Marine Variables Linker

Find relations between changing variables in marine science literature

Event 1

Query

Event types

Event instances

Event 2

Query

Event types

Event instances

Relation

Query

Relation types

Relation instances

Relation graph

Help

1. Introduction

The Marine Variable Linker helps you to find related events in marine science literature. The current demo system searches in article abstracts from a large number of journals. It can find three kinds of events:

  • variables that are increasing
  • variables that are decreasing
  • variables that are changing into an unspecified direction

For example, the following sentence contains two events:

The increase of greenhouse gasses causes a decline of Artic sea ice.

The first event is an increase of the variable greenhouse gasses (marked in red). The second event is a decrease of the variable Artcic sea ice (marked in blue). Technically each event is a combination of a semantic predicate (increase, decrease or change) and a particular variable.

Events can be related to each other. In the example above, the two events occur together in the same sentence, so there exists a co-occurrence relation between them. If two events tend to co-occur frequently in text, this usually indicate that there exists some meaningful connection between them. In the above example, there is in fact a causal relation between the two events, as indicated by the use of the verb causes.

The current version of the Marine Variables Linker allows you to search for both co-occurrence and causal relations between events. Future versions will add searching for correlations and feedbacks.

2. Searching for events

Click on Event1 in menu bar at the top. This brings you to the section for Event 1.

Query

In the panel titled Query you can compose a query to search for events. Each query consists of one or more rules. A rule has a combination of drop-down lists and text boxes for its three parts:

  1. subject
  2. assertion
  3. value

For example, you can select variable as the subject, contains word as the assertion and iron as the value, to search for all events involving a variable containing the word iron.

You can click the green Add rule button to add more rules in order to narrow down your search. For example, you can further restrict events to only those with predicate decrease. Use the red Delete button to remove rules from the query.

By default, rules are combined by conjunction (that is, rule1 and rule 2 must match). If you want to combine rules by disjunction (that is, rule1 or rule 2 must match), then click on the blue OR button in the upper left corner.

Event types

Once your query is finished, click on the Search button. This will bring up a new panel titled Event types that shows all types of event matching the query. It contains a table showing the instance counts, predicates and variables. You can browse through multiple pages (if the table is long enough) or sort rows on a particular column. Note that there is maximimum to the number of event types shown (currently 500).

Event instances

Clicking on a row in the Event types table will bring up the Event instances panel that shows all the actual mentions of this event in journal articles. Each row in the table shows a sentence, year of publication and source. The event is marked in color; red for increasing, blue for decreasing and green for changing. Hovering the mouse over the icon will show a citation for the source article. Clicking on this icon will open the article containing the sentence in a new window. Note that in the case of a full text article, your need to have access rights to view it.

Click on the Reset button to start a new search.

Searching with specialisations and generalisations

The drop down next to the Search button offers three choices:

  • exact only: This means that only events exactly matching the search criteria will be returned. This is the default setting.
  • with specialisations: This means that also more specific types of events will be returned. For example, if the variable is sea ice, then the search result includes more specific types such as winter sea ice, Artic sea ice and frost, snow and baltic sea ice.
  • with generalisations: This means that also more general types of events will be returned. For example, if the variable is summer Arctic sea ice, then the search result includes more general types such as Arctic sea ice, sea ice and ice.

3. Searching for relations

In order to search for relations between events, you first need to define queries for Event 1 and Event 2. Then click on Relation in the meny bar at the top, which brings you to the section for relations.

Query

In the panel titled Query you can compose a query to search for relations. This works in the similar way as for events. A query typically consist of a single rule, although it is possible to use multiple rules where each one specifies conditions on the relation.

The first kind of relation is cooccurs. This means that the two events co-occur in the same sentence. When two events often co-occur in a single sentence, they tend to be associated in some way. This may indicate a correlation between the two types of events. However, it may also indicate a meaningless cooincidence or -- especially in case of low frequency -- the co-occurence may be purely by chance.

The second kind of relation is causes. This means that the two events in a sentence are causally related, where one event is the cause and other the effect. That is, causal relations are directed. Causality must be described in the sentence, for example, by words like causes, therefore or leads to, etc.

Relation types

Clicking on the Search button will bring up two new panels. The first panel is titled Relation types and contains a table with all pairs of event types in the given relation(s), specifying counts, relations, event predicates and event variables.

Relation graph

The second panel, called Relation graph, displays the same event types in the form of a graph. The nodes are event types with

  • red triangles for increasing variables,
  • blue triangles for decreasing variables and
  • green diamonds for changing variables.
Edges indicate relations between events, where the thickness of the edge represents the relative frequency of the relation. Causal relations have an arrow, pointing from cause to effect. The size of a node indicates the relative frequency of the event.

Your view on the graph can be changed using the green navigation icons at the bottom of the graph or by using the mouse:

  • Click and drag on nodes to move them around.
  • Click and drag on the cavas to move the whole graph.
  • Use the mouse wheel to zoom in or out.

Clicking on the icon in the upper right corner of the graph panel will show the graph in full screen mode, while clicking the icon will return it ti its orginal size.

Relation instances

To see the corresponding relation instances, either click on a row in the relation types table or click on an edge in the relation graph. This will open up a new panel called Relation instances, showing sentences, years and citations of articles containing the given relation. The events are marked in color: red for increasing, blue for decreasing and green for changing. Hovering the mouse over the icon will show a citation for the source article. Clicking on this icon will open the article containing the sentence in a new window. Note that in the case of a full text article, your need to have access rights to view it.

Click on the Reset button to start a new search.

Open-ended relation search

Suppose you are interested in the effects of increased temperature. In that case, you have to define Event 1 as an increase of the variable temperature, while leaving Event 2 open, and then search for causal relations. Note that you have to delete all rules for Event 2 to leave it completely open. This search will show all events caused by increased temperature.

More help?

Here is a slide presentation that contains a number of detailed illustrations on how to use the MVL interface: PDF.

About

Credits

The Marine Variable Linker has been developed at Department of Computer and Information Science of the Norwegian University of Science and Technology (NTNU) within the Ocean Certain project, Work Package 1: Data Mining and Knowledge Discovery. Financial aid from the European Commission (OCEAN-CERTAIN, FP7-ENV-2013-6.1-1; no: 603773) is gratefully acknowledged.

Contact

For more information, please contact For more information regarding text mining please contact Erwin Marsi (emarsi@idi.ntnu.no) or Pinar √ėzturk (pinar@idi.ntnu.no) at the Norwegian University of Science and Technology (NTNU). For questions about Ocean-Certain in general, contact Murat Van Ardalan (murat.v.ardelan@ntnu.no).