UROP Proceedings 2020-21

School of Engineering Department of Computer Science and Engineering 97 AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN Gary Shueng Han / CSE Student: CHIU Ka Ho / COSC Course: UROP2100, Fall As crowd-source WiFi indoor fingerprinting has its unique importance in applications such as monitoring people’s locations and indoor localization, related methods are worth studying. This report would focus on its usage in indoor localization, introduce one proposed method and focus more on improvement details. Wi-Fi fingerprinting has two phases, survey and query, but this report would assume that the survey phase is done and some labeled locations with exact coordinates and RSSI vectors are obtained. The focus would be on the second phase, which is to get a model for predicting the target position based on those obtained labels. The model would be based on the similarities, or proximities between nodes, and is divided into two parts, the first part would perform network embedding for feature extraction, The second part would be a springlike system with carefully designed loss to enforce various constraints to further project the nodes from lower-dimensional embedding space to 2 dimensional physical space. This report would focus more on the second phase of the project, and emphasize a number of implementations to improve the final results and some outcomes of trials. AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN Gary Shueng Han / CSE Student: JIANG Tianrui / COMP Course: UROP2100, Fall With the rapid development of modern traffic system, the data collected has become much more diverse and abundant, providing the potential for deeper and more precise analysis, prediction, as well as upgraded recommendation system. In particular, the demand of flow data is currently a hot topic, as there remains a large room for improvement on current solutions. An important category of methods are based on origindestination pairs, which basically have better performance than zoon-based methods with the cost of additional complexity. Inspired by the prevalence of attention mechanism in the area of natural language processing and computer vision, this project seeks to discover a way that integrates it into origin-destination prediction. My work involves researching relevant works and participating in the implementation of experiments for control-group methods. The proposed method is not finalized yet, and needs further enhancement.

RkJQdWJsaXNoZXIy NDk5Njg=