关于计算文本关键字textrank技术点之窗口中的词对

不多说了自己的笔记, 只是计算一个分词组成的列表, 中窗口大小的分词对

def word_combine(words, window=2):
    if window < 2: 
        window = 2

    for x in range(1, window):
        print("\n 跨度:", x)
        if x >= len(words):
            break
            
        words2 = words[x:]
        res = zip(words, words2)
        for r in res:
            yield r
# 测试
res = word_combine(["我", "爱", "中国", "人民", "和", "共产党"], 3)
for i in res:
    print(i)

结果是:

跨度: 1
('我', '爱')
('爱', '中国')
('中国', '人民')
('人民', '和')
('和', '共产党')

 跨度: 2
('我', '中国')
('爱', '人民')
('中国', '和')
('人民', '共产党')

我们可以想象成,整个列表整体向后移动[1 : window - 1]个位置, 然后进行拉链操作

 

===========================================================

转移矩阵和重要度的计算和迭代

matrix = np.array([[0.1, 0, 0.7],
                   [0.4, 0.4, 0.2],
                   [0.01, 0.8, 0.0]])

g = np.array([[0.33], [0.33], [0.33]])

iters = 5
alpha = 0.3
for i in range(iters):
    g = alpha + (1 - alpha) * np.dot(matrix, g)
    g = g / g.sum()
    if i > iters - 10:
        print(g)
        print()

多次迭代后的结果是(这里的迭代次数较少, 不过也可以当做收敛了):

[[0.32257421]
 [0.35331457]
 [0.32411122]]

[[0.31751476]
 [0.35262363]
 [0.32986161]]

[[0.31887964]
 [0.3518058 ]
 [0.32931455]]

[[0.31888035]
 [0.35198255]
 [0.3291371 ]]

[[0.31881529]
 [0.35199036]
 [0.32919434]]