Skip to Main content Skip to Navigation
Conference papers

Early Detection of Depression and Anorexia from Social Media: A Machine Learning Approach

Faneva Ramiandrisoa 1 Josiane Mothe 1
1 IRIT-SIG - Systèmes d’Informations Généralisées
IRIT - Institut de recherche en informatique de Toulouse
Abstract : In this paper, we present an approach on social media mining to help early detection of two mental illnesses: depression and anorexia. We aim at detecting users that are likely to be ill, by learning from annotated examples. We mine texts to extract features for text representation and also use word embedding representation. The machine learning based model we proposed uses these two types of text representation to predict the likelihood of each user to be ill. We use 58 features from state of the art and 198 features new in this domain that are part of our contribution. We evaluate our model on the CLEF eRisk 2018 reference collections. For depression detection, our model based on word embedding achieves the best performance according to the measure ERDE 50 and the model based on features only achieves the best performance according to precision. For anorexia detection, the model based on word embedding achieves the second-best results on ERDE 50 and recall. We also observed that many of the new features we added contribute to improve the results.
Complete list of metadatas

Cited literature [25 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-02877723
Contributor : Josiane Mothe <>
Submitted on : Monday, June 22, 2020 - 4:08:43 PM
Last modification on : Friday, July 3, 2020 - 3:00:12 AM

File

CIRCLE20_13.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02877723, version 1

Citation

Faneva Ramiandrisoa, Josiane Mothe. Early Detection of Depression and Anorexia from Social Media: A Machine Learning Approach. Circle 2020, Iván Cantador; Max Chevalier; Massimo Melucci; Josiane Mothe, Jul 2020, Samatan, France. ⟨hal-02877723⟩

Share

Metrics

Record views

33

Files downloads

29