* If you want to update the article please login/register
The aim of this project is to help eliminate the already-existing gender wage gap by making objective job recommendations based on resumes from job seekers. To remove gender from word2vec and 900k resumes, we use a collaborative adversary network to eliminate word2vec images of 12 million jobs vacancies texts and 900k resumes. We conclude that adversarial debiasing of word representations can improve real-world fairness of applications and therefore may be part of the solution for building fairness-aware recommendation schemes.
Source link: https://arxiv.org/abs/2209.09592v1
Mitigating gender bias in information retrieval is crucial in avoiding propagating stereotypes. In this research, we use a database consisting of two components: relevance of a document to a question and a "gender" of a paper, in which pronouns are replaced by male, female, and neutral conjugations, according to the author. Pre-trained IR models for IR do not do well in zero-shot retrieval tasks, and we found that lightweight fine-tuning with adapter networks raises zero-shot retrieval results almost 20% over baseline.
Source link: https://arxiv.org/abs/2208.01755v3
In several applications, deep learning algorithms for predicting neuroimaging results have demonstrated a great deal. Prior research has shown that deep learning models that use the data's 3D structure can outperform standard machine learning on several learning tasks. We investigate structural MRI results to predict gender and detect gender-related changes in brain structure within the Adolescent Brain and Cognitive Development project, a large longitudinal growth study. Typical frontal areas and the temporal lobe were identified as the most discriminative in the task under investigation, with major brain regions and the temporal lobe.
Source link: https://arxiv.org/abs/2209.07590v1
This article examines inter-women's and gender discrimination in order to investigate the relationship between women and men in French audiovisual media. To develop an overlapped speech detection and a gender detection system, we recommend using the WavLM model, which has the benefit of being pre-trained on a large number of speech samples. This research provides new perspectives for human science researchers investigating the differences of representation of women and men in French media.
Source link: https://arxiv.org/abs/2209.04167v1
In the applications that use these models, the gender bias present in the findings on which language models are pre-trained is captured. Bilingual detection in English is not widely adopted in other gendered languages, where the syntax and semantics differ. We investigate gender stereotypes in occupations in Hindi language models in our paper. Our main contributions in this paper include the establishment of a novel corpus to investigate occupational gender bias in Hindi, quantifying this existing bias in these environments using a well-defined metric, and minimizing it by precisely fine-tuning our model.
Source link: https://arxiv.org/abs/2209.03661v1
Staff shortage has recently emerged as a hot topic in information technology. One way to increase the workforce supply is to increase the number of female IT professionals. This is because of a gender imbalance in the information technology industry. To achieve this, it is vital to reduce the impact of biases, such as the belief that men are more suitable for science and technology than women, as well as the belief that women are more suited for careers in science and technology than women, as well as the assumption that female professionals are more suited for careers in science and technology than women, as well as the belief that female professionals are better prepared for female careers. We investigated the connection between gender and code reading speeds in the field of software engineering to help minimize gender bias's influence. Certain source codes require developers to properly understand them, such as those that often change values frequently.
Source link: https://arxiv.org/abs/2209.03516v1
Despite attempts to raise gender parity in politics, global attempts to ensure equal female representation have failed to ensure equal female representation. This is likely to implicit gender stereotypes against women in authority. We present a comprehensive analysis of gender biases that appear in online political discourse in this essay. We get ten million comments on Reddit in discussions of male and female politicians, which allows for an exhaustive review of automatic gender bias detection. We also conducted a multi-faceted research of gender bias against politicians looking at both linguistic and extra-linguistic cues. We examine five specific aspects of gender biases by examining reporting, combinatorial, nominal, sentimental, and lexical errors that are extant in social media language and discourse. Female politicians are often identified by their first names and are described in connection with their body, clothing, or family; this is a condition that is not readily available to men.
Source link: https://arxiv.org/abs/2112.12014v2
Identifying the factors that influence labor force participation could help people arrive at their labor supply decisions, which is, in turn, of utmost importance in determining how the labour market functions. This paper explores the effect of parenthood status on Labor Force Participation decisions made by an individual-level fixed-effects identification scheme. The relationship between individuals' LFP behavior and parenthood will be determined by the differences between individuals and over time, as well as being or not participating in the labour force, which will lead to the confusion needed to determine the relationship between individuals' LFP participation and parenthood. Parents have different impacts on mothers than fathers, as opposed to fathers. This paper estimates LFP equations using individual-level micro data considering the nature of the LFP variable. Parenthood has a negative overall effect on LFP, according to the study's findings.
Source link: https://arxiv.org/abs/2209.02743v1
We use Natural Language Processing methods to analyze 377808 English song lyrics from the "Two Million Song Database" corpus, focusing on gender identity over five decades and gender bias measurement. We find sexist lyrics on a larger scale than previous studies that used small samples of manually annotated popular songs using a sexism classifier. Songs have also been shown to have different language biases based on the gender of the performer, with male solo artist songs containing more and more biases.
Source link: https://arxiv.org/abs/2208.02052v2
Although the gender-based differences in programming courses have been documented before, for example, showing that girls may be safer in same-sex groups, less is known about gender-specific differences in online programming programs. In order to determine whether gender-specific differences can be detected in online courses, we delivered an online introduction to Scratch, in which we examined participant's gender-specific characteristics, how they communicate, their passion, the code they produce, and the creativity displayed by their services in order to investigate whether gender-specific differences can be observed in online courses. In order to determine whether gender-specific differences can be discovered in online courses, we conducted an online introductory programming course, which tested the participants's online introductory programming course coding. Overall, we found no significant differences between girls who participated in all-female vs. mixed groups, and girls were more engaged with the course than boys, with no significant differences between how girls operated in all-female vs. mixed groups. However, when promoting creative freedom in programming, girls and boys appear to fall back to socially inherited stereotypical conduct even in an online setting, influencing the selection of programming terms used.
Source link: https://arxiv.org/abs/2208.13447v1
* Please keep in mind that all text is summarized by machine, we do not bear any responsibility, and you should always check original source before taking any actions