8.3 Latent Semantic Indexing

CiteSpace provides a somewhat underdeveloped Latent Semantic Analysis function under Text ►Latent Semantic Analysis. The Latent Semantic Analysis is based on a singular value decomposition of the term by document matrix. It is a dimension reduction method (Deerwester, Dumais, Landauer, Furnas, & Harshman, 1990).

Use the browse button to locate at least two data sources, i.e. folders of text files in plain full text or the WoS format. After select each data source, add it to the list using the button “Add to the List” then press the “Analyze” button. Then wait for it to finish …

Once it is done, five most representative words in each dimension are shown in the user interface.

Three coarse visualizations of the latent semantic space are provided for the three most prominent dimensions of the latent semantic space. Each visualization shows a mixture of terms and documents. You can zoom in and out, change the font size of labels, and the length of a label. That is about it. This function has been there for years, but it has not been actively developed.