R语言-文本挖掘 主题模型 文本分类

####需要先安装几个R包，如果有这些包，可省略安装包的步骤。
#install.packages("Rwordseg")
#install.packages("tm");
#install.packages("wordcloud");
#install.packages("topicmodels")

例子中所用数据

数据来源于sougou实验室数据。

数据网址：http://download.labs.sogou.com/dl/sogoulabdown/SogouC.mini.20061102.tar.gz

文件结构

└─Sample

├─C000007 汽车

├─C000008 财经

├─C000010 IT

├─C000013 健康

├─C000014 体育

├─C000016 旅游

├─C000020 教育

├─C000022 招聘

├─C000023

└─C000024 军事

采用Python对数据进行预处理为train.csv文件，并把每个文件文本数据处理为1行。

预处理python脚本
<ignore_js_op> combineSample.zip (720 Bytes, 下载次数: 96)

所需数据
<ignore_js_op> train.zip (130.2 KB, 下载次数: 164) 
大家也可以用R直接将原始数据转变成train.csv中的数据

文章所需stopwords
<ignore_js_op> StopWords.zip (2.96 KB, 下载次数: 114)

1.     读取资料库

1. csv <- read.csv("d://wb//train.csv",header=T, stringsAsFactors=F)
2. mystopwords<- unlist (read.table("d://wb//StopWords.txt",stringsAsFactors=F))

复制代码

2.

数据预处理（中文分词、stopwords处理）

1. library(tm);
2. #移除数字
3. removeNumbers = function(x) { ret = gsub("[0-9０１２３４５６７８９]","",x) }
4. sample.words <- lapply(csv$$$$text, removeNumbers)

复制代码

1. #处理中文分词,此处用到Rwordseg包
2. wordsegment<- function(x) {
3. library(Rwordseg)
4. segmentCN(x)
5. }
6. sample.words <- lapply(sample.words, wordsegment)

复制代码

1. ###stopwords处理
2. ###先处理中文分词，再处理stopwords，防止全局替换丢失信息
3. removeStopWords = function(x,words) {
4. ret = character(0)
5. index <- 1
6. it_max <- length(x)
7. while (index <= it_max) {
8. if (length(words[words==x[index]]) <1) ret <- c(ret,x[index])
9. index <- index +1
10. }
11. ret
12. }
13. sample.words <- lapply(sample.words, removeStopWords, mystopwords)

复制代码

3.    wordcloud展示

1. #构建语料库
2. corpus = Corpus(VectorSource(sample.words))
3. meta(corpus,"cluster") <- csv$$$$type
4. unique_type <- unique(csv$$$$type)
5. #建立文档-词条矩阵
6. (sample.dtm <- DocumentTermMatrix(corpus, control = list(wordLengths = c(2, Inf))))

复制代码

1. #install.packages("wordcloud"); ##需要wordcloud包的支持
2. library(wordcloud);
3. #不同文档wordcloud对比图
4. sample.tdm <-  TermDocumentMatrix(corpus, control = list(wordLengths = c(2, Inf)));
5. tdm_matrix <- as.matrix(sample.tdm);
6. png(paste("d://wb//sample_comparison",".png", sep = ""), width = 1500, height = 1500 );
7. comparison.cloud(tdm_matrix,colors=rainbow(ncol(tdm_matrix)));####由于颜色问题，稍作修改
8. title(main = "sample comparision");
9. dev.off();

复制代码

1. #按分类汇总wordcloud对比图
2. n <- nrow(csv)
3. zz1 = 1:n
4. cluster_matrix<-sapply(unique_type,function(type){apply(tdm_matrix[,zz1[csv$$$$type==type]],1,sum)})
5. png(paste("d://wb//sample_ cluster_comparison",".png", sep = ""), width = 800, height = 800 )
6. comparison.cloud(cluster_matrix,colors=brewer.pal(ncol(cluster_matrix),"Paired")) ##由于颜色分类过少，此处稍作修改
7. title(main = "sample cluster comparision")
8. dev.off()

复制代码

<ignore_js_op>

可以看出数据分布不均匀，culture、auto等数据很少。

1. #按各分类画wordcloud
2. sample.cloud <- function(cluster, maxwords = 100) {
3. words <- sample.words[which(csv$$$$type==cluster)]
4. allwords <- unlist(words)
5. wordsfreq <- sort(table(allwords), decreasing = T)
6. wordsname <- names(wordsfreq)
7. png(paste("d://wb//sample_", cluster, ".png", sep = ""), width = 600, height = 600 )
8. wordcloud(wordsname, wordsfreq, scale = c(6, 1.5), min.freq = 2, max.words = maxwords, colors = rainbow(100))
9. title(main = paste("cluster:", cluster))
10. dev.off()
11. }
12. lapply(unique_type,sample.cloud)# unique(csv$$$$type)

复制代码

<ignore_js_op> 
<ignore_js_op>

4.    主题模型分析

1. library(slam)
2. summary(col_sums(sample.dtm))
3. term_tfidf  <- tapply(sample.dtm$$$$v/row_sums( sample.dtm)[ sample.dtm$$$$i],   sample.dtm$$$$j,  mean)*
4. log2(nDocs( sample.dtm)/col_sums( sample.dtm  >  0))
5. summary(term_tfidf)
6. sample.dtm  <-  sample.dtm[,  term_tfidf  >=  0.1]
7. sample.dtm  <-  sample.dtm[row_sums(sample.dtm)  >  0,]
8. library(topicmodels)
9. k <- 30
10. SEED <- 2010
11. sample_TM <-
12. list(
13. VEM = LDA(sample.dtm, k = k, control = list(seed = SEED)),
14. VEM_fixed = LDA(sample.dtm, k = k,control = list(estimate.alpha = FALSE, seed = SEED)),
15. Gibbs = LDA(sample.dtm, k = k, method = "Gibbs",control = list(seed = SEED, burnin = 1000,thin = 100, iter = 1000)),
16. CTM = CTM(sample.dtm, k = k,control = list(seed = SEED,var = list(tol = 10^-4), em = list(tol = 10^-3)))
17. )

复制代码

<ignore_js_op>

1. sapply(sample_TM[1:2], slot, "alpha")
2. sapply(sample_TM, function(x) mean(apply(posterior(x)$$$$topics,1, function(z) - sum(z * log(z)))))

复制代码

<ignore_js_op>

α估计严重小于默认值，这表明Dirichlet分布数据集中于部分数据，文档包括部分主题。
数值越高说明主题分布更均匀

1. #最可能的主题文档
2. Topic <- topics(sample_TM[["VEM"]], 1)
3. table(Topic)
4. #每个Topic前5个Term
5. Terms <- terms(sample_TM[["VEM"]], 5)
6. Terms[,1:10]

复制代码

<ignore_js_op>

1. ######### auto中每一篇文章中主题数目
2. (topics_auto <-topics(sample_TM[["VEM"]])[ grep("auto", csv[[1]]) ])
3. most_frequent_auto <- which.max(tabulate(topics_auto))
4. ######### 与auto主题最相关的10个词语
5. terms(sample_TM[["VEM"]], 10)[, most_frequent_auto]

复制代码

<ignore_js_op>