Enter your comment...In my opinion, any scientist will strive to have full knowledge of the given scientific field. This implies that any "killer app" must provide the kill in terms of providing information from the totality of the relevant literature. This is of course far from the case for the Biomed Central corpus and similar.

There is one obvious effort that should receive the full attention from academia, and that is to provide the publishers with open source text tagging formats and software. As any paper will necessarily need to go through the proofing process, which is a part of the production line of the publisher, tagging should be a part of the process, and the tags proofed by each scientist. This is in my opinion both realistic and feasible to implement. Given the text mining efforts, still producing F-scores of dubious quality for a number of tasks, money would be well spent to develop such software. Many journals are of course still not open access, but providing the service across the board, would stimulate open access further.

The tagging for items such as genes and proteins, interactions, chemicals, gene ontologies etc. could be provided by international bodies like the Gene Ontology consortium, and would stimulate this development further along.

This would really be the killer, and everything else would follow automagically.