Systems and methods for exchanging information in a large group

• US 8,335,827 B2
• Filed: 06/08/2009
• Issued: 12/18/2012
• Est. Priority Date: 07/11/2008
• Status: Expired due to Fees

• Current Assignee: Yuriy Mishchenko
• Original Assignee: Yuriy Mishchenko
• Inventors: Mishchenko, Yuriy
• Primary Examiner(s): DIVECHA, KAMAL B
• Application Number:
  US12/480,325
• Publication Number:
  US 20100011072A1
• Time in Patent Office: 1,289 Days
• Field of Search: 709/206, 709/207, 709/246
• US Class Current: 709/206
• CPC Class Codes: G06Q 10/10 Office automation; Time management

Abstract

Systems and methods are disclosed for actively all-to-all exchanging messages in a large group of individuals, comprising accumulating messages from individuals in at least one hub, processing and reorganizing accumulated messages and their groups into short representations of the messages, and selectively actively delivering short representations according to a distribution pattern found as a maximum of a goodness function. Processing and reorganizing accumulated messages into short representations, that is summarization, is performed by constraining the stream of messages to a hierarchical stream format, having a hierarchy of messages linked by a directional parent-child relationship, and iteratively replacing the parent messages in such hierarchy with a representative message chosen among all children of that message. Distribution pattern of the messages containing short representations is found by constrained maximization of a predetermined goodness function depending on the entire distribution pattern jointly for all messages and all users. The process of summarizing messages into short representations, and the process of calculating goodness function for distribution patterns are formulated in terms of statistical model of individuals'"'"' expertise and interests, which is learnt from observing past individuals'"'"' activity.

Claims

1. An electronically implemented method for actively exchanging messages in a group of individuals, comprising:

a. accumulating messages from individuals in at least one hub comprising a processor;

b. processing with the processor and reorganizing accumulated messages and groups of accumulated messages with similar contents into short representations without or with insignificant information loss for said single messages or said groups; and

c. selectively distributing, by the processor, messages containing the short representations in place of accumulated messages according to a distribution pattern achieving a constrained maximum of a goodness function depending on the distribution pattern and parameters describing group members and the messages,

wherein the goodness function is defined as a linear function, ƒ

(d)=Σ

<sub>X,Dε

G(X)</sub>PI_X(D), subject to a linear constraint of a form Σ

<sub>X,Dε

G(X)</sub>q_X(D)≦

r, where X identifies different individuals in the group, G(X) identifies a list of messages to be sent to individual X, D identifies one of the messages in the system, PI_X(D) q_x(D) are real valued coefficients depending on features of the message D and individual X, r is an acceptability constraint and the sum f(d) is over all individuals X and messages D in G(X) jointly and

wherein the coefficient PI_X(D) uses a probabilistic model of individuals'"'"' responses to messages, PI_X(D)=1−

Π

<sub>Yε

A(D)</sub>Π

_c(1−

RC_D(c)VI_X^Y(c)), where PI_X(D) is identified with the probability that individual X will respond positively, in a certain sense, to message D, and where Y identifies different individuals in the group, c identifies different categories to which a message can be related, A(D) identifies a list of authors of message D, RC_D(c) for all c identifies a set of features, valued between 0 and 1, describing the message D in relation to the set of categories c, and VI_X^Y(c) for all Y and c is a set of features, valued between 0 and 1, describing the individual X in relation to the set of categories c and message authors Y.

2. The method of claim 1, wherein said processing and reorganizing of accumulated messages is based on a hierarchical conversation format in which a hierarchy of messages comprises a directional parent-child relationship.

3. The method of claim 2, comprising applying a reduction model and a model of individuals'"'"' expertise to generate a short representation of the group of same-parent messages.

4. The method of claim 3, comprising determining the parameters for the model of individuals'"'"' expertise from observations of past individuals'"'"' activity.

5. The method of claim 1, wherein the active communication channel provides online “

live”

conversations between an individual and the group.

6. The method of claim 1, wherein the messages include text messages, voice messages, or video messages.

7. The method of claim 1, wherein the messages are accumulated and distributed by a website, an email, Short Message Service, Instant Messaging, or a mobile communication device.

8. The method of claim 1, comprising actively soliciting feedback from individuals to messages by distributing each message via email, Short Message Service or Instant Messages to a selected number of individuals according to predetermined criteria.

9. The method of claim 1, comprising selecting a distribution pattern of messages to users to maximize the goodness function, which depends on the features of messages and individuals and the set of distribution lists for all members either jointly or separately, and wherein messages are either processed before re-distribution or re-distributed in their original form.

10. A method for distributing messages to individuals in a communication system, comprising

a. accumulating messages from individuals in at least one hub comprising a processor;

b. processing with the processor and reorganizing accumulated messages and groups of accumulated messages with similar contents into short representations without or with insignificant information loss for said single messages or said groups; and

c. distributing, by the processor, messages containing short representations in place of accumulates messages to individuals according to one or more distribution lists, each distribution list specifying which messages are to be delivered to each individual, wherein the one or more distribution lists are chosen by finding a maximum of a goodness function that depends on the distribution lists and the parameters describing the individuals and the messages,

wherein the goodness function is defined as a linear function, ƒ

(d)=Σ

<sub>X,Dε

G(X)</sub>PI_X(D), subject to a linear constraint of the form Σ

<sub>X,Dε

G(X)</sub>q_X(D)≦

r, where X identifies different individuals in the group, G(X) identifies the list of messages to be sent to individual X, D identifies one of the messages in the system, PI_x(D) q_X(D) are real valued coefficients depending on the features of the message D and individual X, r is an acceptability constraint and the sum f(d) is over all individuals X and messages D in G(X) jointly and wherein the coefficient PI_X(D) uses a probabilistic model of individuals'"'"' responses to messages,

PI_X(D)=1−

Π

<sub>Yε

A(D)</sub>Π

_c(1−

RC_D(c)VI_X^Y(c)), where PI_X(D) is identified with the probability that individual X will respond positively, in a certain sense, to message D, and where Y identifies different individuals in the group, c identifies different categories to which a message can be related, A(D) identifies a list of authors of message D, RC_D(c) for all c identifies a set of features, valued between 0 and 1, describing the message D in relation to the set of categories c, and VI_X^Y(c) for all Y and c is a set of features, valued between 0 and 1, describing the individual X in relation to the set of categories c and message authors Y.

11. The method of claim 10, wherein the goodness function comprises a function of features describing all messages and all individuals and the set of all distribution lists for all individuals and all messages, simultaneously and jointly, wherein the maximum is subject to a constraint described in terms of features of all messages and all individuals and the distribution lists for all individuals and all messages.

12. The method of claim 10, comprising generating a probabilistic model of individuals'"'"' responses to distributed messages, that depends both on features describing the individuals and the messages, wherein this model is used to determine the goodness function.

13. The method of claim 10, wherein the goodness function is a sum of predetermined match-scores for assignment pairs, message-to-individual, from all distribution lists for all individuals and all messages.

14. The method of claim 12, comprising determining the parameters of the probabilistic model of individuals'"'"' responses, including both features describing the individuals and the messages, from observations of past individuals'"'"' activity.

15. The method of claim 14, comprising the probabilistic model of individuals'"'"' responses, wherein the features describing the individuals VI_X^Y(c) are updated per each observation of either positive or negative response from an individual to a distributed message according to

16. The method of claim 14, comprising the probabilistic model of individuals'"'"' responses, wherein the features describing the messages RC_D(c) are updated per each observation of either positive or negative response from an individual to a distributed message according to