In our approach, we monitor all the key events that the user types. Typing one key triggers a pair of key events: press and release, which we call a key stroke. We get the latency between pressing and releasing a key for each key that is typed, which is called PR-latency. For each two continuous keys typed, we get the latency between the release event of the first key and the press event of the second key, which is called RP-latency. PR-latency is always positive, because a key can only be released after it's pressed. RP-latency can be negative, because the second key can be pressed before the first key is released. Because of the resolution of the X-server, the resolution of the latencies that we can get is 10 ms. The PR-latencies and RP-latencies are two disjoint data sets. The PR-latencies and RP-latencies are grouped respectively in three different ways: bigram, trigram and word-gram. A Key event is a bigram event, a trigram event or a word-gram event.
Bigram: We group every two continuous key strokes into one bigram event and index it by the two keys.
Trigram: We group every three continuous key strokes into one trigram event and index it by the three keys.
Word-gram: We group every continuous set of key strokes that only contains letters, the CapsLock key and the Shift key into one word-gram event and index it by the keys.
The Cardinality of a key event is the number of keys in the event. A bigram event has a cardinality of two. A trigram event has a cardinality of three. A word-gram event has a cardinality of the length of the word. The Bigram model contains all the bigram events data. A Trigram model contains all the trigram events data. A Word-gram model contains all the word-gram events data. We call the data set of each index in the three models the index-set. The Cardinality of an index-set C[I] is the cardinality of the index of the index-set.