From text saliency to linguistic objects: learning linguistic interpretable markers with a multi-channels convolutional architecture
A lot of effort is currently made to provide methods to analyze and understand deep neural network impressive performances for tasks such as image or text classification. These methods are mainly based on visualizing the important input features taken into account by the network to build a decision....
Αποθηκεύτηκε σε:
| Τόπος έκδοσης: | URI:https://journals.openedition.org/corpus, |
|---|---|
| Κύριοι συγγραφείς: | , , , |
| Μορφή: | Article ou chapitre numérique |
| Γλώσσα: | Anglais |
| Έκδοση: |
Corpus
2023
|
| Θέματα: | |
| Διαθέσιμο Online: | Accès Université d'Orléans et IFPM Accès Université d'Orléans et IFPM |
| id |
openedition-oai:revues.org:corpus_7667 |
|---|---|
| record_format |
openedition |
| spelling |
openedition-oai:revues.org:corpus_7667 https://doi.org/10.4000/corpus.7667 urn:doi:10.4000/corpus.7667 en https://journals.openedition.org/corpus/7667 2023-02-07T01:00:00Z info:eu-repo/semantics/openAccess article info:eu-repo/semantics/article urn:handle:20.500.13089/efw4 All rights reserved From text saliency to linguistic objects: learning linguistic interpretable markers with a multi-channels convolutional architecture Corpus Bases, corpus, langage - UMR 7320 info:eu-repo/semantics/reference/issn/1765-3126 info:eu-repo/semantics/reference/issn/1638-9808 URI:https://journals.openedition.org/corpus Vanni, Laurent Corneli, Marco Mayaffre, Damon Precioso, Frédéric texte cooccurrence motifs deep learning apprentissage profond convolution déconvolution cooccurrence deep learning convolution deconvolution texte A lot of effort is currently made to provide methods to analyze and understand deep neural network impressive performances for tasks such as image or text classification. These methods are mainly based on visualizing the important input features taken into account by the network to build a decision. However these techniques, let us cite LIME, SHAP, Grad-CAM, or TDS, require extra effort to interpret the visualization with respect to expert knowledge. In this paper, we propose a novel approach to inspect the hidden layers of a fitted CNN in order to extract interpretable linguistic objects from texts exploiting classification process. In particular, we detail a weighted extension of the Text Deconvolution Saliency (wTDS) measure which can be used to highlight the relevant features used by the CNN to perform the classification task. We empirically demonstrate the efficiency of our approach on corpora from two different languages: English and French. On all datasets, wTDS automatically encodes complex linguistic objects based on co-occurrences and possibly on grammatical and syntax analysis. De nombreux efforts sont actuellement déployés pour fournir des méthodes d'analyse et de compréhension des performances remarquables des réseaux neuronaux profonds pour des tâches telle que la classification d'images ou de textes. Ces méthodes sont principalement fondées sur la visualisation des imputs - ici, la matière textuelle - prises en compte par le réseau pour construire une décision. Cependant, ces techniques que l'on retrouve dans LIME, SHAP, Grad-CAM ou TDS, nécessitent un effort supplémentaire pour interpréter la visualisation. Dans cet article, nous proposons une nouvelle approche pour inspecter les couches cachées d'un réseau CNN ajusté afin d'extraire des objets linguistiques directement interprétables des textes. En particulier, nous détaillons une extension pondérée de la mesure Text Deconvolution Saliency (wTDS). Nous démontrons empiriquement l'efficacité de notre approche sur des corpus de deux langues différentes : anglais et français. Sur tous les jeux de données, wTDS encode automatiquement des objets linguistiques complexes ou motifs basés sur les co-occurrences et éventuellement sur l'analyse grammaticale et syntaxique. |
| collection |
OpenEdition Journal |
| language |
Anglais |
| format |
Article ou chapitre numérique |
| building |
0/Bibliothèque numérique/ 1/Bibliothèque numérique/OpenEdition Journals/ |
| topic |
texte cooccurrence motifs deep learning apprentissage profond convolution déconvolution cooccurrence deep learning convolution deconvolution texte |
| spellingShingle |
texte cooccurrence motifs deep learning apprentissage profond convolution déconvolution cooccurrence deep learning convolution deconvolution texte Vanni, Laurent Corneli, Marco Mayaffre, Damon Precioso, Frédéric From text saliency to linguistic objects: learning linguistic interpretable markers with a multi-channels convolutional architecture |
| description |
A lot of effort is currently made to provide methods to analyze and understand deep neural network impressive performances for tasks such as image or text classification. These methods are mainly based on visualizing the important input features taken into account by the network to build a decision. However these techniques, let us cite LIME, SHAP, Grad-CAM, or TDS, require extra effort to interpret the visualization with respect to expert knowledge. In this paper, we propose a novel approach to inspect the hidden layers of a fitted CNN in order to extract interpretable linguistic objects from texts exploiting classification process. In particular, we detail a weighted extension of the Text Deconvolution Saliency (wTDS) measure which can be used to highlight the relevant features used by the CNN to perform the classification task. We empirically demonstrate the efficiency of our approach on corpora from two different languages: English and French. On all datasets, wTDS automatically encodes complex linguistic objects based on co-occurrences and possibly on grammatical and syntax analysis. |
| author |
Vanni, Laurent Corneli, Marco Mayaffre, Damon Precioso, Frédéric |
| author_facet |
Vanni, Laurent Corneli, Marco Mayaffre, Damon Precioso, Frédéric |
| author_sort |
Vanni, Laurent |
| title |
From text saliency to linguistic objects: learning linguistic interpretable markers with a multi-channels convolutional architecture |
| title_short |
From text saliency to linguistic objects: learning linguistic interpretable markers with a multi-channels convolutional architecture |
| title_full |
From text saliency to linguistic objects: learning linguistic interpretable markers with a multi-channels convolutional architecture |
| title_fullStr |
From text saliency to linguistic objects: learning linguistic interpretable markers with a multi-channels convolutional architecture |
| title_full_unstemmed |
From text saliency to linguistic objects: learning linguistic interpretable markers with a multi-channels convolutional architecture |
| title_sort |
from text saliency to linguistic objects: learning linguistic interpretable markers with a multi-channels convolutional architecture |
| publisher |
Corpus |
| publishDate |
2023 |
| container_title |
URI:https://journals.openedition.org/corpus |
| url |
https://ezproxy.univ-orleans.fr/login?url=https://doi.org/10.4000/corpus.7667 https://ezproxy.univ-orleans.fr/login?url=https://journals.openedition.org/corpus/7667 |
| _version_ |
1877292901002641408 |