APPROACH TO IDENTIFY SCI COVERED PUBLICATIONS WITHIN NON-PATENT REFERENCES IN PATENTS

In order to evaluate approaches for identifying SCI covered publications within NPRs, I introduce a simple method that uses two key indicators, recall and precision, to evaluate the relevance of information retrieval systems. There are two primary reasons that conventional methods of evaluating matching results are insufficient: there is nothing in place to evaluate accuracy, and there is a direct dependence on the intermediate outcome. The proposed approach consists of five main steps: 1) data collection, 2) creation of supervised data and test data, 3) selection and execution of matching algorithms, 4) evaluation of algorithms and optimization of their combinations, and 5) evaluation of optimized combinations. A comparison of the proposed and conventional methods show that the proposed approach works well; its results (i.e., 99% precision and 69% recall) are better than the target implicitly set in a previous study (Tomizawa, 2008). In that sense, the proposed approach is quite promising.