Question

目的：对美国法院对同性婚姻的历史判断进行情绪分析。＃由于某些用户的推文数量非常高，可能会引入偏见。我们怎样才能删除它们？＃另外，为什么在usafull和total中的独特推文数量不同？

    rm(list=ls())
    library(twitteR)
    library(wordcloud)
    library(tm)

    download.file(url="http://curl.haxx.se/ca/cacert.pem",   destfile="cacert.pem")

    consumer_key <- 'key'
    consumer_secret <- 'secret'
    access_token <- 'key'
    access_secret <- 'secret'
    setup_twitter_oauth(consumer_key, consumer_secret, access_token, access_secret)


    usa <- searchTwitter("#LoveWins", n=1500 , lang="en")

    usa2 <- searchTwitter("#LGBT", n=1500 , lang="en")

    usa3 <- searchTwitter("#gay", n=1500 , lang="en")

#get the text
    tusa <- sapply(usa, function(x) x$getText())
    tusa2 <- sapply(usa2, function(x) x$getText())
    tusa3 <- sapply(usa3, function(x) x$getText())

#join texts
    total <- c(tusa,tusa2,tusa3)

#remove the duplicated tweets
    total <- total[!duplicated(total)]

#no. of unique tweets
    uni <- length(total)

# merging three set of tweets horozontally
    usafull<-c(usa,usa2,usa3)

#convert the tweets into dafa frame
    usafull <- twListToDF(usafull)
    usafull <- unique(usafull)

#to know the dates of the tweets (date formatting)
    usafull$date <- format(usafull$created, format = "%Y-%m-%d")
    table(usafull$date)

#make a table of number of tweets per user in decreasing number of tweets
    tdata <- as.data.frame(table(usafull$screenName))
    tdata <- tdata[order(tdata$Freq, decreasing = T), ]
    names(tdata) <- c("User","Tweets")
    head(tdata)


# plot the freq of tweets over time in two hour windows
    library(ggplot2)
    minutes <-60
    ggplot(data = usafull, aes(x=created))+geom_bar(aes(fill=..count..),    binwidth =60*minutes)+scale_x_datetime("Date")+ scale_y_continuous("Frequency")


#plot the table above for the top 30 to identify any unusual trends
    par(mar=c(5,10,2,2))
    with(tdata[rev(1:30), ], barplot(Tweets, names=User, horiz = T, las =1,     main="Top 30: Tweets per user", col = 1))

# the twitter users with more than 20 tweets for removing bias
    userid <- tdata[(tdata$Tweets>20),]
    userid <- userid[,1]

Answer 1

从您的代码中我了解到您要删除userid中的推文，其中一种方法就是这样，

usafull_nobias <- subset(usafull, !(screenName %in% userid$User))

至于您在total和usafull中获得不同数量的推文的原因，可能是因为在total中你使用推文发现的文字重复，并在usafull中使用完整的推文;考虑到这一点，例如转推可能有相同的文字，但可能来自不同的用户，有不同的ID等。

希望它有所帮助。

我们如何从特定用户（具有大量推文的用户）中删除使用R进行情绪分析的推文？

1 个答案: