The UI is accessible at http://apps.lattice.cnrs.fr/ie/uidev/
The UI has two views, a main view (the default), and the AgreeDisagree view. Both are described below.
This view allows searching for actors' statements, via a structured search for each components of a statement: its speaker (i.e. a negotiation actor), its message, and the predicate relating the actor and the message. It also displays the context (document or sentence) for the statements, as well as an overview of their content based on keyphrases, DBpedia concepts and domain-specific concepts.
We called a proposition the triple that defines a statement, i.e. an ⟨actor, predicate, message⟩ triple. The predicate is the expression (i.e. a reporting verb or reporting noun) relating the actor and the message.
The triples' elements can be searched for individual, as described below. The numbered items correspond to the image above.
1. Actors search box: Restricts the propositions returned to those made by the actor searched
2. Actions search box: Restricts the propositions returned according to those whose predicate starts like the query entered in the search box, or to those whose predicate type matches the predicate type(s) in the selected checkboxes (support, oppose, "report", i.e. neutral reporting).
3a. Points box: Restricts the propositions returned to those containing the query term in the message
3b. Free text: This returns sentences matching the query term(s). Boolean operators can be used.
4. Confidence range: Propositions were assigned a score that indicates how likely they are a sound, informative proposition. The score is in a 0 to 5 scale. The dropdowns can be used to indicate the minimum and maximum confidence scores. By default, only the best propositions (those with confidence 5) are displayed.
5. Date range: The start and end date of the range are defined with dropdowns.
Click on the check button to run the query. The x button resets the form. Individual search boxes are reset by removing the term manually.
The "Point Only" dropdown is currently inactive and can be ignored.
For more details on query term matching for each type of search, see Chapter 6 in Ruiz Fabo (2017)1, esp. p. 175ff.
As can be seen on the image below, the left pane displays the propositions returned by the query.
The right pane (cf. the image below) displays two types of information: the propositions' context, and metadata automatically annotated in the propositions' messages.
The example below illustrates how the filtering functions can be used, also showing one of the metadata tabs returned for the query on the right pane.
This allows us to choose two actors using the dropdown menus. A relation type needs to be chosen too (agreement or disagreement).
This will display, on the left pane, keyphrases, DBpedia entities or Climate Tagger thesaurus concepts extracted from statements where the actors agree or disagree.
Clicking on one of the metadata points will display, on the right, the sentence(s) where the selected actors show the selected relation (agreement or disagreement), and where the metadata point was attested.
The figure below shows an example how to use this view.
1 Ruiz Fabo, Pablo. (2017). Concept-based and Relation-based Corpus Navigation: Applications of Natural Language Processing in Digital Humanities. PhD Thesis. Ecole Normale Supérieure, PSL Research University. Paris, France.