Engineering the Policy-making Life Cycle

Seventh Framework Programme - Grant Agreement 288147

Opinion Mining Software Prototype

If you have any problems with this code/data/files please contact Pedro Coelho (


Here are the necessary steps to install the software:


The stand alone prototype includes a user interface that allows two types of users to login (admins and normal users), and provides a different set of functionalities for each of these types of users. The interface for standard users allows them to select the problem / domain they wish to explore, and then for each domain it provides means for the user to select the topics she/he wants to visualize, as well as the time span to consider in this exploration. After this selection, the system draws a plot of the data in which a user can see the a line with the aggregated sentiment score along time together with a confidence band around the line reflecting the variability in this aggregated score. A second plot is also drawn with the individual sentiment scores assigned to each post, which lead to the aggregated sentiment expressed by the mentioned lines. Figure 1) shows an illustration of this part of the GUI.
Figure 1 - Standard user GUI - exploring the aggregated sentiment of topics.
Standard users can also drill down to individual posts. A table is presented with all posts within the selected time span (Figure 2), with a search box that allows easy filtering of these posts. The user may also select and individual post (through its ID) to obtain the specific text of this post together with the assigned sentiment scores for each topic.
Figure 2 - Standard user GUI - exploring the individual posts.
The stand alone OM prototype also provides and administrator graphical user interface (Figure 3).
Figure 3 - Administrator GUI.
In the first section of this interface admins can select the crawlers (providing a command which will be run by the system), select the topics and tune the parameters of the opinion mining models that are used by our prototype. Currently, it is possible to tune the following parameters of these models: A second section of the admin GUI allows these users to train new opinion mining models by clicking on a button and also to inspect and tag new posts (Figure 4). For this latter task we provide a table where the available posts can be filtered by ID, date and title. Using these filtering facilities the user may drill down to a specific post and eventually tag it for sentiment concerning the available topics.
Figure 4 - Administrator GUI - tagging posts.