-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathslidedeck.Rpres
48 lines (38 loc) · 1.52 KB
/
slidedeck.Rpres
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
Text Prediction Web Application
========================================================
author: Michal Svoboda
date: 27.2.2018
autosize: true
Task
========================================================
- Develop a web application which given a part of the sentence, predicts the next word.
- Deploy the application on shinyapps.io
Text prediction model
========================================================
- N-Gram language model using interpolation.
- Algorithm relies on assigning score for the prediction as a sum of conditional probabilities of the word given 1 to 4 previous words
- Words with the highest scores are returned
Model verification
==================
<br>
- Accuracy (likelihood that one of n suggested words is true)
```{r, echo=FALSE}
plot(x=1:5,xlab="n",ylab = "Accuracy",y=c(.447,.548,.611,.648,.673))
```
***
<br>
- Correlation between score and correct suggestion
```{r,echo=FALSE}
pr <- readRDS("protocoll.rds")
plot(pr$score,pr$match,xlab = "Score",ylab="Match probability")
abline(lm(match~score,data=pr),col="red")
```
Application
========================================================
<br>
- The [application](https://svobom38.shinyapps.io/CapstoneApp/) accepts an input from user and shows suggestions for the next word in the button labels
- The action buttons can be used similarly as the prediction bars of the mobile device`s keyboards
- Unlike the most text predicton applications, this one shows the score, which is proportional to the likelihood of the correct answer
***
<br>
