Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classification

Borkan, Daniel; Dixon, Lucas; Sorensen, Jeffrey; Thain, Nithum; Vasserman, Lucy

doi:10.48550/arxiv.1903.04561

Cited by 6 publications

(11 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The toxic class label contains several subtypes of toxic comments such as identity attacks, insults, explicit sexuality, obscenity, insult, and threats. An LSTM model applied to a civil comments dataset [6] has been used. The LSTM model is composed of an embedding 300-dimensional layer, two bidirectional LSTM layers (with 256 units for each direction), and finally, a dense layer with 128 hidden units.…”

Section: Use Casementioning

confidence: 99%

“…In this section, we evaluate the ability of T-EBAnO to adapt to different architectures and different tasks (use cases [3][4][5][6][7][8]. For this purpose, we defined the following additional tasks.…”

Section: Framework Extendibilitymentioning

confidence: 99%

“…Fig 6. Examples of textual explanation report for the input in Fig.6aoriginally labeled by custom LSTM model as Toxic with a probability of 0.98.…”

mentioning

confidence: 99%

“…The LIME parameter number of permutations has been set to 5000 5. The Partition Explainer version of SHAP has been used 6. https://huggingface.co/.…”

mentioning

confidence: 99%

See 3 more Smart Citations

Trusting deep learning natural-language models via local and global explanations

et al. 2022

View full text Add to dashboard Cite

Despite the high accuracy offered by state-of-the-art deep natural-language models (e.g., LSTM, BERT), their application in real-life settings is still widely limited, as they behave like a black-box to the end-user. Hence, explainability is rapidly becoming a fundamental requirement of future-generation data-driven systems based on deep-learning approaches. Several attempts to fulfill the existing gap between accuracy and interpretability have been made. However, robust and specialized eXplainable Artificial Intelligence solutions, tailored to deep natural-language models, are still missing. We propose a new framework, named T-EBAnO, which provides innovative prediction-local and class-based model-global explanation strategies tailored to deep learning natural-language models. Given a deep NLP model and the textual input data, T-EBAnO provides an objective, human-readable, domain-specific assessment of the reasons behind the automatic decision-making process. Specifically, the framework extracts sets of interpretable features mining the inner knowledge of the model. Then, it quantifies the influence of each feature during the prediction process by exploiting the normalized Perturbation Influence Relation index at the local level and the novel Global Absolute Influence and Global Relative Influence indexes at the global level. The effectiveness and the quality of the local and global explanations obtained with T-EBAnO are proved on an extensive set of experiments addressing different tasks, such as a sentiment-analysis task performed by a fine-tuned BERT model and a toxic-comment classification task performed by an LSTM model. The quality of the explanations proposed by T-EBAnO, and, specifically, the correlation between the influence index and human judgment, has been evaluated by humans in a survey with more than 4000 judgments. To prove the generality of T-EBAnO and its model/task-independent methodology, experiments with other models (ALBERT, ULMFit) on popular public datasets (Ag News and Cola) are also discussed in detail.

show abstract

Section: Use Casementioning

confidence: 99%

Section: Framework Extendibilitymentioning

confidence: 99%

See 2 more Smart Citations

Trusting deep learning natural-language models via local and global explanations

et al. 2022

View full text Add to dashboard Cite

show abstract

“…The first task is a binary toxic comment classification, and it consists of predicting whether the input comment is clean or toxic, i.e., it contains inappropriate content such as obscenity, threat, insult, identity attack, and explicit sexual content. An LSTM model applied to a civil comments dataset [33] has been used. The toxic class label contains several subtypes of toxic comments as identity attacks, insults, explicit sexuality, and threats.…”

Section: Use Casesmentioning

confidence: 99%

Explaining the Deep Natural Language Processing by Mining Textual Interpretable Features

Ventura¹,

Greco²,

Apiletti³

et al. 2021

Preprint

View full text Add to dashboard Cite

Despite the high accuracy offered by state-of-the-art deep natural-language models (e.g. LSTM, BERT), their application in reallife settings is still widely limited, as they behave like a black-box to the end-user. Hence, explainability is rapidly becoming a fundamental requirement of future-generation data-driven systems based on deep-learning approaches. Several attempts to fulfill the existing gap between accuracy and interpretability have been done. However, robust and specialized xAI (Explainable Artificial Intelligence) solutions tailored to deep natural-language models are still missing. We propose a new framework, named T-EBAnO, which provides innovative prediction-local and class-based model-global explanation strategies tailored to black-box deep natural-language models. Given a deep NLP model and the textual input data, T-EBAnO provides an objective, human-readable, domain-specific assessment of the reasons behind the automatic decision-making process. Specifically, the framework extracts sets of interpretable features mining the inner knowledge of the model. Then, it quantifies the influence of each feature during the prediction process by exploiting the novel normalized Perturbation Influence Relation index at the local level and the novel Global Absolute Influence and Global Relative Influence indexes at the global level. The effectiveness and the quality of the local and global explanations obtained with T-EBAnO are proved on (i) a sentiment analysis task performed by a fine-tuned BERT model, and (ii) a toxic comment classification task performed by an LSTM model.

show abstract

Hierarchical multi-view metric learning with HSIC regularization

et al. 2022

View full text Add to dashboard Cite

Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classification

Cited by 6 publications

References 0 publications

Trusting deep learning natural-language models via local and global explanations

Trusting deep learning natural-language models via local and global explanations

Explaining the Deep Natural Language Processing by Mining Textual Interpretable Features

Hierarchical multi-view metric learning with HSIC regularization

Contact Info

Product

Resources

About