如何合并和删除networkx图中的节点？

Question

我使用 networkx 制作了一个图表，该图表具有三个属性（id、publication_authors、publication title）。 我想检查节点属性（作者），如果两个节点之间的公共节点属性大于一个，我想添加这两个节点属性并从图中删除第二个节点。例如，我有以下图表：

[(0, {'publication_authors': {'Bob Johnson', 'Stephen Michell', 'Andy J. Wellings', 'Jorg Kienzle', 'Thomas Wolf', 'Bo Sanden'}, 'title': 'Object-Oriented Programming and Protected Objects in Ada 95'}), (1, {'publication_authors': {'Bob Johnson'}, 'title': 'UNIX Metrics: Is The Data In Open Systems The Same From Platform To Platform?'}), (2, {'publication_authors': {'Bob Johnson'}, 'title': 'Triad Of Computing In The 21st Century Or, Back To The Future Again'}), (3, {'publication_authors': {'Bob Johnson'}, 'title': 'User-centeredness, situatedness, and designing the media of computer documentation'}), (4, {'publication_authors': {'Bob Johnson'}, 'title': "Introduction to commentaries on 'Spurious Coin: A History of Science, Management, and Technical Writing' by Bernadette Longo"}), (5, {'publication_authors': {'Brian Lawrence', 'Bob Johnson'}, 'title': 'Manager: The Project Scoping Gamble'}), (6, {'publication_authors': {'Bob Johnson', 'Stephen Michell', 'Andy J. Wellings', 'Jorg Kienzle', 'Thomas Wolf', 'Bo Sanden'}, 'title': 'Integrating object-oriented programming and protected objects in Ada 95'}), (7, {'publication_authors': {'Robert Johnson', 'Michael Hackett', 'Bob Johnson', 'Hung Quoc Nguyen'}, 'title': 'Testing Applications on the Web: Test Planning for Mobile and Internet-Based Systems, 2 edition'}), (8, {'publication_authors': {'Bob Johnson'}, 'title': 'The Wired Neighborhood: An Extended Multimedia Conversation'}), (9, {'publication_authors': {'Bob Johnson'}, 'title': 'Introduction to the book commentaries'}), (10, {'publication_authors': {'Bob Johnson'}, 'title': 'The cult of ISDN'})]

我想获取第一个节点并将其与剩余的 10 个节点进行比较，如果任何两个节点（0、6）之间的“publication_authors”大于 1，那么我想修改第一个节点（在本例中为 0）并将节点6的属性合并为0。然后从图中删除节点6。我已经实现了以下代码，但它给了我错误。请有人帮我改正。提前致谢。

import networkx as nx

ground_truth_file = 'C:\\Bob Johnson.txt'
G = nx.DiGraph()


f = open(ground_truth_file, mode='r')
lines = f.readlines()
i=0

for line in lines:
    line.strip()
    pub_authors = set()
    tokens = line.split('<>')
    authors = tokens[1]
    title = tokens[2]
    venue = tokens[3]
    num_of_authors = authors.split(',')
    for author in num_of_authors:
        pub_authors.add(author)
    G.add_node(i,publication_authors=pub_authors, title=title)
    i=i+1
num_nodes =G.number_of_nodes()
for node in range (num_nodes-1):
    for next_node in range (node+1,num_nodes):
        a = G.node[node]['publication_authors']
        b = G.node[next_node]['publication_authors']
        common_authors = a.intersection(b)
        if (len(common_authors)>1):
            c = G.node[node]['title']
            d = G.node[next_node]['title']
            cluster_authors = a.union(b)
            cluster_title = c+d
            G.node[node]['publication_authors']=cluster_authors
            G.node[node]['title']=cluster_title
            G.remove_node(next_node)
        else:
            print ('Not enough common authors')

print(G.nodes(data=True))`

我的txt文件是

0<>Andy J. Wellings,Bob Johnson,Bo Sanden,Jorg Kienzle,Thomas Wolf,Stephen Michell<>Object-Oriented Programming and Protected Objects in Ada 95<>Ada-Europe<>2000<>null
1<>Bob Johnson<>UNIX Metrics: Is The Data In Open Systems The Same From Platform To Platform?<>Int. CMG Conference<>1995<>null
1<>Bob Johnson<>Triad Of Computing In The 21st Century Or, Back To The Future Again<>Int. CMG Conference<>1995<>null
2<>Bob Johnson<>User-centeredness, situatedness, and designing the media of computer documentation<>SIGDOC<>1990<>Miami University of Ohio
3<>Bob Johnson<>Introduction to commentaries on 'Spurious Coin: A History of Science, Management, and Technical Writing' by Bernadette Longo<>ACM Journal of Computer Documentation<>2001<>Michigan Technological University
4<>Brian Lawrence,Bob Johnson<>Manager: The Project Scoping Gamble<>IEEE Software<>1997<>null
0<>Andy J. Wellings,Bob Johnson,Bo Sanden,Jorg Kienzle,Thomas Wolf,Stephen Michell<>Integrating object-oriented programming and protected objects in Ada 95<>ACM Trans. Program. Lang. Syst.<>2000<>null
5<>Hung Quoc Nguyen,Bob Johnson,Robert Johnson,Michael Hackett<>Testing Applications on the Web: Test Planning for Mobile and Internet-Based Systems, 2 edition<>null<>2002<>null
2<>Bob Johnson<>The Wired Neighborhood: An Extended Multimedia Conversation<>ACM SIGDOC Asterisk Journal of Computer Documentation<>1997<>Miami University, Oxford, OH
2<>Bob Johnson<>Introduction to the book commentaries<>ACM SIGDOC Asterisk Journal of Computer Documentation<>1998<>Miami University, Oxford, OH
6<>Bob Johnson<>The cult of ISDN<>PC/Computing<>1989<>null

Answer 1

我认为

nx.contracted_nodes

是你需要的功能。

https://networkx.github.io/documentation/networkx-1.10/reference/generated/networkx.algorithms.minors.contracted_nodes.html

Answer 2

如果我理解正确，您希望递归合并节点，直到边缘之间没有重叠。我的想法是从一个完全连接的图开始，然后“递归地”合并节点。这是一个“假”递归实现。编辑：我不太清楚为什么 networkx 在这里是必要的，它可以用 dicts 来完成（也许更清楚）。

import networkx as nx

# file you provided
with open('temp.txt', 'r') as f:
    lines = f.readlines()



nodes = {}
for idx, line in enumerate(lines):
    authors, title, venue = line.split('<>')[1:4]
    authors = set(authors.split(','))
    nodes[idx] = dict(authors = authors, title = (title, ))

G = nx.complete_graph(len(nodes))
nx.set_node_attributes(G, nodes)


def merge_recursive(G, target = 'authors'):
    """
    Keeps merging if there is overlap between nodes
    """
    # check edges
    while G.edges():
        for (i, j) in G.edges():
            overlap = G.nodes()[i][target].intersection(G.nodes()[j][target])
            # copy values
            if overlap:
                tmp = {}
                for k, v in G.nodes()[i].items():
                    if isinstance(v, set):
                        tmp[k] = v.union(G.nodes()[j][k])
                    else:
                        tmp[k] = v + G.nodes()[j][k]

                nx.set_node_attributes(G, {i: tmp})
                G.remove_node(j)
            # no overlap remove edge
            else:
                G.remove_edge(i, j)
            break
    return G

merged = merge_recursive(G.copy())

from matplotlib.pyplot import subplots

fig, (left, right) = subplots(1, 2, figsize = (10,  5))
nx.draw(G, ax = left, with_labels = 1)
nx.draw(merged, ax = right, with_labels = 1)

left.set_title('Before merging')
right.set_title('After merging')
fig.show()