CSIC2010学习Word2vec表示及可视化

1、sudo apt-get install liblapack-dev
2、sudo apt-get install gfortran
3、sudo apt-get install python-pandas
4、sudo pip install --upgrade gensim
5、sudo pip install jieba
6、sudo pip install theano (0.7)

根据给定词生成word2vec词向量

# -*- coding: utf-8 -*-
"""
Created on Thu Jun 15 16:24:01 2017

@author: Jiabao Wang
@description: Generate word2vec model based on given words
"""

import gensim.models.word2vec as w2v  

def train_model(input_file_name, model_file_name):
    #模型训练，生成词向量  
    sentences = w2v.LineSentence(input_file_name)  
    model = w2v.Word2Vec(sentences, size=20, window=5, min_count=5, workers=4)   
    model.save(model_file_name)

input_file_name = 'wPred_word.txt' # Input Words
model_file_name = 'wPred_model.txt' # Output Model
train_model(input_file_name, model_file_name)

# Compute and evaluate similarity and probability
model = w2v.Word2Vec.load(model_file_name)
print model.similarity('eval','@')
for k in model.similar_by_word('eval'):
    print str(k[1])+"\t# "+k[0].decode('utf-8')

可视化词向量

# -*- coding: utf-8 -*-
"""
# This is the visualization for the embedding word vectors:
# Input: the words for visualization, the words labels, and the word2vec model
# Output: the visualization of the given words

Created on Thu Jun 22 01:55:37 2017

@author: Jiabao Wang
"""


import numpy as np
from gensim.models.word2vec import Word2Vec
import matplotlib.pyplot as plt
#import sklearn.manifold.TSNE as tsne

modelpath = 'pub_data/wPred_model.txt' # 词向量模型
model = Word2Vec.load(modelpath)
sentenceFilePath = 'pub_data/wordList.txt' # 可视化词的词典
labelFilePath = 'pub_data/wordName.txt' # 可视化词对应显示名称

visualizeVecs = []
with open(sentenceFilePath, 'r') as f:
    for line in f:
        word = line.strip()
        vec = model[word]
        visualizeVecs.append(vec)

visualizeWords = []
with open(labelFilePath, 'r') as f:
    for line in f:
        word = line.strip()
        visualizeWords.append(word)

visualizeVecs = np.array(visualizeVecs).astype(np.float64)
#Y = tsne(visualizeVecs, 2, 200, 20.0);
## Plot.scatter(Y[:,0], Y[:,1], 20,labels);
## ChineseFont1 = FontProperties(‘SimHei‘)
#for i in xrange(len(visualizeWords)):
#    # if i<len(visualizeWords)/2:
#    #     color=‘green‘
#    # else:
#    #     color=‘red‘
#    color = 'red'
#    plt.text(Y[i, 0], Y[i, 1], visualizeWords[i],bbox=dict(facecolor=color, alpha=0.1))
#plt.xlim((np.min(Y[:, 0]), np.max(Y[:, 0])))
#plt.ylim((np.min(Y[:, 1]), np.max(Y[:, 1])))
#plt.show()


# vis_norm = np.sqrt(np.sum(temp**2, axis=1, keepdims=True))
# temp = temp / vis_norm
temp = (visualizeVecs - np.mean(visualizeVecs, axis=0))
covariance = 1.0 / visualizeVecs.shape[0] * temp.T.dot(temp)
U, S, V = np.linalg.svd(covariance)
coord = temp.dot(U[:, 0:2])
for i in xrange(len(visualizeWords)):
    print i
    print coord[i, 0]
    print coord[i, 1]
    color = 'red'
    plt.text(coord[i, 0], coord[i, 1], visualizeWords[i], bbox=dict(facecolor=color, alpha=0.1),
             fontsize=12)  # fontproperties = ChineseFont1
plt.xlim((np.min(coord[:, 0])-5, np.max(coord[:, 0])+5))
plt.ylim((np.min(coord[:, 1])-5, np.max(coord[:, 1])+5))
plt.savefig('pub_data/distrubution.png', format='png',dpi = 1000,bbox_inches='tight')
plt.show()

可视化效果如下：

图中间部分的词为SQL攻击的关键词，相对其他词更加聚集。