• Here I built an app to predict the next word when gaven a previous phrase (multiple words). The link to the app is: https://aifyer.shinyapps.io/wordPredict/
• On the left panel of the app page, the text box under the header “Please enter your phrases here!” is where to input your phrase (leaving the last word out).
• After input the phrase, press Submit! button and wait. The average time I wait is 12 seconds.
• Then, on the right panel, under the header “Your next word might be:”, you will see the next word predicted.

• To make the prediction, all text samples from twitter/blogs/news were read in as the studying copora.
• First, each text sample was split to single sentences by punctuations “[\.]|[!]|[?]|[,]|[;]”.
• Then each sentence was split to words by “[^[:alpha:]’’]”. Here “’’” was included to keep phrases like I’m/we’ve.
• Sometimes the words were surrounded by apostrophe like ‘hello’, which will make it a totally different word from hello, and these kind of apostrophes were removed. some critical codes for these steps were shown here:

a=unlist(strsplit(t,split="[\\.]|[!]|[?]|[,]|[;]")) #split to sentences
a=unlist(strsplit(t,split="[^[:alpha:]'’]")) #split sentence to words
a=gsub("^['’]|['’]$","",a) #remove unwanted '’

• Bigram and Trigram were got by paste each word with its following one or two words in every sentence.
• Single word (Unigram),Bigram and Trigram probability was calculated separately by the following formulas. Here w is every word. C is its counts in the copora. N: total words in the copora. V: words in vocabulary. \[ P(w_i)=(Cw_i+1)/(N+V) # Unigram P(w_i|w_{i-1}))=(C(w_{i-1},w_i)+1)/(C(w_{i-1})+V) # Bigram P(w_i|w_{i-1},w_{i-2}))=(C(w_{i-2},w_{i-1},w_i)+1)/(C(w_{i-2},w_{i-1})+V) # Trigram \]

I use the Bigram & Trigram statistic probability to predict next word.The predict word is the word with highest Trigram or Bigram (if input phrase contains only one word. or Trigram has no output) probability. In the results, stopword from package(tm) and letters were removed.

some key codes for these steps were:

v2Fm$Ptwo<-(v2Fm$Freq.y+1)/(v2Fm$Freq.x+nrow(v1F)) #calculate bigram probability
v3Fm$Ptri<-(v3Fm$gram3Freq+1)/(v3Fm$gram2Freq+nrow(v1F)) #calculate trigram probability
regex <- paste0("\\s*\\b(", paste(wordsToRemove, collapse="|"), ")\\b") #get word to remove expression

Thanks for trying! And hope you find my app useful!