Work-in-R

News Popularity

rm(list=ls()) #FACEBOOK library(dplyr) library(data.table) data=fread("/Users/tanz/Desktop/News_Final.csv", header = T)

names(data)<-tolower(names(data)) colnames(data) str(data) news_pop<-subset(data, select = -c(idlink,title,headline,source,topic,publishdate,sentimenttitle,sentimentheadline))

#Cleaning data news_pop=data%>% filter(facebook>=0)%>% select(c("headline","facebook"))

#Creating popular/unpopular parameters news_pop$facebook<-ifelse(news_pop$facebook>=10,'Popular','Unpopular')

Sampling the data

news_sample = news_pop[sample(1:nrow(news_pop),nrow(news_pop)/3, replace = FALSE)]

selecting columns

str(news_sample) summary(news_sample)

news_sample$facebook=factor(news_sample$facebook)

library(tm) news_corpus=VCorpus(VectorSource(news_sample$headline))

inspect(news_corpus[1:2])

converted to Character

lapply(news_corpus[1:3], as.character)

making all to lower cases

news_corpus_ready=tm_map(news_corpus, content_transformer(tolower))

as.character(news_corpus_ready[[1]]) news_corpus_ready=tm_map(news_corpus_ready,removeNumbers) news_corpus_ready=tm_map(news_corpus_ready,removeWords,stopwords()) news_corpus_ready=tm_map(news_corpus_ready,removePunctuation)

############### library(SnowballC) news_corpus_ready=tm_map(news_corpus_ready,stemDocument) news_corpus_ready=tm_map(news_corpus_ready,stripWhitespace) news_corpus_ready=tm_map(news_corpus_ready,removeNumbers) ############## #Creates a DTM news_dtm=DocumentTermMatrix(news_corpus_ready)

#converting to a matrix and to see news_mat=as.matrix(news_dtm)

#Create a train and test set dim(news_dtm) #Create a train and test set with round the prediction number(train 80% and test 20%) n = round(nrow(news_dtm)*.80,digits=0) news_dtm_train=news_dtm[1:n,] news_dtm_test=news_dtm[(n+1):nrow(news_dtm),]

Creating labels (6530 train and 1633 test)

news_train_labels=news_sample[1:n,]$facebook news_test_labels=news_sample[(n+1):nrow(news_dtm),]$facebook length(news_test_labels)

Naive bayes

news_freq=findFreqTerms(news_dtm_train,2) library(dplyr) news_dtm_freq_train=news_dtm_train[,news_freq] news_dtm_freq_test=news_dtm_test[,news_freq]

convert_counts= function(x){ x=ifelse(x>0, "Yes", "No") } news_train=apply(news_dtm_freq_train,MARGIN = 2, convert_counts)

news_test=apply(news_dtm_freq_test,MARGIN = 2, convert_counts)

#Phase- 2: Implement Bayes Algo and eval performance

library(e1071) news_algo=naiveBayes(news_train, news_train_labels) news_test_pred=predict(news_algo, news_test)

length(news_train) library(gmodels)

CrossTable(news_test_pred,news_test_labels, prop.chisq = FALSE, prop.c = FALSE, prop.r = FALSE, dnn = c('classified','actual'))