Term Extraction Using Co-occurrence in Abstract and First Claim for Patent Analysis

Patent information analysis need to extract patent terms from a vast amount of patent records rapidly and accurately. This paper proposes a patent term extraction method using co-occurrence in abstract and first claim sections of patent records. The research is carried out in three steps. In experiment 1, our proposed method´s F1 is 34.8% and higher than that of KEA with 21.0%. This demonstrates the good performance of our proposed method from a standard point of view. In experiment 2, our proposed method selects only 10,866 qualified terms from 28,044 candidate strings, with the selection percentage of 38.75%. There are 96.8% of total patent claims that contain the candidate terms at the same time. This illustrates the advantage of the proposed method in both eliminating unqualified candidates and providing a good coverage of patent texts from an actual application perspective.