Quick Start
- 4GB RAM (min), 8GB RAM (recommended)
- Java 8+
A .bat script is provided for Windows users. Simply double-click it to launch SONDY. A .sh script is provided for Linux and Mac OS X users. To launch SONDY, open a terminal, navigate to the directory where the script is and type «./autostart.sh». Both scripts are text files that can be edited to adjust the amount of memory allocated to the application.
The workspace is a directory located on your hard drive where SONDY stores datasets, stop-word lists, etc. Go to "Edit -> Edit configuration" to set up the workspace. By default, the workspace is located in the same directory as SONDY.jar.
A dataset consists in a set of messages (e.g. tweets), and - optionally - a network (e.g. the network which describes the follower-followee relationships among the authors of the tweets). To import a new dataset into SONDY, select a CSV file containing the messages and -optionally- a CSV file containing the directed network structure, fill the required fields and click the "Import" button.
- One message per line
- 3 fields per line, separated by a tab: author \t timestamp (format: YYYY-MM-DD HH:mm:ss) \t message_content
- Example: adrienguille 2009-05-28 22:13:58 Hello Twitter! #myfirstTweet
- One directed edge per line
- 2 fields per line: initial_vertex \t terminal_vertex
- Example : adrienguille NoelGallagher
Once the data have been successfully imported, the dataset will appear in the list of available datasets.
SONDY requires the message set to be prepared before any event detection can be applied. This - at least - consists in discretizing the time-axis so that frequencies can be computed on the message stream. The time-slice length should be given in minutes (e.g. 30 min). Messages can be tokenized into unigrams, bigrams or trigrams. Additional preprocessing, i.e. stemming or lemmatization, can be added if necessary.
To preprocess a dataset, select it in the list of available datasets, fill in the required fields and click the "Preprocess" button.
Once the dataset have been successfully preprocessed, it will appear in the drop down list.
To detect events, select a preprocessed dataset, then select a method (adjust parameters values if needed) and click the "Apply method" button.
The detected events are listed in the table on the right. Left-clicking an event in the table updates the chart on the left with the frequencies of the words describing it. Right-cliking an event opens a new window that allows browsing and searching through the related messages.
To analyse users' influence, select an event, then select a method (adjust the parameter values if needed) and click the "Apply method" button.
The network is drawn on the left and is colored according to the ranks (in terms of influence) computed for the users. Right clicking a vertex opens a new window that reveals the user id and allows browsing and searching through his/her messages. The rank distribution is plotted on the right.