Indexing by Latent Dirichlet Allocation and an Ensemble Model

The contribution of this paper is two-fold. First, we present indexing by Latent Dirichlet Allocation (LDI), an automatic document indexing method with a probabilistic concept search. The probability distributions in LDI utilizes those in Latent Dirichlet Allocation (LDA), which is a generative topic model that has been previously used in applications for document indexing tasks. However, those ad hoc applications, or their variants with smoothing techniques as prompted by previous studies in LDA-based language modeling, would result in unsatisfactory performance as the terms in documents may not properly reflect concept space. To improve the performances, we introduce a new definition of document probability vectors in the context of LDA and present a novel scheme for automatic document indexing based on it. Second, we propose an ensemble model (EnM) for document indexing. The EnM combines basis indexing models by assigning different weights and tries to uncover the optimal weights with which the mean average precision (MAP) is maximized. To solve the optimization problem, we propose three algorithms, EnM.B, EnM,CD and EnM.PCD. EnM.B is derived based on the boosting method, EnM.CD the coordinate descent method, and EnM.PCD the parallel property of the EnM.CD. The results of our computational experiment on a benchmark data set indicate that both the proposed approaches are viable options in the document indexing tasks. c © 2013 Published by Elsevier Ltd.