Transparency and accountability in AI decision support: Explaining and visualizing convolutional neural networks for text information

Kim, Buomsoo; Park, Jinsoo; Suh, Jihae

doi:10.1016/j.dss.2020.113302

Cited by 105 publications

(40 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Several methods based on statistical machine learning (e.g., naive Bayes, decision trees, hidden Markov models, maximum entropy, and look‐up tables) and deep learning (e.g., convolutional neural network and recurrent neural network) algorithms can be employed to enhance the accuracy and efficiency of syntactic and semantic analysis of sentences (Kim, Park, & Suh, 2020; Laylavi, Rajabifard, & Kalantari, 2017; Luo, 2021; Luo & Chen, 2020; Zhang et al., 2011). The main aim of our study was to investigate the potential of UGC for DHM.…”

Section: Methodsmentioning

confidence: 99%

Integrating user‐generated content to track urban flooding hotspots and foster emergency management: A case study in central China

Zeng

Lan

Jiang

et al. 2021

Transactions in GIS

View full text Add to dashboard Cite

This study explores the potential for employing user‐generated content (UGC) during severe flooding to discover and track urban flooding disaster hotspots in a timely manner. A flooding case in central China was selected for this study. Crawlers, natural language processing, and geographic visualization methods were used to extract flood‐related UGC specific to severely flooded areas. Further, the quality of the geo‐tagged content on the web was verified using scientific gauge data (e.g., water level, digital elevation model, and rainfall). Based on our findings, we were able to deduce that UGC on the web is valuable for identifying flooding hotspots. In fact, this approach offers substantial advantages for addressing the emerging needs of data acquisition for flood emergency management. Notably, the values of urban stakeholders that share their observations on the web can be mined rapidly through the integration of various approaches. Overall, the findings of our study can contribute to the efficient mining of web data sources and development of disaster hotspot mapping systems at an urban scale to improve flood management and mitigation. The limitations of UGC and our study's future direction are also discussed.

show abstract

Section: Methodsmentioning

confidence: 99%

Integrating user‐generated content to track urban flooding hotspots and foster emergency management: A case study in central China

Zeng

Lan

Jiang

et al. 2021

Transactions in GIS

View full text Add to dashboard Cite

show abstract

“…Additionally, the improved explainability can lead to a greater support for the work with such systems [6]. Regarding decision support, explainability can also lead to enhanced fairness [32]. To increase the explainability of the system, different techniques and visualizations can be utilized and combined [5][6][7].…”

Section: Definition Of Design Principles and Featuresmentioning

confidence: 99%

“…(DF 2) Provide the confidence for the present classification. The probability of the classification will be represented as the confidence of the AI system which is also an eponymous goal of XAI [32]. This is represented by the probability value of the classification in percentage.…”

Section: Definition Of Design Principles and Featuresmentioning

confidence: 99%

AI-Assisted and Explainable Hate Speech Detection for Social Media Moderators – A Design Science Approach

Bunde¹

2021

Proceedings of the Annual Hawaii International Conference on System Sciences

View full text Add to dashboard Cite

To date, the detection of hate speech is still primarily carried out by humans, yet there is great potential for combining human expertise with automated approaches. However, identified challenges include low levels of agreement between humans and machines due to the algorithms' missing expertise of, e.g., cultural, and social structures. In this work, a design science approach is used to derive design knowledge and develop an artifact, through which humans are integrated in the process of detecting and evaluating hate speech. For this purpose, explainable artificial intelligence (XAI) is utilized: the artifact will provide explanative information, why the deep learning model predicted whether a text contains hate. Results show that the instantiated design knowledge in form of a dashboard is perceived as valuable and that XAI features increase the perception of the artifact's usefulness, ease of use, trustworthiness as well as the intention to use it.

show abstract

“…Imaging is one of the emerging fields in DL, the majority of works tried to explain imaging systems from specific problems [ 70 , 71 ]. However, language processing accompanied with the availability of large text dataset became centre of interest to many researchers, one remarkable work was done by [ 72 ] for huge text corpus explanation; although the imaging system is more clarified and flexible, the way the graph was generated doesn’t benefit from graph-based technologies that optimize the input starting from naive generation.…”

Section: Literature Reviewmentioning

confidence: 99%

“…Example-based approaches ; research in this area is always conducted through a training-example, by specifying some initial observations which will be verified through features’ extraction, this discipline is widely adapted despite the difficulty of verifying the trustworthiness of each example, this covers: ✓ Gradient methods (e.g., Guided-back propagation, Layer-wise relevance propagation [ 72 ]), which aim to a better gradient optimization. ✓ Saliency-feature map [ 73 ] for measuring pattern importance within images and videos.…”

Section: Literature Reviewmentioning

confidence: 99%

An investigation into the deep learning approach in sentimental analysis using graph-based theories

Kentour

2021

PLoS ONE

View full text Add to dashboard Cite

Sentiment analysis is a branch of natural language analytics that aims to correlate what is expressed which comes normally within unstructured format with what is believed and learnt. Several attempts have tried to address this gap (i.e., Naive Bayes, RNN, LSTM, word embedding, etc.), even though the deep learning models achieved high performance, their generative process remains a “black-box” and not fully disclosed due to the high dimensional feature and the non-deterministic weights assignment. Meanwhile, graphs are becoming more popular when modeling complex systems while being traceable and understood. Here, we reveal that a good trade-off transparency and efficiency could be achieved with a Deep Neural Network by exploring the Credit Assignment Paths theory. To this end, we propose a novel algorithm which alleviates the features’ extraction mechanism and attributes an importance level of selected neurons by applying a deterministic edge/node embeddings with attention scores on the input unit and backward path respectively. We experiment on the Twitter Health News dataset were the model has been extended to approach different approximations (tweet/aspect and tweets’ source levels, frequency, polarity/subjectivity), it was also transparent and traceable. Moreover, results of comparing with four recent models on same data corpus for tweets analysis showed a rapid convergence with an overall accuracy of ≈83% and 94% of correctly identified true positive sentiments. Therefore, weights can be ideally assigned to specific active features by following the proposed method. As opposite to other compared works, the inferred features are conditioned through the users’ preferences (i.e., frequency degree) and via the activation’s derivatives (i.e., reject feature if not scored). Future direction will address the inductive aspect of graph embeddings to include dynamic graph structures and expand the model resiliency by considering other datasets like SemEval task7, covid-19 tweets, etc.

show abstract

Transparency and accountability in AI decision support: Explaining and visualizing convolutional neural networks for text information

Cited by 105 publications

References 22 publications

Integrating user‐generated content to track urban flooding hotspots and foster emergency management: A case study in central China

Integrating user‐generated content to track urban flooding hotspots and foster emergency management: A case study in central China

AI-Assisted and Explainable Hate Speech Detection for Social Media Moderators – A Design Science Approach

An investigation into the deep learning approach in sentimental analysis using graph-based theories

Contact Info

Product

Resources

About