Interactive Trend Detection

Exponentially growing patent databases demand the urge of automatic patent analysis or the help of machines to accelerate the process while offering reliable results. One major use of patent analysis is to identify trending topics running under patent pools. Nonetheless there are yet obstacles to accurately discover such trends in technology, uses of devices or other desired criterion of the interest.
This study studies the current issues on trend detection in patent datasets and tackle some major problems in particular feature selection from patent data by designing an interactive system that offers the very professional yet trending features under the studied topic and the given dataset. This helps the user build data-relevant queries more confidently or use other methods to discover trends such as clustering and relevance
feedback.
The author believes incorporating the extensive and valuable knowledge of the expert users of patent analysis systems in one of the less studied stage of feature selection make a big difference in user satisfaction as it enables them to reflect their own exceptions on the system and guide it to the type of information they seek.
The system kicks off by the desired criterion of the user to reduce the patents to their relevant passages. It then organizes trending topics of the given patent dataset into a cooccurrencefeature network and expose it layer by layer to the user providing interactive tools for further communication. The results can be taken into different pipelines of classifiers, IRS or a PRF trend analysis system to produce the output the user wishes to investigate. The results are assessed against a benchmark designed by the WIPO reports queries of five topics on EPFULL.