NAMEXTEND

홈페이지 · 논문 1편

This dataset extends NAMEXACT by including words that can be used as names, but may not exclusively be used as names in every context. ## Dataset Details ### Dataset Description <!-- Provide a longer summary of what this dataset is. --> Unlike NAMEXACT, this datasets contains *words* that are mostly used as *names*, but may also be used in other contexts, such as - *Christian* (believer in Christianity) - *Drew* (simple past of the verb to draw) - *Florence* (an Italian city) - *Henry* (the SI unit of inductance) - *Mercedes* (a car brand) In addition, names with ambiguous gender are included - once for each gender. For instance, Skyler is included as female (F) name with a probability of 37.3%, and as male (M) name with a probability of 62.7%. ### Dataset Sources [optional] <!-- Provide the basic links for the dataset. --> - Repository: github.com/aieng-lab/gradiend - Original Dataset: Gender by Name ## Dataset Structure - name: the name - gender: the gender of the name (M for male and F for female) - count: the count value of this name (raw value from the original dataset) - probability: the probability of this name (raw value from original dataset; not normalized to this dataset!) - gender_agreement: a value describing the certainty that this name has an unambiguous gender computed as the maximum probability of that name across both genders, e.g., $max(37.7%, 62.7%)=62.7%$ for *Skyler*. For names with a unique gender in this dataset, this value is 1.0 - primary_gender: is equal to gender for names with a unique gender in this dataset, and equals otherwise the gender of that name with higher probability - genders: label B if *both* genders are contained for this name in this dataset, otherwise equal to gender - prob_F: the probability of that name being used as a female name (i.e., 0.0 or 1.0 if genders != B) - prob_M: the probability of that name being used as a male name ## Dataset Creation ### Source Data <!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). --> The data is created by filtering Gender by Name. #### Data Collection and Processing <!-- This section describes the data collection and processing process such as data selection criteria, filtering and normalization methods, tools and libraries used, etc. --> The original data is filtered to contain only names with a count of at least 100 to remove very rare names. This threshold reduces the total number of names by $72%, from 133910 to 37425. ## Bias, Risks, and Limitations <!-- This section is meant to convey both technical and sociotechnical limitations. --> The original dataset provides counts of names (with their gender) for male and female babies from open-source government authorities in the US (1880-2019), UK (2011-2018), Canada (2011-2018), and Australia (1944-2019) in these periods