Question

我已经和NaiveBayes分类了好几个小时了。标记为TrainData.tsv的文件由3列和25000行组成。列是id（chr），sentiment（int）和review（chr）。

我试着预测测试数据的情绪，但无论我做什么，它都会返回50/50的猜测。产生的真值表总是如下：

prediction   0   1
         0 199 201
         1   0   0

我做错了什么？代码

library(e1071)
library(tm)

trainData <- read.table("data/labeledTrainData.tsv", header = TRUE, stringsAsFactors = FALSE)
trainData <- trainData[sample(nrow(trainData)),]

range.train <- c(1:1600) 
range.test <- c(1601:2000) 

df <- trainData[1:20000,]
df$sentiment <- as.factor(df$sentiment)
df.train <- df[range.train,]
df.test <- df[range.test,]

corpus <- Corpus(VectorSource(df$review))

corpus.clean <- cleanText(corpus)

dtm <- DocumentTermMatrix(corpus.clean)

trainNB <- as.matrix(dtm[range.train,])
testNB <- as.matrix(dtm[range.test,])

system.time( classifier <- naiveBayes(trainNB, df.train$sentiment) )

system.time( prediction <- predict(classifier, testNB[,-1]) )

table(prediction, df.test$sentiment)

convert_count <- function(x) {
  y <- ifelse(x > 0, 1, 0) # 1 might have to be changed to x
  y <- factor(y, levels=c(0,1), labels=c("No", "Yes"))
}

cleanText <- function(text) {
  text %>%
    tm_map(content_transformer(tolower)) %>%
    tm_map(removeWords, stopwords('en')) %>%
    tm_map(toSpace, "<br />") %>%
    tm_map(toSpace, "/") %>%
    tm_map(removePunctuation) %>%
    tm_map(removeNumbers) %>%
    tm_map(stemDocument)
}

toSpace <- content_transformer(function(x, pattern) {
  return (
    gsub(pattern, " ", x)
  )
})

NaiveBayes算法的准确度达到50/50

0 个答案: