Activity clustering

Activity clustering allows you to automatically group activities using a number of different factors (explained below). In order to start, you need to select at least one matter. Once you select the matter(s), all activities from that matter will be automatically added. After selecting the matter(s), you can also add additional activities manually. 


Once you've added the matter(s) and activities, you need to create the groups. In order for the clustering to work, each group needs at least one activity to start with. 


Select the factor you want to use for your clustering (matter is used for this example), and click "Cluster". 



Activities that have been dragged to a group serve as examples for the clustering. Clicking the cluster button lets the system predict in which group the activity should be. Clustered activities will appear yellow. If an activity is yellow, you can still undo the clustering. To confirm the clustering of an activity, please click on the activity and the yellow will disappear. The activity will remain in the group even if you click "Undo". If you want to refine the grouping, you can repeat the clustering process. Press cluster once your changes are made and confirmed. To confirm all activities, please click on "Confirm".


Factors for activity clustering

When using activity clustering, there are several factors you can use to cluster the activities. Below we explain each factor that can be used to cluster the activities. Please note that using multiple of these might confuse the clustering, as it will assume that each selected factor is equally important. Thus, more data is not necessarily better. 

If you think a form of data should be added to the activity clustering, let Alex know.

 

Matter

The Matter factor compares how often two activities take place in the same matter. If they occur regularly together in the same matters, two activities are considered more similar.

 

Co-occurrence

The co-occurrence looks at the timecard entry-level and determines how often two activities are categorized together. The algorithm also looks at the activities categorized 5 timecard entries before and 5 timecard entries after within the matter. 

The more two activities occur together, the more similar they are. Hence, the name co-occurrence.

 

Practice Group

When you select Practice Group as a factor to cluster activities, the algorithm compares how often two activities take place in the same practice group. If activities occur often together in the same practice groups, these activities are considered similar.

 

Temporal

The Temporal factor looks at the timeline of the selected matters and when each activity occurs within the matters. When using Temporal as a factor, activities that occur at the same time in a matter e.g. start, middle or end, are grouped together. 

 

Properties

The Properties factor looks at the narrative quality of the activity. It can be good, vague, or uncategorizable.