Mining of Association Relations in Text
Gleb Sizov · 2012
We present a method that adopts ideas from association rule mining in database (DB) systems to facilitate reuse of organizational knowledge captured in textual form. Many organizations possess large collections of textual reports representing the episodic memory of an organization, e.g. medical patient records, industrial accident reports, lawsuit records and investigation reports. Effective (re)use of expert knowledge contained in these reports may increase the productivity of the organizations. Our method may support employees in discovering information in textual reports that can be useful for dealing with a new situation or writing a new report. Association rule mining is used in DB systems to discover associations between items in database records. We set out with the hypothesis that a similar approach may be used to discover implicit relations between text units in textual reports. We developed the SmoothApriori algorithm that finds association relations between sentences which may correspond to a cause-effect type of relation or have a more implicit nature. We evaluated the ability of SmoothApriori to restore sentences that were removed, for test purpose, from air investigation reports. SmoothApriori restored significantly more information than any of our baselines. 1