Python_LDA实现方法详解
LDA(LatentDirichletallocation)模型是一种常用而用途广泛地概率主题模型。其实现一般通过Variationalinference和GibbsSamping实现。作者在提出LDA模型时给出了其变分推理的C源码(后续贴出C++改编的类),这里贴出基于Python的第三方模块改写的LDA类及实现。
#coding:utf-8 importnumpyasnp importlda importlda.datasets importjieba importcodecs classLDA_v20161130(): def__init__(self,topics=2): self.n_topic=topics self.corpus=None self.vocab=None self.ppCountMatrix=None self.stop_words=[u',',u'。',u'、',u'(',u')',u'·',u'!',u'',u':',u'“',u'”',u'\n'] self.model=None defloadCorpusFromFile(self,fn): #中文分词 f=open(fn,'r') text=f.readlines() text=r''.join(text) seg_generator=jieba.cut(text) seg_list=[iforiinseg_generatorifinotinself.stop_words] seg_list=r''.join(seg_list) #切割统计所有出现的词纳入词典 seglist=seg_list.split("") self.vocab=[] forwordinseglist: if(word!=u''andwordnotinself.vocab): self.vocab.append(word) CountMatrix=[] f.seek(0,0) #统计每个文档中出现的词频 forlineinf: #置零 count=np.zeros(len(self.vocab),dtype=np.int) text=line.strip() #但还是要先分词 seg_generator=jieba.cut(text) seg_list=[iforiinseg_generatorifinotinself.stop_words] seg_list=r''.join(seg_list) seglist=seg_list.split("") #查询词典中的词出现的词频 forwordinseglist: ifwordinself.vocab: count[self.vocab.index(word)]+=1 CountMatrix.append(count) f.close() #self.ppCountMatrix=(len(CountMatrix),len(self.vocab)) self.ppCountMatrix=np.array(CountMatrix) print"loadcorpusfrom%ssuccess!"%fn defsetStopWords(self,word_list): self.stop_words=word_list deffitModel(self,n_iter=1500,_alpha=0.1,_eta=0.01): self.model=lda.LDA(n_topics=self.n_topic,n_iter=n_iter,alpha=_alpha,eta=_eta,random_state=1) self.model.fit(self.ppCountMatrix) defprintTopic_Word(self,n_top_word=8): fori,topic_distinenumerate(self.model.topic_word_): topic_words=np.array(self.vocab)[np.argsort(topic_dist)][:-(n_top_word+1):-1] print"Topic:",i,"\t", forwordintopic_words: printword, print defprintDoc_Topic(self): foriinrange(len(self.ppCountMatrix)): print("Doc%d:((toptopic:%s)topicdistribution:%s)"%(i,self.model.doc_topic_[i].argmax(),self.model.doc_topic_[i])) defprintVocabulary(self): print"vocabulary:" forwordinself.vocab: printword, print defsaveVocabulary(self,fn): f=codecs.open(fn,'w','utf-8') forwordinself.vocab: f.write("%s\n"%word) f.close() defsaveTopic_Words(self,fn,n_top_word=-1): ifn_top_word==-1: n_top_word=len(self.vocab) f=codecs.open(fn,'w','utf-8') fori,topic_distinenumerate(self.model.topic_word_): topic_words=np.array(self.vocab)[np.argsort(topic_dist)][:-(n_top_word+1):-1] f.write("Topic:%d\t"%i) forwordintopic_words: f.write("%s"%word) f.write("\n") f.close() defsaveDoc_Topic(self,fn): f=codecs.open(fn,'w','utf-8') foriinrange(len(self.ppCountMatrix)): f.write("Doc%d:((toptopic:%s)topicdistribution:%s)\n"%(i,self.model.doc_topic_[i].argmax(),self.model.doc_topic_[i])) f.close()
算法实现demo:
例如,抓取BBC川普当选的新闻作为语料,输入以下代码:
if__name__=="__main__": _lda=LDA_v20161130(topics=20) stop=[u'!',u'@',u'#',u',',u'.',u'/',u';',u'',u'[',u']',u'$',u'%',u'^',u'&',u'*',u'(',u')', u'"',u':',u'<',u'>',u'?',u'{',u'}',u'=',u'+',u'_',u'-',u''''''] _lda.setStopWords(stop) _lda.loadCorpusFromFile(u'C:\\Users\Administrator\Desktop\\BBC.txt') _lda.fitModel(n_iter=1500) _lda.printTopic_Word(n_top_word=10) _lda.printDoc_Topic() _lda.saveVocabulary(u'C:\\Users\Administrator\Desktop\\vocab.txt') _lda.saveTopic_Words(u'C:\\Users\Administrator\Desktop\\topic_word.txt') _lda.saveDoc_Topic(u'C:\\Users\Administrator\Desktop\\doc_topic.txt')
因为语料全部为英文,因此这里的stop_words全部设置为英文符号,主题设置20个,迭代1500次。结果显示,文档148篇,词典1347词,总词数4174,在i3的电脑上运行17s。
Topic_words部分输出如下:
Topic:0
towillandofhebetrumpsthewhatpolicy
Topic:1hewouldinsaidnotnowithmrthisbut
Topic:2fororcansomewhetherhavechangehealthobamacareinsurance
Topic:3thetothatpresidentasofusalsofirstall
Topic:4trumptowhenwithnowwererepublicanmrofficepresidential
Topic:5thehistrumpfromukwhopresidenttoamericanhouse
Topic:6atothatwasitbyissuevotewhilemarriage
Topic:7thetoofanaretheywhichbycouldfrom
Topic:8ofthestatesonevotesplannedwontwonewclinton
Topic:9inusauseforobamalawentrynewinterview
Topic:10andonimmigrationhasthattherewebsitevettingactiongiven
Doc_Topic部分输出如下:
Doc0:((toptopic:4)topicdistribution:[0.029729730.00270270.00270270.164864860.327027030.19189189
0.00270270.00270270.029729730.00270270.029729730.0027027
0.00270270.00270270.029729730.00270270.029729730.0027027
0.137837840.0027027])
Doc1:((toptopic:18)topicdistribution:[0.210.010.010.010.010.010.010.010.110.010.010.01
0.010.010.010.010.010.010.310.21])
Doc2:((toptopic:18)topicdistribution:[0.020754720.001886790.039622640.001886790.001886790.00188679
0.001886790.152830190.001886790.020754720.001886790.24716981
0.001886790.077358490.001886790.001886790.001886790.00188679
0.416981130.00188679])
当然,对于英文语料,需要排除大部分的虚词以及常用无意义词,例如it,this,there,that...在实际操作中,需要合理地设置参数。
换中文语料尝试,采用习大大就卡斯特罗逝世发表的吊唁文章和朴槿惠辞职的新闻。
Topic:0
的同志和人民卡斯特罗菲德尔古巴他了我
Topic:1在朴槿惠向表示总统对将的月国民
Doc0:((toptopic:0)topicdistribution:[0.917141230.08285877])
Doc1:((toptopic:1)topicdistribution:[0.092006660.90799334])
还是存在一些虚词,例如“的”,“和”,“了”,“对”等词的干扰,但是大致来说,两则新闻的主题分布很明显,效果还不赖。
以上就是本文关于Python_LDA实现方法详解的全部内容,希望对大家有所帮助。感兴趣的朋友可以继续参阅本站:python+mongodb数据抓取详细介绍、Python探索之创建二叉树、Python探索之修改Python搜索路径等,有什么问题可以随时留言,欢迎大家一起交流讨论。感谢朋友们对本站的支持!