Accéder directement au contenu Accéder directement à la navigation
Communication dans un congrès

Managing Genetic Algorithm Parameters to Improve SegGen - A Thematic Segmentation Algorithm

Abstract :

SegGen [1] is a linear thematic segmentation algorithm grounded on a variant of the Strength Pareto Evolutionary Algorithm [2] and aims at optimizing the two criteria of the Salton's [3] definition of segments: a segment is a part of text whose internal cohesion and dissimilarity with its adjacent segments are maximal. This paper describes improvements that have been implemented in the approach taken by SegGen by tuning the genetic algorithm parameters according with the evolution of the quality of the generated populations. Two kinds of reasons originate the tuning of the parameters and have been implemented here. First as it could be measured by the values of global criteria of the population quality, the global quality of the generated populations increases as the process goes and it seems reasonable to set values to parameters and define new operators, which favor intensification and diminish diversification factors in the search process. Second since individuals in the populations are plausible segmentations it seems reasonable to weight sentences in the current segmentation depending on their distance to the boundaries of the segment they belong to for the calculus of similarities between sentences implied in the two criteria to be optimized. Although this tuning of the parameters of the algorithm currently rests on estimations based on experiments, first results are promising.

Type de document :
Communication dans un congrès
Liste complète des métadonnées

https://hal.univ-angers.fr/hal-03256756
Contributeur : Okina Université d'Angers <>
Soumis le : jeudi 10 juin 2021 - 14:16:52
Dernière modification le : vendredi 11 juin 2021 - 03:29:58

Identifiants

  • HAL Id : hal-03256756, version 1
  • OKINA : ua13593

Collections

Citation

Sirin Saygili, Tankut Acarman, Tassadit Amghar, Bernard Levrat. Managing Genetic Algorithm Parameters to Improve SegGen - A Thematic Segmentation Algorithm. 24th International Workshop on Database and Expert Systems Applications (DEXA), 2013, Los Alamitos, United States. pp.58-62. ⟨hal-03256756⟩

Partager

Métriques

Consultations de la notice

14