Thetazero Pubs
Gumiltonprediction predicting predict model text word words click sentence abc

Fast Prediction

April 18th 2015DATA_ADSENSE


Natural Language Processing has been implicated as in great need. Text mining and predicting have been widely applied in many fields. Here we developed an integrated tool performing text predicting based on the input. Our software have advantages as:

  • Large and deep Coverage
  • Portable size and fast response
  • Reasonable and accurate prediction

Model and Algorithm

The basic model we applied is Markov model, given the user have typed in a sentence to predict what is the next word. We considered no more than last 3 words in the input sentence, for instance, the user typed in ?I like A B C? where A, B, C stand for any three words, we will predict the next word D with the following model, for which has the top 6 highest probability.

\( Pr(D|ABC) = \frac{Pr(ABCD)}{Pr(ABC)} \)

Algorithm (Continue)

  • The input by user will be cleaned up for better prediction.
  • If the three words A B C can not be found in our current database, we will backoff to use B C for prediction, then C alone.
  • If none of the backoff above can be solved, we will present top 6 most frequently used words.
  • Instead of predict one output, we present the user with top 6 choice to let the user choose from based on specific occasion.
  • The Markov model is modified to ?sense? the situation or occasion when a longer sentence is typed in, for instance, ?we like to go to? will not always predict ?school? but predict ?beach? when some words related with ?sunshine? appear at the beginning of the sentence.

Using the Application

The use of the application is simple.

  1. Type in the sentence in the empty box.DATA_ADSENSE

  2. Click ?Submit?

  3. The prediction result will show up below.

  4. The prediction result is ordered with the most possible one at the top.

Copyright © 2016 All Rights Reserved. Privacy Policy