Simple Framework for Interpretable Fine-grained Text Classification

Munkhtulga Battogtokh; Michael Luck; Cosmin Davidescu; Rita Borgo

doi:10.1007/978-3-031-50396-2_23

Simple Framework for Interpretable Fine-grained Text Classification

Munkhtulga Battogtokh^*, Michael Luck, Cosmin Davidescu, Rita Borgo

^*Corresponding author for this work

ContactEngine Limited

Research output: Chapter in Book/Report/Conference proceeding › Conference paper › peer-review

1 Citation (Scopus)

94 Downloads (Pure)

Abstract

Fine-grained text classification with similar and many labels is a challenge in practical applications. Interpreting predictions in this context is particularly difficult. To address this, we propose a simple framework that disentangles feature importance into more fine-grained links. We demonstrate our framework on the task of intent recognition, which is widely used in real-life applications where trustworthiness is important, for state-of-the-art Transformer language models using their attention mechanism. Our human and semi-automated evaluations show that our approach better explains fine-grained input-label relations than popular feature importance estimation methods LIME and Integrated Gradient and that our approach allows faithful interpretations through simple rules, especially when model confidence is high.

Original language	English
Title of host publication	ECAI 2023
Subtitle of host publication	3rd International Workshop on Explainable and Interpretable Machine Learning (XI-ML)
DOIs	https://doi.org/10.1007/978-3-031-50396-2_23
Publication status	Published - 2023

Access to Document

10.1007/978-3-031-50396-2_23

Battogtokh-et-al-XI-ML-ECAI-2023Accepted author manuscript, 3.64 MBLicence: CC BY-NC

Cite this

@inbook{16c897d15ada41b881571d357098df53,

title = "Simple Framework for Interpretable Fine-grained Text Classification",

abstract = "Fine-grained text classification with similar and many labels is a challenge in practical applications. Interpreting predictions in this context is particularly difficult. To address this, we propose a simple framework that disentangles feature importance into more fine-grained links. We demonstrate our framework on the task of intent recognition, which is widely used in real-life applications where trustworthiness is important, for state-of-the-art Transformer language models using their attention mechanism. Our human and semi-automated evaluations show that our approach better explains fine-grained input-label relations than popular feature importance estimation methods LIME and Integrated Gradient and that our approach allows faithful interpretations through simple rules, especially when model confidence is high.",

author = "Munkhtulga Battogtokh and Michael Luck and Cosmin Davidescu and Rita Borgo",

year = "2023",

doi = "10.1007/978-3-031-50396-2_23",

language = "English",

booktitle = "ECAI 2023",

}

TY - CHAP

T1 - Simple Framework for Interpretable Fine-grained Text Classification

AU - Battogtokh, Munkhtulga

AU - Luck, Michael

AU - Davidescu, Cosmin

AU - Borgo, Rita

PY - 2023

Y1 - 2023

N2 - Fine-grained text classification with similar and many labels is a challenge in practical applications. Interpreting predictions in this context is particularly difficult. To address this, we propose a simple framework that disentangles feature importance into more fine-grained links. We demonstrate our framework on the task of intent recognition, which is widely used in real-life applications where trustworthiness is important, for state-of-the-art Transformer language models using their attention mechanism. Our human and semi-automated evaluations show that our approach better explains fine-grained input-label relations than popular feature importance estimation methods LIME and Integrated Gradient and that our approach allows faithful interpretations through simple rules, especially when model confidence is high.

AB - Fine-grained text classification with similar and many labels is a challenge in practical applications. Interpreting predictions in this context is particularly difficult. To address this, we propose a simple framework that disentangles feature importance into more fine-grained links. We demonstrate our framework on the task of intent recognition, which is widely used in real-life applications where trustworthiness is important, for state-of-the-art Transformer language models using their attention mechanism. Our human and semi-automated evaluations show that our approach better explains fine-grained input-label relations than popular feature importance estimation methods LIME and Integrated Gradient and that our approach allows faithful interpretations through simple rules, especially when model confidence is high.

U2 - 10.1007/978-3-031-50396-2_23

DO - 10.1007/978-3-031-50396-2_23

M3 - Conference paper

BT - ECAI 2023

ER -

Simple Framework for Interpretable Fine-grained Text Classification

Abstract

Access to Document

Fingerprint

Cite this