proj-444's Introduction

Proj-444

Abstract

These are codes for course project of VE444 Networks, UM-SJTU JI, partially adapted from CS224 Machine Learning with Graphs, Standford. In this prohect, we are exploring on network methods for online video recommendation. There are basically 2 ways under consideration:

Link prediction. For a video entry, creating new relevant videos based on existing ones.
Social recommendation. Precit a user's preferences based on the preferences of those who have relationships with him/her.

dataset: contains all the data used in his project. All are of Python dict class stored as json files.
- following: following network of users. (Actually not used)
- user2video & users_all.json: 1344 subscribed bangumis of 492 users, totally 4559 subscriptions. (for social recommendation & link predition)
- video2video_bfs.json & video_all_bfs.json: the video recommendation network of 843 videos, totally 1060 recommendations. (for link prediction)
generation1: scripts to collect Bilibili user and video info. Information is collected by Breath-first-search (BFS).
linkpred: scripts and notebooks of link predictions. Linkpred is run on
- User-video bipartite graph
- Whole video reommendation graph
social: scripts and notebooks of social recommendation.

Acknowledgement

Some key python libraries:

bilibili_api: APIs to collect informations from Bilibili.
NetworkX: A library for studying graphs and networks. It offers a wide range of graph and network-related classes and algorithms.
scikit-learn (sklearn): Simple and efficient tools for machine learning and data analysis.
node2vec Python implementation for node2vec algorith.

Recommend Projects