Approaches

In texts, the links between words (concepts) are implicit. Hence, extracting a network of

concepts from a text, and classifying this network via the meta-matrix ontology, requires

an inference process. The links, or relations, between concepts must be extracted based

on the semantic, syntactic, and contextual information given in a text (Carley, 1986;

Carley, 1988; Popping, 2003). Making the meta-matrix approach available for NTA can

provide analysts with a novel technique for extracting textual networks that reveal the

relationships within and between the elements that compose a network and that were

classified a priori according to the meta-matrix model. The features of the textual data

that are relevant to the analyst can then be represented as a network structure of the meta-

Table 2: Original meta-matrix conceptualization

Based on Carley (2002, 2003)

Meta-Matrix entities People Knowledge/

Resources

Events/

Tasks

Groups/

Organizations

People Social network Knowledge

Network/

Resource

Network

Attendance

Network/

Assignment

Network

Membership

network

Knowledge/Resources Information

Network/

Substitution

Network

Needs

network

Organizational

capability

Events/Tasks Temporal

Ordering/

Task Flow/

Precedence

Institutional

support or

attack

Organizations Interorganizational

network

matrix entity classes and the connections between these classes. Such a network makes

the structure of social systems, which is implicitly contained in texts visible and

analyzable.

How did we combine and formalize the meta-matrix approach and map analysis technique,

which is a specific type of NTA? We utilized the meta-matrix model as an extension of

NTA in general and map analysis in specific by instantiating the following five step

procedure:

1. Concept Identification: identify the concepts in texts that are relevant to the

analyst’s research question. As part of this process, the analyst may first want to

generalize many text-level concepts into higher-level concepts.

2. Entity Identification: define an ontology for capturing the overall structure

described in the text. We use the basic meta-matrix. Other analysts may wish to

adapt this to their research question. Note, step 1 and 2 can also be done in reverse

order.

3. Concept Classification: classify the identified concepts into the relevant entity

classes in the meta-matrix. Given the vagaries of the language it may be that some

concepts need to be cross-classified in two or more entity classes.

4. Perform Map Analysis: automatically extracting the identified concepts and the

relations among them from the specified texts. This results in a map or conceptual

network. Since the concepts are classified by entity classes, the resulting concept

network is hierarchically embedded in the ontology provided by the meta-matrix.

In essence then, there are three networks. First, there is the concept network where

the nodes are concepts (many of which are higher-level concepts). Second, there

is the entity network where the nodes are the entity classes and the links are the

connections among and between the entity classes. Third, there is the network

embodied in the meta-matrix thesaurus, connecting concepts in entity classes to

concepts in the same or other entity classes. Finally, there is the network (embodied

in the meta-matrix thesaurus), connecting concepts to entity classes.

5. Graph and Analyze Data: the final step is to take the extracted data for each text,

the network, and graph and analyze it in general and by cells in the meta-matrix. As

part of this analysis, the resultant networks from different texts can be combined

and compared. Note the analysis can occur at the concept network level (map

analysis), the entire meta-matrix level (meta-matrix text analysis), and the sub-cell

level (sub-matrix text analysis).

We refer to these five steps as the method of meta-matrix text analysis. With this novel

technique we hope to contribute towards the analysis of complex, large-scale data and

social systems and providing profound multi-level access to the meaning of textual data.

We note that these steps begin to bridge the gap between NTA and a more interpretive

analysis of texts. The meaning of concepts is revealed by virtue of other concepts they

are connected to. In the meta-matrix approach, the meaning of concepts is revealed both

by what other concepts they are connected to and by what type of entity classes into

which they fall.