Architectural Knowledge Management: Supporting Architects and Auditors

TLDR: To illustrate the effect of LSA on the document vector-space model, LSA was applied to the 8 documents from the audit that were still available, and both auditors seem to agree that there are two large document clusters.

Abstract: 2 1 1 2 4 5 5 5 5 5 concrete content 1 1 1 1 3 5 5 5 4 4 packing input for development 1 1 1 1 3 4 5 5 4 4 output of development descriptive / static 1 1 1 1 3 3 5 5 4 4 use / time dimension app. functionality 1 1 1 1 3 3 3 4 5 5 system administration design 1 1 1 1 3 4 3 5 5 5 deployment conceptual 1 1 1 2 2 4 3 4 5 5 concrete/instance high level 5 3 1 1 2 5 3 4 4 4 detailed absolute 1 1 1 1 3 5 2 2 2 2 relative (wrt prev. version) application 1 1 1 1 5 2 2 2 2 2 organisation used by me 4 2 1 2 5 2 1 1 3 3 not used by me analyzed, for instance to determine clusters of documents that contain similar documents. Figs. 16.6 and 16.7 depict the document clusters according to auditor 1 and 2 respectively. The clusters have been determined with the single linkage hierarchical clustering method (Johnson, 1967). The axis denotes the difference between the documents, calculated as 1 minus the similarity. For instance, for auditor 1 the similarity between documents FD and FM has been calculated as 0.87, therefore the difference between the two equals 0.13, as shown in Fig. 16.6. Although there are some differences between the two cluster configurations, both auditors seem to agree that there are two large document clusters. One cluster contains documents FD, FM, PD, and XX. The other documents are grouped in the second cluster. Note that we left AS and DB out of the figures to allow for a fair comparison with the effect of LSA. Recall that those two documents were no longer available, due to which LSA was unable to process them. Had we included them in the cluster figures, they would have shown as a small sub-cluster of two very similar documents (similarity according to auditor 1: 0.96; auditor 2: 1.00). For both auditors this sub-cluster is most similar to document IM (auditor 1: 0.79, auditor 2: 0.84). To illustrate the effect of LSA on the document vector-space model, we applied LSA to the 8 documents from the audit that were still available. We determined the