Labeling research fronts

When conducting a case study, labeling research fronts is the most important part of exploratory data analysis. The investigator needs to "dive into" the data, in order to get familiar with the specialty that is being mapped, but more importantly to validate the process of gathering the collection of papers and clustering the papers into research fronts.

If available, the investigator should employ subject matter experts (SMEs) to produce and validate research front labels. However, the investigator should still go through the labeling process to check the gathering of data and the clustering results.

Tools for browsing papers in research fronts

To produce research front labels it is necessary to browse titles of the papers within each research front for themes. This can be done in 3 ways:

1. Use DIVA's interactive exploratoration tools to select and browse papers titles of research fronts on the timeline.
2. Work off of a research front report.
3. Use an interactive timeline webpage.

Using the research front report is probably the most convenient method. The report can be printed out (not recommended as it may be VERY large). It can also be emailed to SMEs with some simple instructions for making the research front labels.

Generating the research front labels.

• It is not necessary to label all research fronts.
• A research front label is specified by its cluster number and a text string corresponding to the label.
• Keep in mind the following guidelines:
  • Because of the characteristics of the dendrogram seriation, easily discriminated research fronts tend to appear at the top and bottom of the timeline. These will be easy to label.
  • Research fronts in the center part of the timeline are less easily distinguished and may nedd to be labeled as indeterminate.
  • Usually a giant cluster of papers appears toward the vertical center of the timeline. This cluster should be labeled indeterminate.
  • In small clusters of papers watch for artifact research fronts corresponding to authors that publish the same papers in several different journals.
  • Remember that research fronts may not correspond to research topics. Research fronts reflect patterns of cited base knowledge, which may change as the knowledge related to a research problem cumulates. Perhaps labels such as early xx research or xx research citing yy may be appropriate.
• Browse the paper titles in each research front, look for repeated phrases that distinguish a theme about the papers. Be on the lookout for abbreviations and phrase variants, e.g., severe acute respiratory syndrome may occur in many titles, while the equivalent term SARS may occur in many other papers.
• Once a theme for labeling of a research front is identified, record the RF number and a concise research front label. There is no limit on the number of characters in the label, but only about 25-30 characters will show on the map. Poorly defined research fronts can be labeled with question marks or some other comment, or ignored.
• Labels can be entered into a table in the database or into a text file. When generating maps, DIVA looks for for the database table first, if that's missing, it checks for the file bib_labels.txt in the project folder, if that's missing, then no labels are put on the map.
  • To put the labels into a database table, open the project database, create table bib_labels with field rf_key (type as number) and label (type as text). For each research front enter RF number into rf_key and the label into the corresponding label entry.

• To put the labels into a file, create a text file bib_labels.txt in the project folder. Edit the file. Each line in the file corresponds to a research front. For each research front label, enter the RF number first, then a comma, then the label string, without quotes. Do not put commas into the label string. The RF labels do not need to be in numerical order, and are typically entered in dendrogram order, the usual sequence for browsing the research front titles.

Example

Below is an example generated from the mad_cow dataset, with research front labels generated and displayed.