Browse Prior Art Database

AIS Biometric Authentication for Instant Messaging Disclosure Number: IPCOM000010129D
Original Publication Date: 2002-Oct-24
Included in the Prior Art Database: 2002-Oct-24
Document File: 3 page(s) / 50K

Publishing Venue



Instant Messaging (IM) is one of the new "disruptible technologies" and its use is becoming very widespread. One problem with IM is how do you know to whom you are "talking" ? Even if a system is password protected it would be possible to sit down at unlocked workstation and pretend to be someone - or even break into the network protocol. A person's style of communication is very particular to them - given someone you know well, you would become suspicious if their style of communication suddenly changed. The idea presented here is use Machine Learning to learn a person's communication style. During a conversation the system would say if it thought the person you're talking to is who you think it is. The suggested implementation is to use a Artificial Immune System (AIS) for the Machine Learning. They are very good at unsupervised classification type problems. In addition they provide the ability to continually learn. This would be very important as communication styles will change over time.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 41% of the total text.

Page 1 of 3

AIS Biometric Authentication for Instant Messaging

Problem Domain

Instant Messaging (IM) has been classified as one of the new "disruptive technologies" Its use has become widespread both between people over the public internet, and also within corporate intranets.

    The problem to be addressed here is "How do you know who you are talking to?" Even if the system (machine or IM client) is password protected it still very possible that someone could sit down at a system and pretend to be someone else.

    If someone was to do this is it is quite possible that the person they were talking to (and trying to fool) might suspect that something was wrong. For example if I was talking to a friend who used a lot of "emticons" or abbreviations and they suddenly stopped using them I might be suspicious.

    The style of a person's communication is particular to them, and it is possible to recognise them by this style. Such techniques are well known for example
1. Morse Code. It was reported many years ago that a person's "fist" could be recognised.
2. Recent court cases have used the style of text messaging to show that messages were actually faked.
3. is a commercial implementation of a Biometric based general password scheme.

    Many features could be taken into account, for example speed of typing, spelling of words, grammar, use of various "texting" abbreviations etc. Some of the features might even not be apparent. For example I might realise that a friend always spelled "realise" with an "s" and not a "z".

Proposed System

    To use an Artificial Intelligent system to learn the style of a person's communication.

    The AI scheme that it is believed would be the most appropriate is an Artificial Immune System (AIS) - primarily because this is an unsupervised learning system capable of classification, in addition it is capable of continually learning and refining its classification.

System Areas AIS

    AIS lend themselves to this application well. They are an unsupervised learning technique - many of the factors that could be used recognise a person may not be consciously apparent. In addition that proportion of them related to a classification may not be apparent.

    They can continually learn. For example style will change as for example more texting abbreviations are learned, or if you develop RSI or break a wrist. Recognition Features

    A variety of features could be used for recognition. Some of the features are only available on the client side (Note that this system would be bi-directional authentication). As well as presence of these features, absence of certain features would be important. The aim would be to establish a profile of the communication, and

Page 2 of 3

then relate future communication to the profile.
1. Spelling - number of correct and incorrect spellings, compared to established traits.
2. Grammar - mistakes in grammar, and style of grammar. E.g. passive or active sentence construction. Or lack of sentence construction. Length...