Evolutionary clustering for categorical data using parametric links among multinomial mixture models

Md Abul Hasnat (1) , Julien Velcin (1, 2) , Stephane Bonnevay (1) , Julien Jacques (3, 1, 4, 2)

1 ERIC - Equipe de Recherche en Ingénierie des Connaissances
2 ERIC - Entrepôts, Représentation et Ingénierie des Connaissances
3 MODAL - MOdel for Data Analysis and Learning
4 LPP - Laboratoire Paul Painlevé - UMR 8524

In this paper, we propose a novel evolutionary clustering method for temporal categorical data based on parametric links among multinomial mixture models. Besides clustering, our main goal is to interpret the evolutions of clusters over time. To this aim, first we propose the formulation of a generalized model that establishes parametric links among two multinomial mixture. Afterward, different parametric sub-models are defined in order to model typical evolutions of the clustering structure. Model selection criteria allow to select the best sub-models and thus to guess the clustering evolution. For the experiments, first we evaluate the proposed method with synthetic temporal data. Next, we apply it to analyze the annotated social media data. Results show that the proposed method is better than the state-of-the-art based on the common evaluation metrics. Additionally, it can provide interpretation about the temporal evolution of the clusters.

Mots clés: Twitter data; evolutionary clustering; multinomial distribution; model-based clustering; mixture model

Domaines: Statistiques [math.ST]; Web; Traitement du texte et du document; Intelligence artificielle [cs.AI]; Recherche d'information [cs.IR]

Md Abul Hasnat, Julien Velcin, Stephane Bonnevay, Julien Jacques. Evolutionary clustering for categorical data using parametric links among multinomial mixture models. Econometrics and Statistics , 2017, 3, pp.141-159. ⟨10.1016/j.ecosta.2017.03.004⟩. ⟨hal-01204613v3⟩