Understanding Social Media Texts with Minimum Human Effort on #Twitter - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2016

Understanding Social Media Texts with Minimum Human Effort on #Twitter

Résumé

Named Entity Recognition (NER) is a traditional Natural Language Processing (NLP) task. But traditional machine learning methods are facing new problems to handle this task with Social Media data like Twitter. In this new context, the performance is often degraded. The Twitter messages have particular features. Consider the example "Today wasz Fun cusz anna Came juss for me <3: hahaha". In this example, the difficulties are manifold: 1) Spelling mistakes: wasz (was), cusz (because), juss (just); 2) Uppercase/lowercase inversion: Fun (fun), 3) anna (Anna), Came (came); 4) Emoticon: <3; 5) Interjection: hahaha. The alternation of uppercase/lowercase is a major problem for the NER task because the only person proper noun "anna" of our tweet begins with a lowercase instead of an uppercase, like in grammatically well-formed texts. In this paper, we present our work on recognizing named entities on Twitter.
Fichier principal
Vignette du fichier
PLIN2016.pdf (191.29 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-01490018 , version 1 (14-03-2017)

Identifiants

  • HAL Id : hal-01490018 , version 1

Citer

Tian Tian, Isabelle Tellier, Marco Dinarelli, Pedro Cardoso. Understanding Social Media Texts with Minimum Human Effort on #Twitter. Language and the new (instant) media (PLIN), May 2016, Louvain-la-Neuve, Belgium. ⟨hal-01490018⟩
173 Consultations
78 Téléchargements

Partager

Gmail Facebook X LinkedIn More