Cluster Analysis View

Introduction

Symphony's Custer Analysis view displays comments grouped together that have something in common. There are a variety of goals Symphony's Cluster Analysis supports, and several approaches (Analysis Methods) you can choose from. Knowing what you are looking to achieve and how to get there are important for getting the most out of Symphony's Cluster Analysis. This document covers common scenarios, suggested procedures for each scenario, and tips for setting the various settings that define the way each analysis is carried out.  
What to Expect from Cluster Analysis View
While we strive to make Symphony's Cluster Analysis View as useful as possible, we want to set the right expectations and be sure you understand the limitations. At best, this tool will reduce the effort it takes to analyze your data, but it will not do a complete job, and will in fact make mistakes. Symphony does not "understand" text the way human's do. It can only look for patterns that may through common usage have the same meaning. Some data is better suited for this type of analysis, while some is not suited at all. Whatever the case, you should consider this tool as only a contributor to your analysis efforts; you still need to spend time in Symphony's Content Explorer or any of the other tools Symphony provides to finish your analysis.

Exact Match Analysis

 The Exact Analysis creates clusters for all of the comments that are the same. "Exact" is loosely defined, in that synonyms are are considered to match, and punctuation and certain common words (e.g. is, to, the) are ignored. As far as Symphony is concerned, "I love my manager" and "We like our boss) are exact matches. You would run an Exact Match analysis if you expect to find a reasonable number of comments that match in this manner. Reasonable is dependent on the nature of your data. For example, an employee engagment survey typically yields a good number of matches because within each organization there tends to be several topics that are deemed important. The matches can be used as the basis for themes and/or coded en masse to an existing coding structure.

To perform an Exact Match Analysis:

  1. Open Cluster Analysis View
  2. Choose the Exact Match option
  3. Click the Refresh button or press function key F5. Results will appear in the left panel of the view.
  4. Click on one of the results and drag it onto your coding structure (on the right-side of the view) and drop it where you want it to go. Dropping it on the top level project node will create a new code and the comments in the cluster will be coded to it. Dropping it on an existing code will create a new sub-code. If the drag is performed using the right mouse button instead of the left, no code will be create; instead the comments will be coded to the code you dropped the result on. Repeat this step until all of the Exact Match results have been coded.

Exploratory Analysis

 The Exploratory Analysis looks for words or phrases of a fixed length and groups comments around them. This type of analysis is good both for themes discovery and coding efficiency. Generally speaking, this analysis is good for any type of project. It is best run after the Exact Match analysis. Often, phrases will be found in the the middle of comments that match up with phrases that surfaced in an Exact Match analysis. The Exploratory Analysis has a couple extra settings:
  • Minimum Matches tells Symphony to include only results where this many matching comments are found.
  • Phrase Range tells Symphony to look for matching phrases within this range. If Symphony finds comments that match the upper range, it looks further to see whether they match over a greater length and use that.

 To peform an Exploratory Analysis:

  1. Open Cluster Analysis view.
  2. Set the Phrase Length. For best results, start by setting it to the longest phrase length you expect to find in your data. If you set it too long, Symphony will go with the last length that it finds results and will display the actual length on the status line at the bottom of the view.
  3. Run the analysis by clicking the Refresh button or by pressing function key F5.
  4. Click on one of the results and drag it onto your coding structure (on the right-side of the view) and drop it where you want it to go. Dropping it on the top level project node will create a new code and the comments in the cluster will be coded to it. Dropping it on an existing code will create a new sub-code. If the drag is performed using the right mouse button instead of the left, no code will be create; instead the comments will be coded to the code you dropped the result on. Repeat this step until all of the Exact Match results have been coded.
  5. Repeat #3-5 for different Phrase Length setttings.

Themed Analysis

  The Themed Analysis uses your existing coding structure as the basis for analysis. With this analysis, Symphony compares the comments being analyzed with your coded comments and makes determinations o where they best fit.  This analysis is best performed after your data has been analyzed to the point where you feel you have identified the most important themes. Generally speaking, you should set a filter so that the analysis is performed only on your uncoded comments. If your coding structure has codes that don't contain any comments, Symphony will use the text of the codes as the basis for comparison.

To perform a Themed Analysis

 
  1. Open Cluster Analysis view.
  2. In the Filter Group, set a filter to the content you want to analyze. Usually this will be the Unassigned code if you imported unstructured data.
  3. Run the analysis by clicking on the Refresh button or by pressing function key F5.
  4. Click on a code in the left panel. The comments contained included in the cluster will appear in the middle of the view. Each comment has a score. There is no absolute meaning to the scores, but higher values suggest better fits to the cluster. It is up to you to make the determination as to whether comments in the cluster should be coded to the selected code. Scroll down the comments, inspecting them. Once you reach a point where you don't feel that they fit, you can select the ones you want to code, then drag and drop them onto the corresponding code in you coding structure in the right panel.
  5.