Authorship technologies
Novel distractorless authorship verification technology optionally combines with novel algorithms to solve authorship attribution as to an open set of candidates—such as without limitation by analyzing the voting of “mixture of experts” and outputting the result to a user using the following: if z (z=p i −p j √p i +p j −(p i −p j ) 2 /n) is larger than a first predetermined threshold then author j cannot be the correct author; or if z (z=p i −p j √p i +p j −(p i −p j ) 2 /n) is smaller than a second predetermined threshold then author i cannot be the correct author; or if no author garners significantly more votes than all other contenders then none of the named authors is the author of a document in question—in a number of novel applications. Personality profiling and authorship attribution may also be used to verify user identity to a computer.
1. A method for increasing the efficiency of a computer's verifying authorship of a first writing D, comprising computer-driven comparison of stylometrics in said first writing for which authorship verification is desired to stylometrics in at least one known authored material attributable to an author candidate for whom verification or negation of authorship of said first writing is desired, wherein only those writings of said author candidate are used for such comparison, comprising the steps of:
a) compiling a set of training data known to be written by author A by selecting at least one text authored by A and inputting said text into a computer such that said text constitutes said training data;
b) compiling via computer model from said training data by extracting linguistic or token-level features from said training data to create a feature vector and labeling said feature vector M=<m1, m2, . . . , mn>;
c) extracting a feature set, F, from said writing D for which authorship verification is desired, in the form of F=<f1, f2, . . . , fn>, where fi corresponds to all mi for all i;
d) choosing a distance function, δ, such that if δ(x,y)<δ(x,z), then x is “close to” or “more similar to” y than to z; and
e) assigning a threshold, t, by empirically analyzing the average δ values between writings authored by A, such that if δ(M,F)≤t, we accept that d and the training data are written by the same author, A, and, wherein when a second threshold p is added, with p<t, the authorship is accepted only if δ(M,F)≤p, and if p≤δ(M,F)≤t, an answer is declined, whereupon the results of step e) are output to a user, and further wherein by using only those writings of said author candidate for comparison the method yields benefits not possible with computer comparisons conducted with use of writings of two or more author candidates, followed by said method steps a) through e) resulting either or both of verification of an author's identity or verification of authorship of one or more texts as a consequence of said method steps.
2. A method for increasing the efficiency of a computer's verifying authorship of a first writing D, comprising computer-driven comparison of stylometrics in said first writing for which authorship verification is desired to stylometrics in at least one known authored material attributable to an author candidate for whom verification or negation of authorship of said first writing is desired, wherein only those writings of said author candidate are used for such comparison, comprising the steps of:
a) compiling a set of training data known to be written by author A by selecting at least one text authored by A and inputting said text into a computer such that said text constitutes said training data;
b) compiling via computer model from said training data by extracting linguistic or token-level features from said training data to create a feature vector and labeling said feature vector M=<m1, m2, . . . , mn>,
c) extracting a feature set, F, from said writing D for which authorship verification is desired, in the form of F=<f1, f2, . . . , fn>, where fi corresponds to all mi for all i;
d) choosing a distance function, δ, such that if δ(x,y)<δ(x,z), then x is “close to” or “more similar to” y than to z; and
e) selecting a monotonically decreasing function C(x) such that C((δ(M,F)) is the probability or confidence that D and the training data are written by the same person, and rendering said data as an output to a user, and further wherein by using only those writings of said author achieves benefits not possible compared to computer comparisons conducted with use of writings of two or more author candidates, followed by said method steps a) through e) resulting either or both of verification of an author's identity or verification of authorship of one or more texts as a consequence of said method steps.