#### DMCA

## Exemplar-based Visualization of Large Document Corpus

Citations: | 15 - 2 self |

### Citations

3316 |
Principal Component Analysis
- Jolliffe
- 2002
(Show Context)
Citation Context ...erve in the projected space the distance relationships among the documents in their original space. Depending on the choice of mapping functions, both linear (e.g., principle component analysis (PCA) =-=[13]-=-) and nonlinear (e.g., ISOMAP [24]) dimensionality reduction techniques have been proposed in the literature. Facing the ever-increasing amount of available documents, a major challenge of text visual... |

2761 |
Pattern Classification
- Duda, Hart, et al.
- 2001
(Show Context)
Citation Context ... 3283 9 Migraine 3703 10 Otitis 2596 4.2 Evaluation Measurement We evaluated the visualization results quantitatively based on the label predication accuracy with the k-nearest neighbor (k-NN) method =-=[8]-=- in the visualization space. Documents are labeled with discussion groups in the 20Newsgroups data, and with disease names in the 10PubMed data. Majority voting among the training documents in the k n... |

2419 |
A Global Geometric Framework for Nonlinear Dimensionality
- Tenenbaum, Silva, et al.
- 2000
(Show Context)
Citation Context ...stance relationships among the documents in their original space. Depending on the choice of mapping functions, both linear (e.g., principle component analysis (PCA) [13]) and nonlinear (e.g., ISOMAP =-=[24]-=-) dimensionality reduction techniques have been proposed in the literature. Facing the ever-increasing amount of available documents, a major challenge of text visualization is to develop scalable app... |

2403 |
An algorithm for suffix stripping
- Porter
- 1980
(Show Context)
Citation Context ...2 shows the 10 document sets (15,565 documents) retrieved. From all the retrieved abstracts, the common and stop words are removed, and the words are stemmed using Porter’s suffix-stripping algorithm =-=[19]-=-. Finally, we built a word-document matrix of the size 22437 × 15565. 3 http://www.cs.uiuc.edu/homes/dengcai2/Data/TextData.html 4 http://www.ncbi.nlm.nih.gov/pubmed/ Table 1. Summary of data subsets ... |

2381 | Nonlinear dimensionality reduction by locally linear embedding
- Roweis, Saul
- 2000
(Show Context)
Citation Context ...D( f (xi), f (x j)) is the Euclidean distance between the corresponding two points in the projected space, and f : X → Y is a mapping function [23]. In general, multidimensional projection techniques =-=[13, 4, 24, 20]-=- can be divided into two major categories based on the function f employed: Linear Projection methods and Non-linear Projection methods. Linear projection creates an orthogonal linear transformation t... |

1686 |
Finite Mixture Models
- McLachlan, Peel
- 2000
(Show Context)
Citation Context ...35 40 45 50 number of neighbors (b) Fig. 1. Accuracy with k-NN in the two-dimensional visualization space with different k: (a) 20Newsgroups-I (3 topics), (b) 20Newsgroups-II (20 topics). be found in =-=[15]-=-. In our experiments, the number of topics for all the topic models is simply set based on the ground truth. Another important observation from Figure 1 is that EV-844 constantly provides a higher acc... |

753 | Probabilistic latent semantic analysis
- Hofmann
- 1999
(Show Context)
Citation Context ...ed objective function. Finally, knowledge or information is usually sparsely encoded in document collections. Thus, main topics of a text corpus are more accurately described by a probabilistic model =-=[10]-=-. That is, a document is modeled as a mixture of topics, and a topic is modeled based on the probabilities of words. In the paper, we propose an Exemplar-based approach to Visualize (EV) extremely lar... |

590 |
Multidimensional Scaling
- Cox, Cox
- 2000
(Show Context)
Citation Context ...D( f (xi), f (x j)) is the Euclidean distance between the corresponding two points in the projected space, and f : X → Y is a mapping function [23]. In general, multidimensional projection techniques =-=[13, 4, 24, 20]-=- can be divided into two major categories based on the function f employed: Linear Projection methods and Non-linear Projection methods. Linear projection creates an orthogonal linear transformation t... |

399 | A survey of clustering data mining techniques
- Berkhin
- 2006
(Show Context)
Citation Context ...ccount of the latent structure in the given data, i.e., topics in the document collection. To this end, Least Square Projection (LSP) [17] first chooses a set of control points using k-medoids method =-=[1]-=- based on the number of topics and then obtains the projection through the least square approximation, in which the data are projected following the geometry defined by the control points. Recently, i... |

251 | Visualizing knowledge domains
- BÖRNER, CHEN, et al.
(Show Context)
Citation Context ...bles us to browse intuitively through huge amounts of data and thus provides a very powerful tool for expanding the human ability to comprehend high dimensional data. A number of different techniques =-=[21, 3, 5]-=- were proposed in the literature for visualizing a large data set, among which multidimensional projection is the most popular one. In document visualization, let X = {x1,x2,..xn} ∈ Rm×n be a word-doc... |

185 |
Hierarchical parallel coordinates for exploration of large datasets
- Fua, Ward, et al.
- 1999
(Show Context)
Citation Context ...ons present unprecedented challenges for the development of highly scalable methods that can be implemented in a linear polynomial time. Therefore, hierarchical-clustering based visualization methods =-=[9, 16]-=- are proposed to partially solve the memory and computation problem, in which a hierarchical cluster tree is first constructed using a recursive partitioning process, and then the elements of that tre... |

108 | M.: Convex and semi-nonnegative matrix factorizations
- Ding, Li, et al.
- 2010
(Show Context)
Citation Context ...Equation (1) is monotonically decreasing under the updating rules in Equations (2)- (3). Due to the space limit, we give an outline of the proof of the propositions and omit the details. Motivated by =-=[6]-=-, we plan to render the proof based on optimization theory, auxiliary function and several matrix inequalities. First, following the standard theory of constrained optimization, we fix one variable G ... |

84 |
From visual data exploration to visual data mining: A survey
- Oliveira, Levkowitz
(Show Context)
Citation Context ...ide logic of the model and offers users a chance to interact with the mining model so that questions can be answered. In general, it is convenient to transform document collections into a data matrix =-=[5]-=-, where the columns represent documents and the row vectors denote keyword counting after pre-processing. Thus, textual data sets have a very high dimensionality. A common way of visualizing text corp... |

49 | On the equivalence between non-negative matrix factorization and probabilistic latent semantic indexing. Computational Stat. & Data Analysis 52(8):3913–3927
- Ding, Li, et al.
- 2008
(Show Context)
Citation Context ... set). In the second step, we need to obtain the “soft” cluster indicators in the low-rank exemplar subspace, representing the probability of each document proportion to the topics in the topic model =-=[7]-=-. We formulate this task as an optimization problem, J = min W≥0,G≥0 ‖ ˜X − CWG T ‖ 2 F (1) = Tr( ˜X T ˜X − ˜X T CWG T − GW T C T ˜X + GW T C T CWG T ) where W is the weight matrix and G is the cluste... |

46 |
News weeder: Learning to filter netnews
- Lang
- 1995
(Show Context)
Citation Context ...ation results by EV in Section 4.3, in which we also compared the computational speed of all the algorithms. 4.1 Data Sets For the experiments on document visualization, we used the 20Newsgroups data =-=[14]-=- and 10PubMed data. 20Newsgroups data consists of documents in the 20 Newsgroups corpus. The corpus contains 18,864 articles categorized into 20 discussion groups 3 with a vocabulary size 26,214. Note... |

28 | Probabilistic latent semantic visualization: topic model for visualizing documents
- Iwata, Yamada, et al.
- 2008
(Show Context)
Citation Context ... are projected following the geometry defined by the control points. Recently, incorporating probabilistic topic models into analyzing documents has attracted great interest in the research community =-=[12]-=- since it can provide a higher quality (i.e., more meaningful) visualization. In Probabilistic Latent Semantic Analysis (PLSA) [10], a topic is modeled as a probability distribution over words, and do... |

26 |
Visual Data Mining: Techniques and Tools for Data Visualization
- Soukop, Davidson
- 2002
(Show Context)
Citation Context ...bles us to browse intuitively through huge amounts of data and thus provides a very powerful tool for expanding the human ability to comprehend high dimensional data. A number of different techniques =-=[21, 3, 5]-=- were proposed in the literature for visualizing a large data set, among which multidimensional projection is the most popular one. In document visualization, let X = {x1,x2,..xn} ∈ Rm×n be a word-doc... |

23 | Least square projection: A fast high precision multidimensional projection technique and its application to document mapping
- Paulovich, Nonato, et al.
(Show Context)
Citation Context ... of a document based on the word frequency, most of them take no account of the latent structure in the given data, i.e., topics in the document collection. To this end, Least Square Projection (LSP) =-=[17]-=- first chooses a set of control points using k-medoids method [1] based on the number of topics and then obtains the projection through the least square approximation, in which the data are projected ... |

22 | Parametric embedding for class visualization
- Iwata, Saito, et al.
- 2005
(Show Context)
Citation Context ... , where W is the weight matrix, and G is the cluster indicator matrix. To reduce the clutter in the visualization, the exemplars in each cluster are first visualized through Parameter Embedding (PE) =-=[11]-=-, providing an overview of the distribution of the entire document collection. When desired, on the clicking of an exemplar, documents in the associated cluster or in a user-selected neighborhood are ... |

22 |
On improved projection techniques to support visual exploration of multidimensional data sets
- Tejada, Minghim, et al.
- 2003
(Show Context)
Citation Context ...here δ(xi,xj) is the original dissimilarity distance and D( f (xi), f (x j)) is the Euclidean distance between the corresponding two points in the projected space, and f : X → Y is a mapping function =-=[23]-=-. In general, multidimensional projection techniques [13, 4, 24, 20] can be divided into two major categories based on the function f employed: Linear Projection methods and Non-linear Projection meth... |

17 |
Less is more: Sparse graph mining with compact matrix decomposition
- Sun, Xie, et al.
- 2008
(Show Context)
Citation Context ...imal low-rank approximation has gained increasing popularity in recent years due to its great computational and storage efficiency. The representative ones include Algorithm 844 [2], CUR [18] and CMD =-=[22]-=-. Typically, a near-optimal low-rank approximation algorithm first selects a set of columns C and a set of rows R as the left and right matrices of the approximation. Then, the middle matrix U is comp... |

14 | A novel hierarchical point placement strategy and its application to the exploration of document collections
- PAULOVICH, MINGHIM
(Show Context)
Citation Context ...ons present unprecedented challenges for the development of highly scalable methods that can be implemented in a linear polynomial time. Therefore, hierarchical-clustering based visualization methods =-=[9, 16]-=- are proposed to partially solve the memory and computation problem, in which a hierarchical cluster tree is first constructed using a recursive partitioning process, and then the elements of that tre... |

13 | Algorithm 844: Computing sparse reduced-rank approximations to sparse matrices
- Berry, Pulatova, et al.
- 2005
(Show Context)
Citation Context ...tion methods, near-optimal low-rank approximation has gained increasing popularity in recent years due to its great computational and storage efficiency. The representative ones include Algorithm 844 =-=[2]-=-, CUR [18] and CMD [22]. Typically, a near-optimal low-rank approximation algorithm first selects a set of columns C and a set of rows R as the left and right matrices of the approximation. Then, the ... |

2 |
Fast monte carlo algorithms for matrices iii: computing a compressed approximate matrix decomposition
- PETROS, RAVI, et al.
(Show Context)
Citation Context ...ods, near-optimal low-rank approximation has gained increasing popularity in recent years due to its great computational and storage efficiency. The representative ones include Algorithm 844 [2], CUR =-=[18]-=- and CMD [22]. Typically, a near-optimal low-rank approximation algorithm first selects a set of columns C and a set of rows R as the left and right matrices of the approximation. Then, the middle mat... |