Modeling Latent Topics and Temporal Distance for Story Segmentation of Broadcast News

Abstract

This paper studies a strategy to model latent topics and temporal distance of text blocks for story segmentation, that we call Graph Regularization in Topic Modeling or GRTM. We propose two novel approaches that consider both temporal distance and lexical similarity of text blocks, collectively referred to as data proximity, in learning latent topic representation, where a graph regularizer is involved to derive the latent topic representation while preserving data proximity. In the first approach, we extend the idea of Laplacian probabilistic latent semantic analysis (LapPLSA) by introducing a distance penalty function in the affinity matrix of a graph for latent topic estimation. The estimated latent topic distributions are used to replace the traditional term-frequency vectors as the data representation of the text blocks and to measure the cohesive strength between them. In the second approach, we perform Laplacian eigenmaps, which makes use of the graph regularizer for dimensionality reduction, on latent topic distributions estimated by conventional topic modeling. We conduct the experiments on the automatic speech recognition (ASR) transcripts of the TDT2 English broadcast news corpus. The experiments show the proposed strategy outperforms the conventional techniques. LapPLSA performs the best with the highest F1-measure of 0.816. The effects of the penalty constant in the distance penalty function, the number of latent topics, and the size of training data on the segmentation performances are also studied.

Fuzzy-Gibbs latent Dirichlet allocation model for feature extraction on Indonesian documents

“…LSI has been widely discussed in study [8,11,29]. pLSI is an enhancement of LSI which able to model every word as a representation of some topics that could overcome the problem of synonyms and polysemy [4,14,18,31]. However, modeling which was done by pLSI only at document level.…”

Domain-agnostic document and question classification using natural language processing techniques

“…This feature vector is then classified into its respective class label via a machine learning algorithm [10,[15][16][17][18]. Since BoW features are highly sparse and lack diversity [19], topic modeling approaches such as latent Dirichlet allocation (LDA) [20] have been developed. These admixture approaches were originally employed for document classification as they offer linguistic insights into language patterns by grouping associated words into topics and, thereafter, computing the probabilities of topics occurring in each document [21,22].…”

“…The scatter plot in Figure 6.3(c), on the other hand, exhibits a less skewed set of points based on the scaling in (6. 19) and is reflected via the histogram with a lighter-tailed distribution. The above implies that moderate emphasis is given to less significant topic probabilities which is subsequently shown to result in better feature representation for classification.…”

Knowledge graph population from news streams

“…SPIGA [83] represents documents by doing EDL and constructing a weighted bag-of-concepts with these entities. Also, graph based models like and-or-graphs (AOG) [120] or graph regularization methods [32] have recently been proving high accuracy results and are useful for multimodal topic modeling. For example, in [120] a novel representation using a Multimodal Topic And-Or Graph (MT-AOG) is presented.…”