Interpretable Ranking with Generalized Additive Models

Zhuang, Honglei; Wang, Xuanhui; Bendersky, Michael; Grushetsky, Alexander; Wu, Yonghui; Mitrichev, Petr; Sterling, Ethan; Bell, Nancy; Ravina, Walker; Qian, Hai

doi:10.1145/3437963.3441796

Cited by 23 publications

(13 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Thus, webpage and document ranking methods are being used to sort and recommend corresponding information for any web inquiries effectively [21]- [23]. Few of recent work employed advanced ML techniques such as Generalized Additive Models to provide global explanations for rankings [24]. Such mechanisms are considered non-competitive ranking since web pages or documents are not necessarily in direct competition to boost their rankings for cyclic evaluations.…”

Section: A Non-competitive Rankingmentioning

confidence: 99%

Local Explanations of Global Rankings: Insights for Competitive Rankings

Anahideh

Mohabbati-Kalejahi

2022

IEEE Access

View full text Add to dashboard Cite

Explaining complex algorithms and models has recently received growing attention in various domains to support informed decisions. Ranking functions are widely used for almost every form of human activity to enable effective decision-making processes. Hence, explaining ranking indicators and their importance are essential properties to enhance performance. Local explanation techniques have recently become a prominent way to interpret individual predictions of machine learning models. However, there has been limited investigation into explaining competitive rankings. This work proposes a hierarchical ranking explanation framework to capture local explanations for competitive rankings by defining a proper neighborhood construction approach. We explore various explanation techniques to identify the local contribution of ranking indicators based on the position of an instance in the ranking as well as the size of the neighborhood around the instance of interest. We evaluate the generated explanations for the Times Higher Education university ranking dataset as a benchmark of competitive ranking. The results reveal insights for a wide range of instances in the ranking list and indicate the importance of local explanations for competitive rankings.

show abstract

Section: A Non-competitive Rankingmentioning

confidence: 99%

Local Explanations of Global Rankings: Insights for Competitive Rankings

Anahideh

Mohabbati-Kalejahi

2022

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Intrinsically interpretable models , such as linear regression, logistic regression, decision trees, general additive models, or combinations of business decision rules, are characterized by their transparency and by a self-explainable structure. They are generally applied for use cases with legal or policy constraints (Zhuang et al, 2020), but they may well be not accurate enough for tasks such as fraud detection, which have high financial stakes. This explains why more accurate black box models look appealing as soon as a post hoc interpretability method is applied to provide explanations on either how they work or on their results.…”

Section: Sota Review Of Fraud Detectionmentioning

confidence: 99%

The accuracy versus interpretability trade-off in fraud detection model

Nesvijevskaia

Ouillade²,

Guilmin³

et al. 2021

Data & Policy

View full text Add to dashboard Cite

Like a hydra, fraudsters adapt and circumvent increasingly sophisticated barriers erected by public or private institutions. Among these institutions, banks must quickly take measures to avoid losses while guaranteeing the satisfaction of law-abiding customers. Facing an expanding flow of operations, effective banking relies on data analytics to support established risk control processes, but also on a better understanding of the underlying fraud mechanism. In addition, fraud being a criminal offence, the evidential aspect of the process must also be considered. These legal, operational, and strategic constraints lead to compromises on the means to be implemented for fraud management. This paper first focuses on the translation of practical questions raised in the banking industry at each step of the fraud management process into performance evaluation required to design a fraud detection model. Secondly, it considers a range of machine learning approaches that address these specificities: the imbalance between fraudulent and nonfraudulent operations, the lack of fully trusted labels, the concept-drift phenomenon, and the unavoidable trade-off between accuracy and interpretability of detection. This state-of-the-art review sheds some light on a technology race between black box machine learning models improved by post-hoc interpretation and intrinsic interpretable models boosted to gain accuracy. Finally, it discusses how concrete and promising hybrid approaches can provide pragmatic, short-term answers to banks and policy makers without swallowing up stakeholders with economical and ethical stakes in this technological race.

show abstract

“…𝑥 𝑖 ) is modulated by a function of another "context" feature (i.e. 𝑐 𝑖 ) [31]. In other cases, the interaction amongst features is adequately approximated by additive models of univariate functions nested within univariate functions, 𝑓 (𝑥 1 , 𝑥 2 , 𝑥 3 ) ≈ 𝑔 1 (𝑓 1,1 (𝑥 1 ) + 𝑓 1,2 (𝑥 2 ) + 𝑓 1,3 (𝑥 3 )), or 𝑓 (𝑥 1 , 𝑥 2 , 𝑥 3 ) ≈ 𝑔 1 (𝑓 1,1 (𝑥 1 ) + 𝑓 1,2 (𝑥 2 ) + 𝑓 1,3 (𝑥 3 )) + 𝑔 2 (𝑓 2,1 (𝑥 1 ) + 𝑓 2,2 (𝑥 2 ) + 𝑓 2,3 (𝑥 3 )) + .…”

Section: Introductionmentioning

confidence: 99%

“…For example, the proposed distillation methodology can be applied to additive models trained using bagged boosted decision trees, as demonstrated in our results. Similarly, the technique can be applied to additive neural nets [2], as demonstrated by Zhuang et al [31]. Listing 1.1 and Figure 1.1 show textual and graphical representations of a model obtained by applying our approach to a decision forest GAM learned from the COMPAS dataset.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Distilling Interpretable Models into Human-Readable Code

Ravina¹,

Sterling²,

Oryeshko³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

The goal of model distillation is to faithfully transfer teacher model knowledge to a model which is faster, more generalizable, more interpretable, or possesses other desirable characteristics. Humanreadability is an important and desirable standard for machinelearned model interpretability. Readable models are transparent and can be reviewed, manipulated, and deployed like traditional source code. As a result, such models can be improved outside the context of machine learning and manually edited if desired. Given that directly training such models is difficult, we propose to train interpretable models using conventional methods, and then distill them into concise, human-readable code.The proposed distillation methodology approximates a model's univariate numerical functions with piecewise-linear curves in a localized manner. The resulting curve model representations are accurate, concise, human-readable, and well-regularized by construction. We describe a piecewise-linear curve-fitting algorithm that produces high-quality results efficiently and reliably across a broad range of use cases. We demonstrate the effectiveness of the overall distillation technique and our curve-fitting algorithm using three publicly available datasets COMPAS, FICO, and MSLR-WEB30K. CCS CONCEPTS• Computing methodologies → Machine learning approaches.

show abstract

Interpretable Ranking with Generalized Additive Models

Cited by 23 publications

References 34 publications

Local Explanations of Global Rankings: Insights for Competitive Rankings

Local Explanations of Global Rankings: Insights for Competitive Rankings

The accuracy versus interpretability trade-off in fraud detection model

Distilling Interpretable Models into Human-Readable Code

Contact Info

Product

Resources

About