Motivation

Helping users with a better typing experience by suggestion words bases on the content that the user had previously wrote:

• Reduce typing time
• Provide comfort with suitable suggestions
• User friendly solution for elders with problems wriing

Keyboatd.

Algorithm

The currently used algorithm is an ngram model. Particullary I have decided to used a trigram model with backoff. Which basically tries to use a trigram model but if no word is available it will backoff to use a bigram model. Ngram model tries to maximize likelihood based in probabilities and wordcounts of the current corpus. We dediced to use just the ~16,000 words which will give ~95% coverage os used words.

Functioning app

The current app gives suggestion of the next three words as the user is typing in the textbox. To select any suggestion just click the respective button. To have a better UX suggestions are only give when the last character of the input is a blank space.

As we are using a backoff model, It is guaranteed that the user always will have an input

trigrm

Future work

Although the current algorith is functional, it has a lot of improvement opportunities. Here are some action items to improve the model

• Increase n-gram (4-5). Currently we are using a trigram model
• Use a better tokenizer to compress the model. Lematize, stem, etc
• Reduce the vocabulary size so we can increase the number of ngrams and have a more context-aware mode
• Change algorithm to Neural Network bases approach