Difference between revisions of "3.3:Network"

Revision as of 15:03, 23 February 2016

The Network output options are useful for exploring and measuring the associations between two or more concepts (search strings) within the articleset(s) used in your query. AmCAT has two options for doing so: clustermaps and network (association) graphs. Both require two or more search terms in the Keyword Search window to work. Remember to put each term on a new line in the window, like this:

Clustermaps

Clustermaps show the overlap between concepts in absolute numbers (e.g. there are 104 articles in which search term X occurs but not search term Y, 3490 articles in which search term Y occurs but not X, and 6753 articles in which both occur). To create one, enter two or more search terms in the Keyword Search window, and select ‘Clustermap’ from the dropdown menu under the ‘Network’ output option.

Then, select ‘diagram’ as output type, and click 'Query'. AmCAT will create a visual clustermap.

Below, you see a clustermap for the search terms ‘nuclear’ and ‘danger’ shown in the previous screenshot. As you can see, the term ‘danger’ occurs in 7084 articles, and the term ‘nuclear’ in 57742 articles. Of these 57742 articles, 7070 also contain the term ‘danger’ (the purple circle), while 50672 contain only ‘nuclear’ but not ‘danger’. The little yellow balls in the red ‘danger’ circle each represent an article. So, there are 14 articles that are about ‘danger’ but not ‘nuclear’. You can click on a yellow ball to see a table with these 14 articles. Clicking a row in the table opens the full text of the corresponding article in a new tab. In this way, you can conveniently explore when, or in which context, certain concepts (search terms) co-occur and when they do not.

If you want to export the data for the clustermap from AmCAT, go back to the output type and select one of the CSV options, or SPSS. This exports the numbers in the clustermap as a table in the chosen file format. You can also display the clustermap as a table instead of a diagram, by selecting the 'Table' output option.

Association

The Association option measures the strength of the association between two concepts in an articleset using conditional probability, and displays the results as a network graph or a table. Conditional probabilities basically measure the chance that, given that B occurs, A also occurs, and the other way around (what is the change that, given that A occurs, B also occurs). This measure is asymmetrical (i.e. the two probabilities are usually not the same). The higher these conditional probabilities, the stronger the concepts are associated with each other. AmCAT can display conditional probabilities as a table (which you can export for use outside of AmCAT) or as a graph.

After entering two or more search terms in the Keyword Search window, click on Association from the Network dropdown menu to open a window where you can specify what your graph or table will look like.

Let’s start with output type, the last option.

• From-to tables are tables that show the condition (If search term A occurs..) on the left and the other term (..what is then the chance that term B also occurs?) on the right. The probability is shown in the rightmost column.

• Cross tables instead put the condition in the rows and the other term in the columns. You thus read them as follows: if [row] occurs, what is then the chance that [column] also occurs?

• Table (from-to) or Table (cross table) display the results within AmCAT, the two CSV options create a CSV file that exports these results from AmCAT.

• Graph, lastly, creates a network graph that visually displays conditional probabilities using nodes and arrows. There, the condition is the first node, and the arrow points towards the other term. For example, in the graph below, the chance that Cameron is mentioned in an article that mentions Obama is 0,112; the chance that given that Cameron is mentioned, Obama is also mentioned is 0,066.

Interval is another important option. By default, AmCAT shows you the conditional probabilities for the full time period included in the query (Interval ‘None’). Specifying an interval allows you to calculate separate conditional probabilities for each time period with this duration (week, month or year), and see how the association between search terms changes over time. It works for both graphs and tables.

By default, AmCAT calculates conditional probabilities based on the amount of articles. If you want to weigh for the number of hits in this calculation (e.g. articles in which the search terms occur together often are counted more heavily than those in which they only occur together once), tick the box ‘Weigh for number of hits’. You can see that this results in different probabilities.

Number format allows you to choose between proportion and percentage.

For graphs, you also have the option to show only associations above a certain threshold, for example only conditional probabilities higher than 10%. Enter this probability (as a proportion) in the ‘Graph: threshold’ box.

Lastly, you can also choose if you want the network graphs to display the probability for each tie or not (the default is not to show them – tick the box if you want to see the probabilities like in the examples shown here).

If you want to download and save an Associations graph, click on the ‘save’ box that appears to the right of the graph. You can save graphs as SVG, JPEG or PNG files.