In search for a method to decrease time spent responding to emails, Georgia Tech inventors have developed DejaVu, an algorithm that relies on inverse document frequency (IDF) and keyword matching to provide relevant suggestions during reply construction. DejaVu’s system architecture includes its client, information curator, and its information database. The DejaVu client is located on the user’s smartphone and interacts with the Information Curator on the cloud. The Information Curator maintains the Information Database and retrieves suggestions from it. 

The Information Curator parses emails in their entirety and stores them in the Information Database. Each entry in the database is indexed by a set of keywords extracted from it. The text of an entry in the information database is initially converted to lowercase and then split into constituent words, W(i). To capture the core context of the text in the index and to avoid duplicates of words that are close in meaning with each other, every word in W(i) is trimmed to its root. Each entry in the information database is then indexed on the set of roots of words in W(i). The email whose suggestions are to be extracted is parsed and the core context in the form of a list of keywords is extracted from it. The Information Curator then matches the keyword with the index of entries in the information database. The DejaVu client then uses a hybrid push/pull model for retrieving suggestions from the Information Curator.