APP
App interface.
Kalpana Rupani
12/2/2017
The goal of this exercise is to create a product to highlight the prediction algorithm that you have built and to provide an interface that can be accessed by others. For this project you must submit:
A Shiny app that takes as input a phrase (multiple words) in a text box input and outputs a prediction of the next word.
A slide deck consisting of no more than 5 slides created with R Studio Presenter pitching your algorithm and app as if you were presenting to your boss or an investor.
A key point here is that the predictive model must be small enough to load onto the Shiny server. So pay attention to model size when creating and uploading your model.
The User Input panel on the left contains three items. First is a text box to enter the phrase you’d like analyzed. Second is a drop down menu where you can select the maximum number of words to return. Third is a button, ‘Analyze Text’, that is the algorithm’s call to action and initiates the analysis. Analysis will only be performed upon pressing this button.
The algorithm returns three things. First the original text that the user provided, second is the filtered text provided to the algorithm, third is a table. In the lefthand column of this table we have the predicted words, in the right hand column we have the log probability. The table is sorted from the most likely word in the first row to the least likely in the last row.
The first task is to filter the input, this is same text cleaning process we used on the SwiftKey data. This includes removing numbers, punctuation, foreign characters, profanity, single letter words (b, c, y, etc.), and contractions.
Next we search of matches based on the user input. For example if we have the input ‘looking forward seeing’ a match is defined as ‘looking forward seeing you’. Based on the maximum number of results to be returned we may shorten the user input to find more matches. An example of shortening would be ‘looking forward seeing’ to ‘forward seeing’. If matches are found with shortened phrases a penalty is used to alter the final log probability score.
Once we have the desired number of matches we use a probability model (an example of this model can be found at X FILL X Slide 4) to give a score for each predicted word. Scores are then sorted in descending order so the most likely word is at the top of the table returned.
App interface.