Projects

Historical Map Text Label Connection [paper, poster,slides]

• An automatic algorithm that combined single words recognized from historical maps into meaningful phrases, which represent complete location descriptions and can be used to link historical sites to other datasets. This algorithm is potential to improve digital map processing by increasing the automation of text extraction on maps.
• Characterized map labels by location size and text content using Support Vector Machines. Built the back-end with Python and Postgres.
• Wrote an evaluation program on Strabo, a semi­automatic system that extracts and recognizes text labels in map images into text and integrated tesseract into Strabo.

An Uncertainty Aware Method for Geographic Data Conflation [paper]

• Analyzed uncertainty generated in the vector to vector conflation of geospatial sources.
• Our algorithm is unique in incorporating uncertainty value from a geographical feature recognition model that using Convolutional Neural Network and centerline extraction of linear feature. The uncertainty value is commonly overlooked by current vector data generation and process. Our algorithm proves that the uncertainty value can improve the accuracy and efficiency of the conflation processcation size and text content using Support Vector Machines. Built the back-end with Python and Postgres.

Cluster analysis of automobile innovative users based on interactive innovation value

• Classified the innovative users in Automobile forums and analyzed the characteristics of different user groups. The algorithm is effective in identifying different category users based on their innovative score from large-scale of data and proposed corresponded interactive method that can get effective feedbacks to improve new product development.
• The algorithm classified users based on six characteristics: community activity, network connectivity, user experience, domain knowledge, ahead of market trend, willingness to innovate and combined Balanced Iterative Reducing and Clustering Using Hierarchies (BIRCH) algorithm with Agglomerative Nesting (AGENES) algorithm for clustering.