def integrate_graphs(mention_graph, retweet_graph, node_to_id, restart_probability, number_of_threads):
"""
A bit of post-processing of the graphs to end up with a single aggregate graph.
Inputs: - mention_graph: The mention graph as a SciPy sparse matrix.
- retweet_graph: The retweet graph as a SciPy sparse matrix.
- user_lemma_matrix: The user lemma vector representation matrix as a SciPy sparse matrix.
- number_of_threads:
Outputs: - adjacency_matrix: An aggregate, post-processed view of the graphs.
- node_to_id: A node to Twitter id map as a python dictionary.
- features: The graph structure proximity features as calculated by ARCTE in scipy sparse matrix format.
- node_importances: A vector containing node importance values.
"""
# Form the adjacency matrix.
adjacency_matrix = 0.25*mention_graph +\
0.25*mention_graph.transpose() +\
0.25*retweet_graph +\
0.25*retweet_graph.transpose()
# Here is where I need to extract connected components or something similar.
adjacency_matrix, node_to_id, old_node_list = extract_connected_components(adjacency_matrix, "weak", node_to_id)
# Extract features
features = arcte(adjacency_matrix=adjacency_matrix,
rho=restart_probability,
epsilon=0.00001,
number_of_threads=number_of_threads)
node_importances = calculate_node_importances(adjacency_matrix)
return adjacency_matrix, node_to_id, features, node_importances
def integrate_graphs(mention_graph,
retweet_graph,
user_lemma_matrix,
node_to_id,
lemma_to_attribute,
restart_probability,
number_of_threads):
"""
A bit of post-processing of the graphs to end up with a single aggregate graph.
Inputs: - mention_graph: The mention graph as a SciPy sparse matrix.
- retweet_graph: The retweet graph as a SciPy sparse matrix.
- user_lemma_matrix: The user lemma vector representation matrix as a SciPy sparse matrix.
- number_of_threads:
Outputs: - adjacency_matrix: An aggregate, post-processed view of the graphs.
- node_to_id: A node to Twitter id map as a python dictionary.
- features: The graph structure proximity features as calculated by ARCTE in scipy sparse matrix format.
- node_importances: A vector containing node importance values.
"""
text_graph = make_text_graph(user_lemma_matrix,
dimensionality=50,
metric="angular",
number_of_estimators=5,
number_of_neighbors=3)
# Form the adjacency matrix.
adjacency_matrix,\
laplacian_matrix = graph_fusion_directed(adjacency_matrix_list=[mention_graph, retweet_graph, text_graph],
weights=[1.0, 1.0, 1.0],
fusion_type="zhou",
laplacian_type="directed")
# Here is where I need to extract connected components or something similar.
adjacency_matrix, node_to_id, old_node_list = extract_connected_components(adjacency_matrix, "weak", node_to_id)
# Extract features
features = arcte(adjacency_matrix=adjacency_matrix,
rho=restart_probability,
epsilon=0.00001,
number_of_threads=number_of_threads)
node_importances = calculate_node_importances(adjacency_matrix)
return adjacency_matrix, node_to_id, features, node_importances, old_node_list
def weakly_connected_graph(full_graph_folder, weakly_connected_graph_folder):
# Read relevant data.
mention_graph = load_pickle(full_graph_folder + "/mention_graph" + ".pkl")
mention_graph = spsp.coo_matrix(spsp.csr_matrix(mention_graph))
retweet_graph = load_pickle(full_graph_folder + "/retweet_graph" + ".pkl")
retweet_graph = spsp.coo_matrix(spsp.csr_matrix(retweet_graph))
user_lemma_matrix = load_pickle(full_graph_folder + "/user_lemma_matrix" + ".pkl")
user_lemma_matrix = spsp.coo_matrix(spsp.csr_matrix(user_lemma_matrix))
user_id_set = load_pickle(full_graph_folder + "/user_id_set" + ".pkl")
node_to_id = load_pickle(full_graph_folder + "/node_to_id" + ".pkl")
# Extract weakly connected graph for the mention graph.
weakly_connected_men_ret_graph, weakly_connected_node_to_id, old_node_list = extract_connected_components(
spsp.coo_matrix(spsp.csr_matrix(mention_graph + retweet_graph)), "weak", node_to_id
)
# Calculate the user twitter id set for the weakly connected component.
weakly_connected_user_id_set = set(list(weakly_connected_node_to_id.values()))
node_array = np.array(old_node_list, dtype=np.int64)
# Extract corresponding retweet graph and user lemma matrix.
weakly_connected_mention_graph = submatrix_pull_via_networkx(
spsp.coo_matrix(mention_graph), node_array, directed=True
)
weakly_connected_retweet_graph = submatrix_pull_via_networkx(
spsp.coo_matrix(retweet_graph), node_array, directed=True
)
user_lemma_matrix = spsp.csr_matrix(user_lemma_matrix)
weakly_connected_user_lemma_matrix = user_lemma_matrix[node_array, :]
# Change sparse matrices to coordinate format in order to save as an edge list.
weakly_connected_mention_graph = spsp.coo_matrix(weakly_connected_mention_graph)
weakly_connected_retweet_graph = spsp.coo_matrix(weakly_connected_retweet_graph)
weakly_connected_user_lemma_matrix = spsp.coo_matrix(weakly_connected_user_lemma_matrix)
# Store weakly connected data.
scipy_sparse_to_csv(
weakly_connected_graph_folder + "/mention_graph.tsv",
weakly_connected_mention_graph,
separator="\t",
directed=True,
)
scipy_sparse_to_csv(
weakly_connected_graph_folder + "/retweet_graph.tsv",
weakly_connected_retweet_graph,
separator="\t",
directed=True,
)
scipy_sparse_to_csv(
weakly_connected_graph_folder + "/user_lemma_matrix.tsv",
weakly_connected_user_lemma_matrix,
separator="\t",
directed=True,
)
store_pickle(weakly_connected_graph_folder + "/user_id_set" + ".pkl", weakly_connected_user_id_set)
store_pickle(weakly_connected_graph_folder + "/node_to_id" + ".pkl", weakly_connected_node_to_id)
