Designing and Implementing hundreds of cool mobility-related features for User Profiling
Details
| Supervisor: | Shichang Ding,Please contact sding@gwdg.de or scdingwork@gmail.com |
| Duration: | at least 3 MONTH, actually it includes several phases, the first phases begins at April 1st and ends at May 15th. |
| Type: | Student Project (plus potential Master's thesis) |
| Status: | closed (for students who are interested in new topics, contact sding@gwdg.de) |
Title: Mobility Feature Engineering for User Profiling (为用户画像进行移动数据的特征工程)
What to do: Designing and Implementing many cool mobility-related features for User Profiling
Why important: User Profiling is an important research area, especially for online-shopping and digital banks (check Webank or 微众银行). Human Mobility data reflects an important part of people's lifestyle. If we can predict user characteristics based on mobility, which can help lots of organizations.
Requirements: Python, interesting in Machine Learning (ML) and data mining (DM)
Do not worry if you do not have too much knowledge of Feature Engineering: 90% of the designing work is already done. So at first, you are mainly responsible for implementing, then you can try to design by your own idea. I think it is an easier way for students to learn how to design features (and why it works).
WHY you should pick this project: For anyone who is interested in ML, you will soon find out Feature Engineering is one of the most important techniques for ML. However, it is not fully discussed in most ML books because it highly depends on area knowledge,'real' projects and 'real' data. So this is really a good chance. What's more, there are chances for paper publishing.
Further task: if faster than expected, we can try to design autoencoder to automatically extract features, and compare it with handy-craft features.
WHY you should not pick this project: if you do not like coding, emm...; if you only want to focus at Computer Vision (CV), no CV work included in the next several months.
About Deep Learning (DL): This project may contain CNN and transformers designing if the feature engineering work is done faster than expected. If not, the major part of this project will be done based on GDBT and LSTM.
The work is predominantly targeted at an approximately 3-month student project, but can be extended into a Master's thesis as well.