب بـ "كراهية" إن احتوت على خطاب كراهية، و أجب بـ "ليس كراهية" إن لم تكن كذلك. \n'
+            + "التغريدة: "
+            + input_sample
+            + "\n"
+            + "التصنيف: ",
+        },
+    ]
+
+
+def post_process(response):
+    out = response["choices"][0]["message"]["content"]
+    label = out.lower().strip()
+
+    if (
+        "ليس" in label
+        or "ليس كراهية" in label
+        or "لا" in label
+        or "no" in label
+        or "not" in label
+        or "don't" in label
+        or "not_hs" in label
+        or "not_hatespeech" in label
+        or "not_hate_speech" in label
+    ):
+        return "NOT_HS"
+    elif (
+        "كراهية" in label
+        or "نعم" in label
+        or "أجل" in label
+        or "yes" in label
+        or "contins" in label
+        or "hs" in label
+        or "hatespeech" in label
+        or "hate speech" in label
+    ):
+        return "HS"
+    else:
+        return None