This is a prototypical demonstration of the text-search capabilities of the CEPT-Retina technology. For each document in a collection (4M english Wikipedia documents in this case) a document-fingerprint is computed. A user-input (word, text-snippet or even a whole document) is converted on the fly in its corresponding fingerprint.
The system now finds the most similar documents by comparing the fingerprints. Please note that this is a plain comparison not taking into account the length or structure of the documents. Therefore large and rich docs will show many (often different) topics and will match only partially. Small docs or single words will of course lack context and the system will most probably respond with a very divergent set of findings.
Nevertheless, this demo shows that CEPT-technology allows to search for text without the use of keyword-matching, because only contexts are compared and not keyword-strings.