opensubtitles_inferred

OpenSubtitles- OpenSubtitles contains subtitles for movies in different languages. The curators retained English subtitles that contain a character name or identity. They annotated the character's gender using gender kinship terms such as daughter and gender probability distribution calculated by counting the masculine and feminine names of baby names in the United States. Using the character's gender, they produced labels for the AS dimension. They produced labels for the TO dimension by taking the gender of the next character to speak if there is another utterance in the conversation; otherwise, they take the gender of the last character to speak. They impute ABOUT labels on this dataset using a classifier trained on the datasets 1-4. The Multi-Dimensional Gender Bias Classification dataset is based on a general framework that decomposes gender bias in text along several pragmatic and semantic dimensions: bias from the gender of the person being spoken about, bias from the gender of the person being spoken to, and bias from the gender of the speaker. It contains seven large scale datasets automatically annotated for gender information (there are eight in the original project but the Wikipedia set is not included in the HuggingFace distribution), one crowdsourced evaluation benchmark of utterance-level gender rewrites, a list of gendered names, and a list of gendered words in English. text-classification-other-gender-bias: The dataset can be used to train a model for classification of various kinds of gender bias. The model performance is evaluated based on the accuracy of the predicted labels as compared to the given labels in the dataset. Dinan et al's (2020) Transformer model achieved an average of 67.13 accuracy in binary gender prediction across the ABOUT, TO, and AS tasks. This is the dataset 'opensubtitles_inferred' from 'Yelp inferred' split, it description is as follows: text: The actual text data from Yelp reviews. binary_label: Indicates the inferred gender (e.g., "ABOUT:female (0)" or "ABOUT:male (1)") based on the content. binary_score: A confidence score associated with the binary_label. ternary_label: Provides a more detailed classification, possibly including "ABOUT:gender-neutral" in addition to the binary labels. ternary_score(target): A confidence score associated with the ternary_label. paper_url = "https://arxiv.org/pdf/1509.01626" original_data_url = "https://huggingface.co/datasets/facebook/md_gender_bias/tree/10c34c50ef78b4a42f6d4eeac80a0ef2d190cd07/opensubtitles_inferred"