The network extraction driven by a text-based stimulus
If both the network and the text are structures built from words, then we may look for an efficient algorithm that can identify in the text the stimulus word used in an experiment performed to build a network and a reasonable number of direct associations with this stimulus. Words identified in the text may serve as the starting point to extract a sub-graph from the network, which will contain as many associations as possible. The semantic relationship between the nodes of a returned sub-graph will be the subject of an evaluation.
In more technical language, the algorithm should take a graph (association network) and the subset of its nodes identified in a text (extracting nodes) as input. Then, the algorithm creates a sub-graph with all extracting nodes as an initial node set. After that, all the connections between the extracting nodes which exist in the network are added to the resulting sub-graph - these connections are said to be direct. Finally, every direct connection is checked in the network to determine whether it can be replaced with a shortest path, i.e. a path which has a path weight lower than the weight of the direct connection and a node number less than or equal to the predefined path length. If such a path is found, it is added to the subgraph - which means adding all the path’s nodes and connections. If we apply this procedure to each text of a large text collection, and if we merge the resulting text sub-graphs, we may evaluate the sub-graph created for a particular stimulus word.