Broadcast News Story Segmentation Using Manifold Learning on Latent Topic Distributions
Lu, Xiaoming and Xie, Lei and Leung, Cheung-Chi and Ma, Bin and Li, Haizhou

Article Structure

Abstract

We present an efficient approach for broadcast news story segmentation using a manifold learning algorithm on latent topic distributions.

Introduction

Story segmentation refers to partitioning a multimedia stream into homogenous segments each embodying a main topic or coherent story (Allan, 2002).

Our Proposed Approach

In this paper, we propose to apply LE on the LDA topic distributions, each of which is estimated from a text block.

Experimental setup

Our experiments were evaluated on the ASR transcripts provided in TDT2 English Broadcast news corpusl, which involved 1033 news programs.

Topics

LDA

Appears in 22 sentences as: LDA (22)
In Broadcast News Story Segmentation Using Manifold Learning on Latent Topic Distributions
  1. The latent topic distribution estimated by Latent Dirichlet Allocation ( LDA ) is used to represent each text block.
    Page 1, “Abstract”
  2. We evaluate two approaches employing LDA and probabilistic latent semantic analysis (PLSA) distributions respectively.
    Page 1, “Abstract”
  3. To deal with this issue, Latent Dirichlet Allocation ( LDA ) (Blei et al., 2003) has been proposed.
    Page 1, “Introduction”
  4. LDA has been proved to be effective in many segmentation tasks (Arora and Ravindran, 2008; Hall et al., 2008; Sun et al., 2008; Riedl and Biemann, 2012; Chien and Chueh, 2012).
    Page 1, “Introduction”
  5. In this paper, we propose to apply LE on the LDA topic distributions, each of which is estimated from a text block.
    Page 2, “Our Proposed Approach”
  6. 2.1 Latent Dirichlet Allocation Latent Dirichlet allocation ( LDA ) (Blei et al., 2003) is a generative probabilistic model of a corpus.
    Page 2, “Our Proposed Approach”
  7. In LDA , given a corpus D 2 {d1, d2, .
    Page 2, “Our Proposed Approach”
  8. An LDA model is characterized by two sets of prior parameters 04 and [3.
    Page 2, “Our Proposed Approach”
  9. Given the ASR transcripts of N text blocks, we apply LDA algorithm to compute the corresponding latent topic distributions X = [x1, x2, .
    Page 2, “Our Proposed Approach”
  10. ,x N] in RK , where K is the number of latent topics, namely the dimensionality of LDA distributions.
    Page 2, “Our Proposed Approach”
  11. We use G to denote an N -n0de (N is number of LDA distributions) graph which represents the relationship between all the text block pairs.
    Page 2, “Our Proposed Approach”

See all papers in Proc. ACL 2013 that mention LDA.

See all papers in Proc. ACL that mention LDA.

Back to top.

topic distributions

Appears in 16 sentences as: topic distribution (2) topic distributions (14)
In Broadcast News Story Segmentation Using Manifold Learning on Latent Topic Distributions
  1. We present an efficient approach for broadcast news story segmentation using a manifold learning algorithm on latent topic distributions .
    Page 1, “Abstract”
  2. The latent topic distribution estimated by Latent Dirichlet Allocation (LDA) is used to represent each text block.
    Page 1, “Abstract”
  3. We employ Laplacian Eigenmaps (LE) to project the latent topic distributions into low-dimensional semantic representations while preserving the intrinsic local geometric structure.
    Page 1, “Abstract”
  4. To further improve the segmentation performance, using latent topic distributions and LE instead of term frequencies to represent text blocks is studied in this paper.
    Page 2, “Introduction”
  5. In this paper, we propose to apply LE on the LDA topic distributions , each of which is estimated from a text block.
    Page 2, “Our Proposed Approach”
  6. [3 is a K x V matrix, which defines the latent topic distributions over terms.
    Page 2, “Our Proposed Approach”
  7. Given the ASR transcripts of N text blocks, we apply LDA algorithm to compute the corresponding latent topic distributions X = [x1, x2, .
    Page 2, “Our Proposed Approach”
  8. ,yN] (yi is a column vector) to indicate the low-dimensional representation of the latent topic distributions X.
    Page 2, “Our Proposed Approach”
  9. The projection from the latent topic distribution space to the target space can be defined as:
    Page 2, “Our Proposed Approach”
  10. i=1 mesegt where y, and y j are the latent topic distributions of text blocks i and j respectively, and y, — yj H2 is the Euclidean distance between them.
    Page 3, “Our Proposed Approach”
  11. 0 PLSA-DP: PLSA topic distributions were used to compute sentence cohesive strength.
    Page 3, “Experimental setup”

See all papers in Proc. ACL 2013 that mention topic distributions.

See all papers in Proc. ACL that mention topic distributions.

Back to top.

development set

Appears in 5 sentences as: developent set (1) Development Set (1) development set (3)
In Broadcast News Story Segmentation Using Manifold Learning on Latent Topic Distributions
  1. We separated this corpus into three non-overlapping sets: a training set of 500 programs for parameter estimation in topic modeling and LE, a development set of 133 programs for empirical tuning and a test set of 400 programs for performance evaluation.
    Page 3, “Experimental setup”
  2. A number of parameters were set through empirical tuning on the developent set .
    Page 3, “Experimental setup”
  3. Figure 1 shows the results on the development set and the test set.
    Page 4, “Experimental setup”
  4. Development Set
    Page 4, “Experimental setup”
  5. All the four approaches had similar performances on the development set and the test set.
    Page 4, “Experimental setup”

See all papers in Proc. ACL 2013 that mention development set.

See all papers in Proc. ACL that mention development set.

Back to top.

topic modeling

Appears in 5 sentences as: topic model (2) topic modeling (3)
In Broadcast News Story Segmentation Using Manifold Learning on Latent Topic Distributions
  1. To deal with these problems, some topic model techniques which provide conceptual level matching have been introduced to text and story segmentation task (Hearst, 1997).
    Page 1, “Introduction”
  2. We separated this corpus into three non-overlapping sets: a training set of 500 programs for parameter estimation in topic modeling and LE, a development set of 133 programs for empirical tuning and a test set of 400 programs for performance evaluation.
    Page 3, “Experimental setup”
  3. When evaluating the effects of different size of the training set, the number of latent topics in topic modeling process was set to 64.
    Page 3, “Experimental setup”
  4. When evaluating the effects of different number of latent topics in topic modeling computation, we fixed the size of the training set to 500 news programs and changed the number of latent topics from 16 to 256.
    Page 3, “Experimental setup”
  5. We observe that LE projection makes the topic model more stable with different numbers of latent topics.
    Page 4, “Experimental setup”

See all papers in Proc. ACL 2013 that mention topic modeling.

See all papers in Proc. ACL that mention topic modeling.

Back to top.

latent semantic

Appears in 3 sentences as: latent semantic (3)
In Broadcast News Story Segmentation Using Manifold Learning on Latent Topic Distributions
  1. We evaluate two approaches employing LDA and probabilistic latent semantic analysis (PLSA) distributions respectively.
    Page 1, “Abstract”
  2. Probabilistic latent semantic analysis (PLSA) (Hofman-n, 1999) is a typical instance and used widely.
    Page 1, “Introduction”
  3. PLSA is the probabilistic variant of latent semantic analysis (LSA) (Choi et al., 2001), and offers a more solid statistical foundation.
    Page 1, “Introduction”

See all papers in Proc. ACL 2013 that mention latent semantic.

See all papers in Proc. ACL that mention latent semantic.

Back to top.

objective function

Appears in 3 sentences as: objective function (3)
In Broadcast News Story Segmentation Using Manifold Learning on Latent Topic Distributions
  1. The objective function can be transformed
    Page 2, “Our Proposed Approach”
  2. Given the low-dimensional semantic representation of the test data, an objective function can be defined as follows:
    Page 3, “Our Proposed Approach”
  3. The story boundaries which minimize the objective function 8 in Eq.
    Page 3, “Our Proposed Approach”

See all papers in Proc. ACL 2013 that mention objective function.

See all papers in Proc. ACL that mention objective function.

Back to top.

PDF available from www.aclweb.org.
Send us an email