Prior Art Search in Chemistry Patents Based On Semantic Concepts and Co-Citation Analysis

Prior Art Search is a task of querying and retrieving the patents in order to uncover any knowledge existing prior to the inventor’s question or invention at hand. For addressing this task, we present a contemporary approach that has been evaluated during Trecchem for its ability to adapt to text containing chemistry-based information. The core of the framework is an index of 1.3 million chemistry patents provided as a data set by Trecchem. For the prior art search task, the information of normalized noun phrases, biomedical and chemical entities are added to the full text index. Altogether, 7 runs were submitted for this task that were based on automatic querying with tokens, noun phrases and entities. In addition, the co-citation information was exploited in a systematic way to generate ranked citation sets from the retrieved documents. Querying with noun phrases and entities coupled with co-citation based post-processing performed considerably well with the best MAP score of 0.23.