Automated Essay Scoring in the Presence of Biased Ratings

Abstract

Studies in Social Sciences have revealed that when people evaluate someone else, their evaluations often reflect their biases. As a result, rater bias may introduce highly subjective factors that make their evaluations inaccurate. This may affect automated essay scoring models in many ways, as these models are typically designed to model (potentially biased) essay raters. While there is sizeable literature on rater effects in general settings, it remains unknown how rater bias affects automated essay scoring. To this end, we present a new annotated corpus containing essays and their respective scores. Different from existing corpora, our corpus also contains comments provided by the raters in order to ground their scores. We present features to quantify rater bias based on their comments, and we found that rater bias plays an important role in automated essay scoring. We investigated the extent to which rater bias affects models based on hand-crafted features. Finally, we propose to rectify the training set by removing essays associated with potentially biased scores while learning the scoring model.

Measuring Bias in Contextualized Word Representations

“…The results prove that BERT expresses strong preferences for male pronouns, raising concerns with using BERT in downstream tasks like resume filtering. Table 5: Percentage of attributes associated more strongly with the male gender 6 Related Work NLP applications ranging from core tasks such as coreference resolution (Rudinger et al, 2018) and language identification (Jurgens et al, 2017), to downstream systems such as automated essay scoring (Amorim et al, 2018), exhibit inherent social biases which are attributed to the datasets used to train the embeddings (Barocas and Selbst, 2016;Zhao et al, 2017;Yao and Huang, 2017).…”

Towards context-aware automated writing evaluation systems

“…Amorim et al [4] raise awareness about the readers' biases. In order to examine the extent to which these biases affect the ratings and thus affect the algorithms based on these ratings, they investigate a corpus of scored texts which are commented by raters.…”

An Analysis of Subjectivity in Brazilian News

“…Essa metodologia é usada em alguns trabalhos relacionados [Amorim et al 2018;Sales et al 2019;Moraes et al 2016;Jha et al 2016]. Amorim et al [Amorim et al 2018], por exemplo, utilizam a abordagem de cálculo de subjetividade através de léxicos para avaliar comentários de avaliadores de redações do ENEM (Exame Nacional do Ensino Médio). Para avaliar a subjetividade, a pesquisadora utiliza léxicos de argumentação, sentimento, pressuposição, modalização e valoração.…”

“…Em trabalho prévio Sales et al [Sales et al 2019] propuseram o uso de léxicos de subjetividade para medir, por meio de word embeddings, a subjetividade de textos jornalísticos. E stes léxicos foram construídos por [Amorim et al 2018] através da análise manual de expressões que frequentemente aparecem em textos quando o interlocutor aparenta expressar alguma subjetividade. Cada léxico encapsula um aspecto de subjetividade, mais especificamente esses aspectos são:…”

“…Em trabalho prévio, léxicos de subjetividade construídos manualmente por especialistas linguistas [Amorim et al 2018] foram utilizados para medir, por meio de word embeddings, a subjetividade de textos linguísticos [Sales et al 2019]. Nesse artigo, nós estendemos esse trabalho caracterizando a subjetividade de notícias em cinco portais de notícias populares no Brasil: Estadão, Folha de São Paulo, O Antagonista, O Globo e Veja.…”