This article presents the Semantic Vector Space Model
based system is at least as good as, and sometimes better (SVSM), a text representation and searching technique than, conventional IR systems in an experimental environbased on the combination of Vector Space Model (VSM) ment. However, some difficulties with VSM have also with heuristic syntax parsing and distributed representabeen voiced (Cooper, 1991; Sutcliffe 1991; Wong, Zition of semantic case structures. In this model, both docarko, Raghavan, & Wong, 1987; Wong, Ziarko, & Wong, uments and queries are represented as semantic matrices. A search mechanism is designed to compute the 1985).
similarity between two semantic matrices to predict rele-
The major difficulty with VSM is its inadequacy in vancy. A prototype system was built to implement this disambiguating the meaning of terms used in natural lanmodel by modifying the SMART system and using the guage texts, which is a direct consequence of the oversim-Xerox Part-Of-Speech (P-O-S) tagger as the pre-processor of the indexing process. The prototype system was plicity of its purely term-based representation of content.
used in an experimental study to evaluate this technique
In this model, keywords are identified, taken out of conin terms of precision, recall, and effectiveness of reletext, and further processed to generate term vectors. All vance ranking. The results of the study showed that if the non-keyword terms, syntactic structures, and other documents and queries were too short (typically less linguistic elements of text are discarded. However, meanthan 2 lines in length), the technique was less effective than VSM. But with longer documents and queries, espeing is conveyed not only by keywords, but also by other cially when original documents were used as queries, linguistic elements such as functional words and syntactic we found that the system based on our technique had patterns. In natural language, rich compositional strucsignificantly better performance than SMART.