摘要
人们通常将某一类性别占主导的职业与该性别联系在一起。在算法得到广泛应用的今天,这种刻板印象从社会数据进入算法,并在算法与用户交互的界面被呈现为偏见。本文关注中国本土的搜索引擎,通过搜索获取搜索引擎中不同排序的图片,将搜索引擎图片与人口普查数据中的职业性别比例进行对比,展开算法审计。研究发现,两个主要搜索引擎的图片搜索结果中都存在符合职业性别刻板印象的性别比例夸大现象,且这种夸大现象在女性占比较低的职业中更加明显;同时,二者都存在职业女性整体占比偏低的情况。在搜索结果排序中,有的搜索引擎靠前图片中刻板印象的比例夸张更严重。搜索算法吸收着职业性别刻板印象且直观地呈现在图片搜索结果中,并可能在算法和用户的认知循环之间进一步加深。
People tend to associate certain gender-dominant occupations with specific genders.In today's widespread application of algorithms,these stereotypes transition from social data into algorithms,presenting themselves as distorted biases in the interfaces where algorithms interact with users.This paper focuses on Chinese search engines,conducting algorithm auditing by retrieving images from different ranking positions through image search,and contrasting them with gender proportions from the population census data.The study reveals that both major search engines'image search results exhibit an exaggeration of gender ratio according to occupational gender stereotypes,with this exaggeration being more pronounced in occupations with lower female proportions.Additionally,both platforms display an overall underrepresentation of women.In the ranking of search results,the occupational gender stereotype exaggeration is more severe in images presented in the front positions.Search algorithms absorb occupational gender stereotypes,presenting them in image search results and potentially deepening these stereotypes as they perpetuate in the loop between algorithms and users.
作者
周葆华
罗沛
ZHOU Bao-hua;LUO Pei
出处
《新闻大学》
CSSCI
北大核心
2024年第6期1-17,118,共18页
Journalism Research
基金
国家社科基金人才项目(22VRC186)
教育部重点研究基地重大项目(22JJD860004)
复旦大学文科先导和创新团队项目(IDH3353070)
复旦大学新闻学院科研创新项目(2023—2024)。
关键词
搜索引擎
职业性别
刻板印象
算法偏见
算法审计
search engines
occupational gender
stereotypes
algorithmic bias
algorithm auditing