CRIC: Kontextbasierte Empfehlung unstrukturierter Texte in Echtzeitumgebungen : ein Verfahren zur Bestimmung der semantischen Proximität von Textobjekten auf Basis eines heuristischen asymmetrischen Distanzmaßes

The idea to link texts with related content in an automated way is not new. The approach
developed and presented here has two main goals: to work automatically on unstructured texts
and to support a large number of parallel accesses. It is distinct from other approaches in that
it determines the semantic distance between texts on the basis of an asymmetrical precalculated
distance matrix. The relations between unstructured text objects are generated by a
language independent heuristic algorithm for feature selection. The resulting bag of words is
used in a query to select matching texts. The user receives recommendations to texts the
content of which is related to the text that appears in front of him. Performance was analyzed
on basis of the algorithms runtime complexity. Extensive real-life tests over a period of 12
months were conducted on the website of an industrial magazine in order to check the
efficiency of the procedure with regard to quality of the recommendations. Results show that
the presented approach nearly equals the quality of manual made recommendations by
professional editors.