For the latest version of JMP Help, visit JMP.com/help.


Basic Analysis > Text Explorer > Topic Analysis
Publication date: 11/10/2021

Image shown hereTopic Analysis

The Topic Analysis, Rotated SVD option performs a varimax rotation on the partial singular value decomposition (SVD) of the document term matrix (DTM). You must specify a number of rotated singular vectors, which corresponds to the number of topics that you want to retain from the DTM. After you specify a number of topics, the Topic Analysis report appears.

Topic analysis is equivalent to a rotated principal component analysis (PCA). The varimax rotation takes a set of singular vectors and rotates them to make them point more directly in the coordinate directions (toward the terms). This rotation makes the vectors help explain the text as each rotated vector orients toward a set of terms. Negative values indicate a repulsion force. The terms with negative values occur in a topic less frequently compared to the terms with positive values.

Image shown hereTopic Analysis Report

The Topic Analysis report shows the terms that have the largest loadings in each topic after rotation. There are additional reports that show the components of the rotated singular value decomposition.

The Top Loadings by Topic report shows a table of terms for each topic. The terms in each table are the ones that have the largest loadings in absolute value for each topic. Each table is sorted in descending order by the absolute value of the loading. These tables can be used to determine conceptual themes that correspond to each topic.

The Topic Analysis report also contains the following reports:

Topic Loadings

Contains a matrix of the loadings across topics for each term. This matrix is equivalent to the factor loading matrix in a rotated PCA.

Word Clouds by Topic

Contains a matrix of word clouds, one for each topic.

Topic Scores

Contains a matrix of document scores for each topic. Documents with higher scores in a topic are more likely to be associated with that topic.

Topic Scores Plots

Contains a Show Text button and a plot of topic scores for each document. The Show Text button opens a window that contains the text of the selected documents.

The Topic Scores Plots report is a visual representation of the matrix in the Topic Scores report. Each panel in the plot corresponds to one of the topics, or one of the columns of the Topic Scores matrix. Within each panel, each point corresponds to one of the documents in the corpus, or one of the rows of the Topic Scores matrix.

Variance Explained by Each Topic

Contains a table of the variance explained by each topic. The table also contains columns for the percent and cumulative percent of the variation explained by each topic.

Rotation Matrix

Contains the rotation matrix for the varimax rotation.

Image shown hereTopic Analysis Report Options

The Topic Analysis red triangle menu in the Text Explorer platform contains the following options:

Topic Scatterplot Matrix

Shows or hides a scatterplot matrix of the rotated singular value decomposition vectors. The Show Text button opens a window that contains the text of the selected documents.

Display Options

Contains options to show or hide content that appears in the Topic Analysis report. See Topic Analysis Report.

Rename Topics

Enables you to add descriptive names for one or more of the topics.

Save Document Topic Vectors

Saves a user-specified number of singular vectors from the rotated singular value decomposition as columns to the data table.

Save Topic Vector Formula

Saves a formula column with the Vector modeling type that contains the rotated singular value decomposition to the data table. The resulting column uses the Text Score() JSL function. For more information about this function, see Help > Scripting Index.

Save Term Topic Vectors

Saves the topic vectors as columns to the data table created by the Save Term Table option.

Remove

Removes the Topic Analysis report from the SVD report.

Want more information? Have questions? Get answers in the JMP User Community (community.jmp.com).