import numpy as np 
import jieba 
import jieba.posseg as pseg 
 
class TextRank(object): 
   
  def __init__(self, sentence, window, alpha, iternum): 
    self.sentence = sentence 
    self.window = window 
    self.alpha = alpha 
    self.edge_dict = {} #記錄節(jié)點(diǎn)的邊連接字典 
    self.iternum = iternum#迭代次數(shù) 
 
  #對句子進(jìn)行分詞 
  def cutSentence(self): 
    jieba.load_userdict('user_dict.txt') 
    tag_filter = ['a','d','n','v'] 
    seg_result = pseg.cut(self.sentence) 
    self.word_list = [s.word for s in seg_result if s.flag in tag_filter] 
    print(self.word_list) 
 
  #根據(jù)窗口，構(gòu)建每個節(jié)點(diǎn)的相鄰節(jié)點(diǎn),返回邊的集合 
  def createNodes(self): 
    tmp_list = [] 
    word_list_len = len(self.word_list) 
    for index, word in enumerate(self.word_list): 
      if word not in self.edge_dict.keys(): 
        tmp_list.append(word) 
        tmp_set = set() 
        left = index - self.window + 1#窗口左邊界 
        right = index + self.window#窗口右邊界 
        if left < 0: left = 0 
        if right >= word_list_len: right = word_list_len 
        for i in range(left, right): 
          if i == index: 
            continue 
          tmp_set.add(self.word_list[i]) 
        self.edge_dict[word] = tmp_set 
 
  #根據(jù)邊的相連關(guān)系，構(gòu)建矩陣 
  def createMatrix(self): 
    self.matrix = np.zeros([len(set(self.word_list)), len(set(self.word_list))]) 
    self.word_index = {}#記錄詞的index 
    self.index_dict = {}#記錄節(jié)點(diǎn)index對應(yīng)的詞 
 
    for i, v in enumerate(set(self.word_list)): 
      self.word_index[v] = i 
      self.index_dict[i] = v 
    for key in self.edge_dict.keys(): 
      for w in self.edge_dict[key]: 
        self.matrix[self.word_index[key]][self.word_index[w]] = 1 
        self.matrix[self.word_index[w]][self.word_index[key]] = 1 
    #歸一化 
    for j in range(self.matrix.shape[1]): 
      sum = 0 
      for i in range(self.matrix.shape[0]): 
        sum += self.matrix[i][j] 
      for i in range(self.matrix.shape[0]): 
        self.matrix[i][j] /= sum 
 
  #根據(jù)textrank公式計算權(quán)重 
  def calPR(self): 
    self.PR = np.ones([len(set(self.word_list)), 1]) 
    for i in range(self.iternum): 
      self.PR = (1 - self.alpha) + self.alpha * np.dot(self.matrix, self.PR) 
 
  #輸出詞和相應(yīng)的權(quán)重 
  def printResult(self): 
    word_pr = {} 
    for i in range(len(self.PR)): 
      word_pr[self.index_dict[i]] = self.PR[i][0] 
    res = sorted(word_pr.items(), key = lambda x : x[1], reverse=True) 
    print(res) 
 
if __name__ == '__main__': 
  s = '程序員(英文Programmer)是從事程序開發(fā)、維護(hù)的專業(yè)人員。一般將程序員分為程序設(shè)計人員和程序編碼人員，但兩者的界限并不非常清楚，特別是在中國。軟件從業(yè)人員分為初級程序員、高級程序員、系統(tǒng)分析員和項(xiàng)目經(jīng)理四大類。' 
  tr = TextRank(s, 3, 0.85, 700) 
  tr.cutSentence() 
  tr.createNodes() 
  tr.createMatrix() 
  tr.calPR() 
  tr.printResult()