Loading Necessary Packages

require(httr)
require(jsonlite)
require(dplyr)
require(rtweet)
require(ggplot2)
require(dplyr)
require(tm)
require(wordcloud)
require(clarifai)
require(twitteR)
require(tidytext)
require(stringr)
library(tibbletime)
library(dplyr)
library(lubridate)
library(alphavantager)

Getting AlphaVantage API access

av_api_key('6KXR3Y47EGL14D1')
#av_get('BYND')

Retrieving stock data from AlphaVantage based on stock abbreviation. THe data is then sorted and plotted in the interval of Dec 28 to Dec 2. The code below is for Microsoft but 4 other companies were also analyzed.

Microsoft_Stock<-av_get(symbol = "MSFT", av_fun = "TIME_SERIES_INTRADAY_EXTENDED", interval = "60min", outputsize = "compact", slice='year1month1')

Microsoft_Stock$Day<-day(as.POSIXlt.POSIXct(Microsoft_Stock$time))
Microsoft_Stock<-  filter(Microsoft_Stock, Microsoft_Stock$Day==28 | Microsoft_Stock$Day==29 | Microsoft_Stock$Day==30 | Microsoft_Stock$Day==1 | Microsoft_Stock$Day==2)
Microsoft_Stock<-as.data.frame(Microsoft_Stock)
Microsoft_Stock_Plot<- as.data.frame(unique(Microsoft_Stock$Day))

Microsoft_Stock_Plot$DayAverage <- 1.0
for(i in (1:nrow(Microsoft_Stock_Plot))){
  Microsoft_Stock_Plot[i,2] <- mean(Microsoft_Stock[Microsoft_Stock$Day == Microsoft_Stock_Plot[i,1], 'close'])
}

Microsoft_Stock_Plot %>% arrange(factor(`unique(Microsoft_Stock$Day)`,levels =c(29,30,1,2,3)))
Microsoft_Stock_Plot$day = c(1,2,3,4,5)
plot(Microsoft_Stock_Plot$DayAverage~Microsoft_Stock_Plot$day,type='l', xlab = "Day", ylab=("Average Stock Price"), main="Microsoft Stock Nov29-Dec2")
Tesla_Stock<-av_get(symbol = "TSLA", av_fun = "TIME_SERIES_INTRADAY_EXTENDED", interval = "60min", outputsize = "compact", slice='year1month1')

Tesla_Stock$Day<-day(as.POSIXlt.POSIXct(Tesla_Stock$time))
Tesla_Stock<-  filter(Tesla_Stock, Tesla_Stock$Day==28 | Tesla_Stock$Day==29 | Tesla_Stock$Day==30 | Tesla_Stock$Day==1 | Tesla_Stock$Day==2)
Tesla_Stock<-as.data.frame(Tesla_Stock)
Tesla_Stock_Plot<- as.data.frame(unique(Tesla_Stock$Day))

Tesla_Stock_Plot$DayAverage <- 1.0
for(i in (1:nrow(Tesla_Stock_Plot))){
  Tesla_Stock_Plot[i,2] <- mean(Tesla_Stock[Tesla_Stock$Day == Tesla_Stock_Plot[i,1], 'close'])
}

Tesla_Stock_Plot %>% arrange(factor(`unique(Tesla_Stock$Day)`,levels =c(29,30,1,2,3)))
Tesla_Stock_Plot$day = c(1,2,3,4,5)
plot(Tesla_Stock_Plot$DayAverage~Tesla_Stock_Plot$day,type='l', xlab = "Day", ylab=("Average Stock Price"), main="Tesla Stock Nov29-Dec2")
Apple_Stock<-av_get(symbol = "TSLA", av_fun = "TIME_SERIES_INTRADAY_EXTENDED", interval = "60min", outputsize = "compact", slice='year1month1')

Apple_Stock$Day<-day(as.POSIXlt.POSIXct(Apple_Stock$time))
Apple_Stock<-  filter(Apple_Stock, Apple_Stock$Day==28 | Apple_Stock$Day==29 | Apple_Stock$Day==30 | Apple_Stock$Day==1 | Apple_Stock$Day==2)
Apple_Stock<-as.data.frame(Apple_Stock)
Apple_Stock_Plot<- as.data.frame(unique(Apple_Stock$Day))

Apple_Stock_Plot$DayAverage <- 1.0
for(i in (1:nrow(Apple_Stock_Plot))){
  Apple_Stock_Plot[i,2] <- mean(Apple_Stock[Apple_Stock$Day == Apple_Stock_Plot[i,1], 'close'])
}

Apple_Stock_Plot %>% arrange(factor(`unique(Apple_Stock$Day)`,levels =c(29,30,1,2,3)))
Apple_Stock_Plot$day = c(1,2,3,4,5)
plot(Apple_Stock_Plot$DayAverage~Apple_Stock_Plot$day,type='l', xlab = "Day", ylab=("Average Stock Price"), main="Apple Stock Nov29-Dec2")
Target_Stock<-av_get(symbol = "TGT", av_fun = "TIME_SERIES_INTRADAY_EXTENDED", interval = "60min", outputsize = "compact", slice='year1month1')

Target_Stock$Day<-day(as.POSIXlt.POSIXct(Target_Stock$time))
Target_Stock<-  filter(Target_Stock, Target_Stock$Day==28 | Target_Stock$Day==29 | Target_Stock$Day==30 | Target_Stock$Day==1 | Target_Stock$Day==2)
Target_Stock<-as.data.frame(Target_Stock)
Target_Stock_Plot<- as.data.frame(unique(Target_Stock$Day))

Target_Stock_Plot$DayAverage <- 1.0
for(i in (1:nrow(Target_Stock_Plot))){
  Target_Stock_Plot[i,2] <- mean(Target_Stock[Target_Stock$Day == Target_Stock_Plot[i,1], 'close'])
}

Target_Stock_Plot %>% arrange(factor(`unique(Target_Stock$Day)`,levels =c(29,30,1,2,3)))
Target_Stock_Plot$day = c(1,2,3,4,5)
plot(Target_Stock_Plot$DayAverage~Target_Stock_Plot$day,type='l', xlab = "Day", ylab=("Average Stock Price"), main="Target Stock Nov29-Dec2")

Loading tweet data required me applying for a twitter developer account to which I was given access keys to the twitter API. I

bearer_token <- "AAAAAAAAAAAAAAAAAAAAAL2gjwEAAAAACuACbul%2BLX4asrb%2FwiPmOWf1trw%3DO0WhnmZqH6eJneMMDOvV7me8GyGhgXSxlCHFex73dsbey02lI5"

auth <- rtweet_app()

Reading raw tweets data

Doge_Tweets_raw <- search_tweets(q = "#Doge", n = 40000, lang ="en")
Apple_Tweets_raw <- search_tweets(q="#AAPL",n =40000, lang="en")
Target_Tweets_raw <- search_tweets(q="#TGT",n =40000, lang="en")
MSFT_Tweets_raw <- search_tweets(q="#MSFT",n =40000, lang="en")

Function to clean tweets, using the afinn sentiment library and placing all data into a dataframe.

FindSentiment <- function(raw){
  
clean <- raw[,c(1,4)]
cleandf <- as.data.frame(clean)




cleandf$full_text <- removeNumbers(cleandf$full_text)
cleandf$full_text <- removePunctuation(cleandf$full_text)
cleandf$full_text <- tolower(cleandf$full_text)
cleandf$full_text <- removeWords(cleandf$full_text,stopwords())
cleandf$full_text <- gsub("[^\x01-\x7F]", "", cleandf$full_text)


one <- summarise(cleandf,text=paste(full_text,collapse = " "))
one <- strsplit(one[1,1], " ")[[1]]
one <- as.data.frame(one)
one <- one[nzchar(one$one),]
one <- as.data.frame(one)

names(one)[1]<-"word"

sentimentdf<-merge(x = one, y = afinn, by = c('word') )# Equivalent

sentimentScore <- sum(sentimentdf$value)


return(sentimentScore)
}

Raw data is read into the function, cleaned and found sentiments through the outside function call. The data is then placed in another data frame that is averaged by the 5 day interval of Nov 28 - Dec 2. A plot of this resulting data frame is then done.

TSLA_Sentiments <- TSLA_Tweets_raw[,c(1,4)]
TSLA_Sentiments[,3]= 0
names(TSLA_Sentiments)[3]<-"Sentiment"
TSLA_Sentiments$Day <- day(TSLA_Sentiments$created_at)

for (i in (1:nrow(TSLA_Sentiments))){
TSLA_Sentiments[i,3] <- FindSentiment(TSLA_Tweets_raw[i,]) 
}

TSLA_Sentiments_Plot <- as.data.frame(unique(TSLA_Sentiments$Day))
TSLA_Sentiments$DayAverage <- 1.0
TSLA_Sentiments_Plot<-as.data.frame(TSLA_Sentiments_Plot)
TSLA_Sentiments <- as.data.frame(TSLA_Sentiments)
for(i in (1:nrow(TSLA_Sentiments_Plot))){
  TSLA_Sentiments_Plot[i,2] <- mean(TSLA_Sentiments[TSLA_Sentiments$Day == TSLA_Sentiments_Plot[i,1], 'Sentiment'])
}

TSLA_Sentiments_Plot
TSLA_Sentiments_Plot<- arrange(TSLA_Sentiments_Plot,`unique(TSLA_Sentiments$Day)`)
TSLA_Sentiments_Plot<-head(TSLA_Sentiments_Plot,5)
plot(TSLA_Sentiments_Plot$V2~TSLA_Sentiments_Plot$`unique(TSLA_Sentiments$Day)`,type='l', xlab = "day", ylab=("Average Sentiment Score"), main="TSLA Sentiment vs Day")
Apple_Sentiments <- Apple_Tweets_raw[,c(1,4)]
Apple_Sentiments[,3]= 0
names(Apple_Sentiments)[3]<-"Sentiment"
Apple_Sentiments$Day <- day(Apple_Sentiments$created_at)

for (i in (1:nrow(Apple_Sentiments))){
Apple_Sentiments[i,3] <- FindSentiment(Apple_Tweets_raw[i,]) 
}

Apple_Sentiments_Plot <- as.data.frame(unique(Apple_Sentiments$Day))
Apple_Sentiments$DayAverage <- 1.0


for(i in (1:nrow(Apple_Sentiments_Plot))){
  Apple_Sentiments_Plot[i,2] <- mean(Apple_Sentiments$Day==Apple_Sentiments_Plot[i,1])
}

Apple_Sentiments_Plot
Apple_Sentiments_Plot<- arrange(Apple_Sentiments_Plot,`unique(Apple_Sentiments$Day)`)
Apple_Sentiments_Plot<-head(Apple_Sentiments_Plot,5)
plot(Apple_Sentiments_Plot$V2~Apple_Sentiments_Plot$`unique(Apple_Sentiments$Day)`,type='l', xlab = "day", ylab=("Average Sentiment Score"), main="Apple Sentiment vs Stock")
Target_Sentiments <- Target_Tweets_raw[,c(1,4)]
Target_Sentiments[,3]= 0
names(Target_Sentiments)[3]<-"Sentiment"
Target_Sentiments$Day <- day(Target_Sentiments$created_at)

for (i in (1:nrow(Target_Sentiments))){
Target_Sentiments[i,3] <- FindSentiment(Target_Tweets_raw[i,]) 
}

Target_Sentiments_Plot <- as.data.frame(unique(Target_Sentiments$Day))
Target_Sentiments$DayAverage <- 1.0


for(i in (1:nrow(Target_Sentiments_Plot))){
  Target_Sentiments_Plot[i,2] <- mean(Target_Sentiments$Day==Target_Sentiments_Plot[i,1])
}
print(unique(Target_Sentiments$Day))
Target_Sentiments_Plot
Target_Sentiments_Plot<- arrange(Target_Sentiments_Plot,`unique(Target_Sentiments$Day)`)
Target_Sentiments_Plot<-head(Target_Sentiments_Plot,5)
plot(Target_Sentiments_Plot$V2~Target_Sentiments_Plot$`unique(Target_Sentiments$Day)`,type='l', xlab = "day", ylab=("Average Sentiment Score"), main="Target Sentiment vs Stock")
TSLA_Sentiments <- TSLA_Tweets_raw[,c(1,4)]
TSLA_Sentiments[,3]= 0
names(TSLA_Sentiments)[3]<-"Sentiment"
TSLA_Sentiments$Day <- day(TSLA_Sentiments$created_at)

for (i in (1:nrow(TSLA_Sentiments))){
TSLA_Sentiments[i,3] <- FindSentiment(TSLA_Tweets_raw[i,]) 
}

TSLA_Sentiments_Plot <- as.data.frame(unique(TSLA_Sentiments$Day))
TSLA_Sentiments$DayAverage <- 1.0


for(i in (1:nrow(TSLA_Sentiments_Plot))){
  TSLA_Sentiments_Plot[i,2] <- mean(TSLA_Sentiments$Day==TSLA_Sentiments_Plot[i,1])
}

TSLA_Sentiments_Plot
TSLA_Sentiments_Plot<- arrange(TSLA_Sentiments_Plot,`unique(TSLA_Sentiments$Day)`)
TSLA_Sentiments_Plot<-head(TSLA_Sentiments_Plot,5)
plot(TSLA_Sentiments_Plot$V2~TSLA_Sentiments_Plot$`unique(TSLA_Sentiments$Day)`,type='l', xlab = "day", ylab=("Average Sentiment Score"), main="TSLA Sentiment Dec1-Dec5")
Apple_Sentiments <- Apple_Tweets_raw[,c(1,4)]
Apple_Sentiments[,3]= 0
names(Apple_Sentiments)[3]<-"Sentiment"
Apple_Sentiments$Day <- day(Apple_Sentiments$created_at)

for (i in (1:nrow(Apple_Sentiments))){
Apple_Sentiments[i,3] <- FindSentiment(Apple_Tweets_raw[i,]) 
}

Apple_Sentiments_Plot <- as.data.frame(unique(Apple_Sentiments$Day))
Apple_Sentiments$DayAverage <- 1.0


for(i in (1:nrow(Apple_Sentiments_Plot))){
  Apple_Sentiments_Plot[i,2] <- mean(Apple_Sentiments$Day==Apple_Sentiments_Plot[i,1])
}

Apple_Sentiments_Plot
Apple_Sentiments_Plot<- arrange(Apple_Sentiments_Plot,`unique(Apple_Sentiments$Day)`)
Apple_Sentiments_Plot<-head(Apple_Sentiments_Plot,5)
plot(Apple_Sentiments_Plot$V2~Apple_Sentiments_Plot$`unique(Apple_Sentiments$Day)`,type='l', xlab = "day", ylab=("Average Sentiment Score"), main="Apple Sentiment Dec1-Dec5")

In Conclusion, there was no correlation of stock price and sentiment score. Considering we were only limited to a five day interval based on api restrictions, more accurate results would maybe arise if the interval was longer and sentiment bins were placed in a weekly or monthly basis. Further data cleaning and fine tuning of the sentiment library could also be used when determining sentiment score.