Skip to Main content Skip to Navigation
Master Thesis

Une exploration de l'architecture des réseaux de neurones pour la modélisation de la compositionnalité sémantique

Abstract : This dissertation presents an evaluation of a neural network model called autoencoder in order to capture the meaning of adjective-noun couples in English. This model works on the representation of words meaning by a vector countaining the index of the words' context lemmas. These indexes are assigned to lemmas according to their frenquency in our Wikipedia-extracted corpus. Our model is evaluated on similarity task between two adjective-noun couples, and then on a task of recomposition of adjective-noun couples vector from their separated components context vectors. These two task results were eventually compared to the following already existing models : the vectors sum (additive model), the weighted additive with a stronger rating on the noun vector, the baseline model where only the nouns vector is taken, and the multiplicative model (multiplication of vectors).
Document type :
Master Thesis
Complete list of metadatas

Cited literature [25 references]  Display  Hide  Download

https://dumas.ccsd.cnrs.fr/dumas-01212786
Contributor : Uga - Bulles <>
Submitted on : Wednesday, October 7, 2015 - 11:36:05 AM
Last modification on : Wednesday, July 15, 2020 - 1:06:02 PM
Long-term archiving on: : Friday, January 8, 2016 - 10:22:49 AM

Identifiers

  • HAL Id : dumas-01212786, version 1

Citation

Chloé Cimpello. Une exploration de l'architecture des réseaux de neurones pour la modélisation de la compositionnalité sémantique. Sciences de l'Homme et Société. 2015. ⟨dumas-01212786⟩

Share

Metrics

Record views

398

Files downloads

745