# Twitter Neighbours dataset

This repository contains the dataset assembled as part of a APPRAISE project (H2020-SU-SEC-2020H2020-SU-SEC-2020)

# Description
The dataset contains a sampled graph, extracted from Twitter, starting from a list of seed users.
Each user is initially represented by a semantic embedding (vector) computed as the average text embedding of a sample of its tweets, while users are connected with each other through the 'following/follower' property on Twitter.

# Statistics
- number of  users: 36122
- number of  edges: 84026
- user initial embeddings size: 768

# Files:
graph_train_and_test.pygeodata: is a compact representation of the graph for PyG usage.

twitter_neighs_graph.json: a dictionary containing:
  - 'adj_sparse': adjacency matrix in sparse representation
  - 'user_init_embs': initial user embedding, computed as average of text embedding of user tweets
  - 'train_test_split': 0/1 1D list, where 0 represents the user is in teh training set and 1 the user is in the test set