From A Property Of The Average Of Fractions To A Text-Processing Interface

The average of two natural numbers always falls between those two numbers. Partitioning a document set into two non overlapping subsets, the words in the set will appear only in one subset or on both. These properties can be used to present users with choices that can allow them to build a phrase where chosen terms have context. The average frequency of a term can be used to study relevance by comparing it to the same term's average in the relevant and the non-relevant subsets. A coefficient of variability (VAR) is defined as the normalized distance between these two values. The vast majority of words seem not to be significant because VAR as their relative comparative value is minimal. But a few words show very high values. This could be exploited by a system to find strong word instances that represent relevant concepts. It is possible to imagine an iterative procedure based on these properties through which a user identifies significant words with high VAR values. Such procedure would be desirable for a diversity of text-related computer-based tasks such as content analysis, thesauri construction, data mining, computer-based indexing and feature selection. A software instance to help users build context has been developed as a prototype to show the concept. Knowledge is always related to a given context and requires a support structure which has some cognitive elements such as other knowledge, data, concepts, information, etc. Using objective and subjective measures, users derive conceptual relationships. Users gauge and build topical relevance by engaging the system, which can then offer more suggestions. There is great advantage in reducing cognitive load and extraneous information. Users deal directly with easier to identify words, phrases and their combinations to form information capsules.