Metadata Based Figure Search Engine for US Patent User Interface
Description/Abstract/Artist Statement
Most online search engines support searching for text-based documents, such as web pages. In this project, we investigate an approach to build a search engine that allows users to search figures in patents. Besides keyword-based searching, the user interface (UI) offers a set of functionalities allowing users to tag, create customized lists, like patent figures, and perform advanced searches based on metadata of figures, such as patent ID, text reference, figure ID, aspect, and the object depicted in the figure. The metadata is queried by JSON objects that are indexed by the state-of-the-art search platform called Elasticsearch. The interface and functionality are implemented with HTML, CSS, JavaScript, PHP, AJAX, and MySQL. The UI was tested and the items in the search engine result page return relevant patent figures. Such UI can be used for building a large corpus of multi-labeled figures, which can further be used for training and evaluating computer vision models for automatic object recognition.
Faculty Advisor/Mentor
Jian Wu
College Affiliation
College of Sciences
Presentation Type
Poster
Disciplines
Databases and Information Systems | Graphics and Human Computer Interfaces | Software Engineering
Session Title
Engineering Research #1
Location
Zoom Room K
Start Date
3-20-2021 10:00 AM
End Date
3-20-2021 10:55 AM
Metadata Based Figure Search Engine for US Patent User Interface
Zoom Room K
Most online search engines support searching for text-based documents, such as web pages. In this project, we investigate an approach to build a search engine that allows users to search figures in patents. Besides keyword-based searching, the user interface (UI) offers a set of functionalities allowing users to tag, create customized lists, like patent figures, and perform advanced searches based on metadata of figures, such as patent ID, text reference, figure ID, aspect, and the object depicted in the figure. The metadata is queried by JSON objects that are indexed by the state-of-the-art search platform called Elasticsearch. The interface and functionality are implemented with HTML, CSS, JavaScript, PHP, AJAX, and MySQL. The UI was tested and the items in the search engine result page return relevant patent figures. Such UI can be used for building a large corpus of multi-labeled figures, which can further be used for training and evaluating computer vision models for automatic object recognition.