1. biolab
  2. Untitled project
  3. orange-text


orange-text / docs / old-html / widgets / catalog / Text / TextFeatureSelection.htm

<title>Text feature selection</title>
<link rel=stylesheet href="../../../style.css" type="text/css" media=screen>
<link rel=stylesheet href="style-print.css" type="text/css" media=print></link>


<h1>Text feature selection</h1>

<img class="screenshot" src="../icons/TextFeatureSelection.png">
<p>Selection of textual features.</p>



<DL class=attributes>
<DT>Examples (ExampleTable)</DT>
<dd>Attribute-valued data set with text features as metaatributes.</dd>

<DL class=attributes>
<DT>Examples (ExampleTable)</DT>
<DD>Attribute-valued data set with the selected text features as metaatributes.</DD>


<p>This widget is used to select a subset of textual features which will be
used in further analysis. It is also used for selection of documents. Feature selection
can be performed using three measures: term frequency, random, and term document
frequency. Term frequency measure selects terms based on their frequency in the
document collection, random measure randomly selects terms, whereas the term document
frequency measure selects terms based on the number of documents they appear in.
Document selection measures are word frequency and number of features. Word frequency
measure selects documents based on the number of words in the document, whereas
number of features measure discriminates based on the number of different features
in the document. The mentioned measures can be chosen in the Select measure box.
The Select operator box is used to specify whether features/documents with the
score above or below the number set in the threshold box are eliminated. If the
percentage checkbox is selected, then the number in the threshold box is treated
as the percent of features/documents that have to be removed. For example,
if we select TF (term frequency) measure, MAX operator, a threshold of 10, and
leave the percentage box ticked, we will remove 10% of features that appear most
frequently in the collection. Selecting, for example, WF (word frequency) measure,
MIN operator, a threshold of 3, while unselecting percentage checkbox will remove
those documents that have less than three words. Statistics for features box shows
some basic statistics about features, whereas the Statistics for documents box
shows similar information, only for documents.</p>

<a href="TextFeatureSelection.png"><img class="schema" src="TextFeatureSelection.png" alt="Text feature selection widget"></a>


<p>Below is a simple example how to use this widget. The input is fed
directly from the <a href="BagOfWords.htm">Bag of words</a> widget, and the output
is sent to the Correspondence analysis widget for visualization.</p>

<a href="TextFeatureSelection-Example.png"><img src="TextFeatureSelection-Example.png" alt="Schema with TextFeatureSelection"