- Set up the folder and file system with pdf in Positive and Negative
- Set up the path for both positive and negative folder
- Set up all keywords
- Create a word count matrix of correct dimensions
- Loop through all pdf files and count the keywords of each files one by one
- Export the word Count as data frame
- Using SVM for classification
- Using Neural Network for classification
- Using Logistic regression for classification
- Test model on new data
getwd()## [1] "D:/Users/Documents/GitHub/Text-Mining-With-R/Text-Mining-With-R"
PositiveFiles <- paste0("./Positive/",list.files(path = "./Positive",pattern = ".pdf"))
NegativeFiles <- paste0("./Negative/",list.files(path = "./Negative",pattern = ".pdf"))
files <- c(PositiveFiles,NegativeFiles)
files## [1] "./Positive/journal.pone.0252003.pdf" "./Positive/journal.pone.0253080.pdf"
## [3] "./Positive/journal.pone.0254293.pdf" "./Positive/journal.pone.0255083.pdf"
## [5] "./Positive/journal.pone.0255531.pdf" "./Positive/journal.pone.0255674.pdf"
## [7] "./Positive/journal.pone.0255874.pdf" "./Positive/journal.pone.0256329.pdf"
## [9] "./Positive/journal.pone.0256579.pdf" "./Positive/journal.pone.0256679.pdf"
## [11] "./Positive/journal.pone.0256861.pdf" "./Positive/journal.pone.0256989.pdf"
## [13] "./Positive/journal.pone.0257209.pdf" "./Positive/journal.pone.0257243.pdf"
## [15] "./Positive/journal.pone.0257338.pdf" "./Positive/journal.pone.0257362.pdf"
## [17] "./Positive/journal.pone.0257392.pdf" "./Positive/journal.pone.0257419.pdf"
## [19] "./Positive/journal.pone.0257445.pdf" "./Positive/journal.pone.0257460.pdf"
## [21] "./Positive/journal.pone.0257510.pdf" "./Positive/journal.pone.0257519.pdf"
## [23] "./Positive/journal.pone.0257572.pdf" "./Positive/journal.pone.0257575.pdf"
## [25] "./Positive/journal.pone.0257582.pdf" "./Positive/journal.pone.0257624.pdf"
## [27] "./Positive/journal.pone.0257642.pdf" "./Positive/journal.pone.0257803.pdf"
## [29] "./Positive/journal.pone.0257826.pdf" "./Positive/journal.pone.0257858.pdf"
## [31] "./Positive/journal.pone.0257890.pdf" "./Positive/journal.pone.0257968.pdf"
## [33] "./Positive/journal.pone.0258003.pdf" "./Positive/journal.pone.0258062.pdf"
## [35] "./Positive/journal.pone.0258081.pdf" "./Positive/journal.pone.0258148.pdf"
## [37] "./Positive/journal.pone.0258149.pdf" "./Positive/journal.pone.0258150.pdf"
## [39] "./Positive/journal.pone.0258153.pdf" "./Positive/journal.pone.0258169.pdf"
## [41] "./Positive/journal.pone.0258183.pdf" "./Positive/journal.pone.0258186.pdf"
## [43] "./Positive/journal.pone.0258193.pdf" "./Positive/journal.pone.0258198.pdf"
## [45] "./Positive/journal.pone.0258212.pdf" "./Positive/journal.pone.0258220.pdf"
## [47] "./Positive/journal.pone.0258221.pdf" "./Positive/journal.pone.0258230.pdf"
## [49] "./Positive/journal.pone.0258233.pdf" "./Positive/journal.pone.0258249.pdf"
## [51] "./Positive/journal.pone.0258267.pdf" "./Positive/journal.pone.0258273.pdf"
## [53] "./Positive/journal.pone.0258275.pdf" "./Positive/journal.pone.0258285.pdf"
## [55] "./Positive/journal.pone.0258325.pdf" "./Positive/journal.pone.0258335.pdf"
## [57] "./Positive/journal.pone.0258383.pdf" "./Positive/journal.pone.0258394.pdf"
## [59] "./Positive/journal.pone.0258395.pdf" "./Positive/journal.pone.0258402.pdf"
## [61] "./Positive/journal.pone.0258567.pdf" "./Negative/journal.pone.0255231.pdf"
## [63] "./Negative/journal.pone.0255281.pdf" "./Negative/journal.pone.0255491.pdf"
## [65] "./Negative/journal.pone.0256136.pdf" "./Negative/journal.pone.0256159.pdf"
## [67] "./Negative/journal.pone.0256431.pdf" "./Negative/journal.pone.0256534.pdf"
## [69] "./Negative/journal.pone.0256567.pdf" "./Negative/journal.pone.0256575.pdf"
## [71] "./Negative/journal.pone.0256624.pdf" "./Negative/journal.pone.0256675.pdf"
## [73] "./Negative/journal.pone.0256683.pdf" "./Negative/journal.pone.0256710.pdf"
## [75] "./Negative/journal.pone.0256737.pdf" "./Negative/journal.pone.0256764.pdf"
## [77] "./Negative/journal.pone.0256837.pdf" "./Negative/journal.pone.0256881.pdf"
## [79] "./Negative/journal.pone.0256922.pdf" "./Negative/journal.pone.0256968.pdf"
## [81] "./Negative/journal.pone.0257101.pdf" "./Negative/journal.pone.0257246.pdf"
## [83] "./Negative/journal.pone.0257277.pdf" "./Negative/journal.pone.0257399.pdf"
## [85] "./Negative/journal.pone.0257412.pdf" "./Negative/journal.pone.0257428.pdf"
## [87] "./Negative/journal.pone.0257451.pdf" "./Negative/journal.pone.0257505.pdf"
## [89] "./Negative/journal.pone.0257539.pdf" "./Negative/journal.pone.0257631.pdf"
## [91] "./Negative/journal.pone.0257665.pdf" "./Negative/journal.pone.0257757.pdf"
## [93] "./Negative/journal.pone.0257912.pdf" "./Negative/journal.pone.0257954.pdf"
## [95] "./Negative/journal.pone.0257988.pdf" "./Negative/journal.pone.0258070.pdf"
## [97] "./Negative/journal.pone.0258405.pdf"
keywords <-
c(
"taxonomy",
"pcr",
"amplification",
"primer",
"molecular",
"cloud computing ",
"phylogenomic",
"genome",
"illumina",
"assembly",
"species",
"negative binomial",
"molecular biology",
"dna",
"rna-seq",
"degs",
"sequencing",
"population",
"education",
"participants"
)
head(keywords)## [1] "taxonomy" "pcr" "amplification" "primer"
## [5] "molecular" "cloud computing "
library(dplyr)##
## Attaching package: 'dplyr'
## The following objects are masked from 'package:stats':
##
## filter, lag
## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, union
library(stringr)
library(pdftools)## Using poppler version 21.04.0
filelength <- length(files)
wordlength <- length(keywords)
word_count <- seq(1,filelength*wordlength)
dim(word_count) <- c(filelength,wordlength)
head(word_count)## [,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10] [,11] [,12] [,13] [,14]
## [1,] 1 98 195 292 389 486 583 680 777 874 971 1068 1165 1262
## [2,] 2 99 196 293 390 487 584 681 778 875 972 1069 1166 1263
## [3,] 3 100 197 294 391 488 585 682 779 876 973 1070 1167 1264
## [4,] 4 101 198 295 392 489 586 683 780 877 974 1071 1168 1265
## [5,] 5 102 199 296 393 490 587 684 781 878 975 1072 1169 1266
## [6,] 6 103 200 297 394 491 588 685 782 879 976 1073 1170 1267
## [,15] [,16] [,17] [,18] [,19] [,20]
## [1,] 1359 1456 1553 1650 1747 1844
## [2,] 1360 1457 1554 1651 1748 1845
## [3,] 1361 1458 1555 1652 1749 1846
## [4,] 1362 1459 1556 1653 1750 1847
## [5,] 1363 1460 1557 1654 1751 1848
## [6,] 1364 1461 1558 1655 1752 1849
for (j in 1:length(files)) {
P1 <- pdftools::pdf_text(pdf = files[j]) %>%
str_to_lower() %>%
str_replace_all("\\t", "") %>%
str_replace_all("\n", " ") %>%
str_replace_all(" ", " ") %>%
str_replace_all(" ", " ") %>%
str_replace_all(" ", " ") %>%
str_replace_all(" ", " ") %>%
str_replace_all("[:digit:]", "") %>%
str_replace_all("[:punct:]", "") %>%
str_trim()
for (i in 1:length(keywords)) {
word_count[j,i] <- P1 %>% str_count(keywords[i]) %>% sum()
}
}
head(word_count)## [,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10] [,11] [,12] [,13] [,14]
## [1,] 61 0 0 0 0 0 0 0 0 0 0 0 0 0
## [2,] 0 0 0 0 0 0 0 0 0 0 1 0 0 1
## [3,] 0 2 1 5 12 0 0 1 0 0 19 0 0 3
## [4,] 6 1 2 3 3 0 0 9 3 0 17 0 0 16
## [5,] 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## [6,] 11 0 0 0 5 0 0 0 0 0 62 0 0 6
## [,15] [,16] [,17] [,18] [,19] [,20]
## [1,] 0 0 0 1 13 2
## [2,] 0 0 0 8 0 0
## [3,] 0 0 1 1 2 0
## [4,] 0 0 15 0 0 0
## [5,] 0 0 0 0 0 73
## [6,] 0 0 0 0 1 0
word_matrix <- as.data.frame(word_count)
SeqPos <- paste0("P",seq(1,length(PositiveFiles)))
SeqNeg <- paste0("N",seq(1,length(NegativeFiles)))
SeqName <- c(SeqPos,SeqNeg)
rownames(word_matrix) <- SeqName
colnames(word_matrix) <- keywords
Label <- c(rep("Positive",length(PositiveFiles)),rep("Negative",length(NegativeFiles)))
Label## [1] "Positive" "Positive" "Positive" "Positive" "Positive" "Positive"
## [7] "Positive" "Positive" "Positive" "Positive" "Positive" "Positive"
## [13] "Positive" "Positive" "Positive" "Positive" "Positive" "Positive"
## [19] "Positive" "Positive" "Positive" "Positive" "Positive" "Positive"
## [25] "Positive" "Positive" "Positive" "Positive" "Positive" "Positive"
## [31] "Positive" "Positive" "Positive" "Positive" "Positive" "Positive"
## [37] "Positive" "Positive" "Positive" "Positive" "Positive" "Positive"
## [43] "Positive" "Positive" "Positive" "Positive" "Positive" "Positive"
## [49] "Positive" "Positive" "Positive" "Positive" "Positive" "Positive"
## [55] "Positive" "Positive" "Positive" "Positive" "Positive" "Positive"
## [61] "Positive" "Negative" "Negative" "Negative" "Negative" "Negative"
## [67] "Negative" "Negative" "Negative" "Negative" "Negative" "Negative"
## [73] "Negative" "Negative" "Negative" "Negative" "Negative" "Negative"
## [79] "Negative" "Negative" "Negative" "Negative" "Negative" "Negative"
## [85] "Negative" "Negative" "Negative" "Negative" "Negative" "Negative"
## [91] "Negative" "Negative" "Negative" "Negative" "Negative" "Negative"
## [97] "Negative"
library(pheatmap)
pheatmap(word_matrix)
indices <- colSums(word_matrix) > 10
word_matrix2 <- word_matrix[,indices]
pheatmap(word_matrix2)
library(e1071)
svm_x <- word_matrix2
svm_y <- as.factor(Label)
dat <- as.data.frame(cbind(svm_x,svm_y))
svm_model = svm(svm_y ~ ., data = dat, kernel = "linear", cost = 10, scale = FALSE)
svm_pred <- predict(svm_x, object = svm_model)
caret::confusionMatrix(svm_pred, svm_y)## Confusion Matrix and Statistics
##
## Reference
## Prediction Negative Positive
## Negative 33 17
## Positive 3 44
##
## Accuracy : 0.7938
## 95% CI : (0.6997, 0.8693)
## No Information Rate : 0.6289
## P-Value [Acc > NIR] : 0.0003538
##
## Kappa : 0.5909
##
## Mcnemar's Test P-Value : 0.0036504
##
## Sensitivity : 0.9167
## Specificity : 0.7213
## Pos Pred Value : 0.6600
## Neg Pred Value : 0.9362
## Prevalence : 0.3711
## Detection Rate : 0.3402
## Detection Prevalence : 0.5155
## Balanced Accuracy : 0.8190
##
## 'Positive' Class : Negative
##
library(glmnet)## Loading required package: Matrix
## Loaded glmnet 4.1-2
glmmodel<-glm(svm_y~., family = binomial(), data=dat)## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred
glmprop <- predict(svm_x,object=glmmodel)
glm.pred <- ifelse(glmprop > 0.5, "Positive", "Negative")
glm.pred <- as.factor(glm.pred)
glm_y <- as.factor(Label)
glm_cm <-caret::confusionMatrix(glm.pred, glm_y)
glm_cm$table## Reference
## Prediction Negative Positive
## Negative 10 4
## Positive 26 57
#https://www.statology.org/glm-fit-fitted-probabilities-numerically-0-or-1-occurred/###Create neural network
library(tensorflow)
library(keras)
copydata <- function(array,n)
{
for(i in 1:n){array <- rbind(array,array)}
return(array)
}
train_data <- word_matrix2 %>% unlist()
dim(train_data) <- dim(word_matrix2)
train_data <- train_data %>% copydata(5)
# Convert labels to categorical one-hot encoding
y <- as.numeric(as.factor(Label)) -1
train_label <- to_categorical(y,num_classes = 2)
train_label <- train_label %>% copydata(5)model <- keras_model_sequential() %>%
layer_dense(units = 28, activation = "sigmoid", input_shape = c(ncol(train_data))) %>%
layer_dense(units = 56, activation = "sigmoid") %>%
layer_dense(units = 28, activation = "sigmoid") %>%
layer_dense(units = 14, activation = "sigmoid") %>%
layer_dense(units = 2, activation = "softmax")
summary(model)## Model: "sequential"
## ________________________________________________________________________________
## Layer (type) Output Shape Param #
## ================================================================================
## dense_4 (Dense) (None, 28) 504
## ________________________________________________________________________________
## dense_3 (Dense) (None, 56) 1624
## ________________________________________________________________________________
## dense_2 (Dense) (None, 28) 1596
## ________________________________________________________________________________
## dense_1 (Dense) (None, 14) 406
## ________________________________________________________________________________
## dense (Dense) (None, 2) 30
## ================================================================================
## Total params: 4,160
## Trainable params: 4,160
## Non-trainable params: 0
## ________________________________________________________________________________
model %>% compile(
optimizer = 'rmsprop',
loss = 'categorical_crossentropy',
metrics = c('accuracy')
)history <- model %>%
fit(
x = train_data, y = train_label,
epochs = 20,
use_multiprocessing=TRUE,
batch_size = 20
)TestFiles <- paste0("./Test/",list.files(path = "./Test",pattern = ".pdf"))
TestFiles## [1] "./Test/Negative (1).pdf" "./Test/Negative (2).pdf"
## [3] "./Test/Negative (3).pdf" "./Test/Positive (1).pdf"
## [5] "./Test/Positive (2).pdf" "./Test/Positive (3).pdf"
filelength <- length(TestFiles)
wordlength <- length(keywords)
word_count_test <- matrix(nrow = filelength,ncol = wordlength)
for (j in 1:length(TestFiles)) {
P1 <- pdftools::pdf_text(pdf = TestFiles[j]) %>%
str_to_lower() %>%
str_replace_all("\\t", "") %>%
str_replace_all("\n", " ") %>%
str_replace_all(" ", " ") %>%
str_replace_all(" ", " ") %>%
str_replace_all(" ", " ") %>%
str_replace_all(" ", " ") %>%
str_replace_all("[:digit:]", "") %>%
str_replace_all("[:punct:]", "") %>%
str_trim()
for (i in 1:length(keywords)) {
word_count_test[j,i] <- P1 %>% str_count(keywords[i]) %>% sum()
}
}
word_matrix_test <- as.data.frame(word_count_test)
rownames(word_matrix_test) <- TestFiles
colnames(word_matrix_test) <- keywords
label_test <- c("Negative","Negative","Negative","Positive","Positive","Positive")
y <- as.factor(label_test)
x <- unlist(word_matrix_test[,indices])
dim(x) <- c(filelength,ncol(word_matrix2))#Predict Outcome of Testing Documents Against Ground Truth
NN_prediction <- predict(model,x=x)
NN_prediction <- round(NN_prediction)
NN_outcome <- c()
for(i in 1:nrow(NN_prediction)){
if(NN_prediction[i,1]){
NN_outcome <- c(NN_outcome,"Negative")
}else
{NN_outcome <- c(NN_outcome,"Positive")
}
}
NN_outcome <- as.factor(NN_outcome)
cm <- caret::confusionMatrix(NN_outcome, y)
cm$table## Reference
## Prediction Negative Positive
## Negative 3 0
## Positive 0 3
colnames(NN_prediction) <- c("Negative","Positive")
rownames(NN_prediction) <- TestFiles
NN_results <- NN_prediction[NN_prediction[,2]==1,]
NN_results## Negative Positive
## ./Test/Positive (1).pdf 0 1
## ./Test/Positive (2).pdf 0 1
## ./Test/Positive (3).pdf 0 1