Method and system for measuring interest levels of digital messages

• US 7,251,640 B2
• Filed: 09/14/2004
• Issued: 07/31/2007
• Est. Priority Date: 09/15/2003
• Status: Active Grant

• Current Assignee: Philippe Baumard
• Original Assignee: Philippe Baumard
• Inventors: Baumard, Philippe
• Primary Examiner(s): Tsang; Fan
• Assistant Examiner(s): Patel; Hemant
• Application Number:
  US10/940,463
• Publication Number:
  US 20050058261A1
• Time in Patent Office: 1,050 Days
• Field of Search: 706/45
• US Class Current: 706/45
• CPC Class Codes: G06Q 10/107 Computer-aided management of electronic mailing [e-mailing]
• CPC Class Codes: H04L 51/212 using filtering or selective blocking

Abstract

In an electronic communication system, relevance levels of an incoming or outgoing message for presenting it to an interlocutor is measured without having to actually interact with the interlocutor, by a method comprising the steps of extracting from the message, a flow of digital signals pertaining to transmission/reception context features, to content of the message and/or to other interlocutors with the interlocutor; weighting probabilistically the digital signals by means of indicators of relative and interrelated frequencies of occurrences of the same digital signals extracted from previous messages; from the results of the above steps, auto-generating a Bayesian network that allows the interlocutor to obtain a probabilistic prediction on the attractiveness of sent/received signals or messages, or find most probably interested interlocutors for a given information or message, each node of the Bayesian network being associated with a signal.

Claims

1. In an electronic communication system, a method for measuring attractiveness of incoming and outgoing messages, the method comprising the steps of:

extracting from each of the incoming and outgoing messages, a flow of digital signals pertaining to transmission or reception context features, to content of the message, and to other interlocutors in relation with a particular interlocutor;

measuring, by machine capture and computation, relative and interrelated frequencies of occurrences of the same digital signals extracted from previous messages;

measuring, by machine capture and computation, a frequency of feedback response, for each data content, interlocutor, and context feature;

measuring, by machine capture and computation, a frequency of feedback response for presence of a combination of content and interlocutor and context feature;

calculating, from the frequencies of feedback response, interrelated and relative probability of triggering an interaction between two interlocutors based on identification of presence of the combination of content, interlocutor, and context feature; and

from results of the extracting and measuring steps, auto-generating a Bayesian network without an initial model, the Bayesian network to be used to obtain a probabilistic prediction on an attractiveness of the incoming and outgoing messages, and to find most probably interested interlocutors for each outgoing message, each node of the Bayesian network being associated with a respective digital signal that represents one of a data content, interlocutor, and context feature.

2. A method according to claim 1 wherein new machine-captured incoming and outgoing signals that have not been previously captured by the machine, are added to the Bayesian network as representing new data, for which the relative frequencies and a shape of interrelations of the Bayesian network are updated in real-time, so as to generate newly discovered relative probabilistic frequencies of interrelation between the particular interlocutor and content and context feature, and the new incoming and outgoing signals so as to generate automated and autonomous machine learning, without prior model and without human handcrafted input and teaching.

3. A method according to claim 1 further comprising the step of extracting from an output of the machine learning the range from the highest to the lowest probabilistic levels of triggering an interaction for all learned combinations of data contents and context features with interlocutors, so as to find interlocutors having a highest likelihood and propensity to communicate on a given combination of subjects and context features.

4. A method according to claim 1, further comprising the step of displaying most probable other interlocutors for given signals, and most probable signals for given other interlocutors, by using a full reversibility between contents and interlocutors allowed by the step of auto-generating the Bayesian network.

5. A method according to claim 1, further comprising extending the step of auto-generating the Bayesian network to multiple interactions, with multiple signals and multiple interlocutors, by separating knowledge from inferences made in said step of auto-generating the Bayesian network, to find a most probable receiver of a message in a large population of other interlocutors.

6. A method according to claim 1, further comprising at least one of the steps of:

permitting the particular interlocutor to interrogate a learning module on most favorable features of his or her communication in order to increase the likelihood of a strong propensity to communicate from other interlocutors, including scoring in real time a message being composed, and showing different scores according to the message to be sent, before the message is sent;

permitting the particular interlocutor to compose a message without giving the name of a particular interlocutor, and interrogating a learning module to obtain a hierarchical probabilistic list of other interlocutors who would be most interested in the message;

permitting the particular interlocutor to fix a probability level of attractiveness under which the particular interlocutor does not want to interact, for each other interlocutor, whereby a learning module continuously learns automatically to carry on with the step of auto-generating without further human intervention.

7. A method according to claim 1, wherein the step of auto-generating a Bayesian network comprises generating a unique proper and specific model of communication behavior for each interlocutor where the models of interrelation between specific signals, contents, context of communication and characteristics and interlocutors are built from any data, through continuous learning without relying on an initial model.

8. A method according to claim 7, wherein the step of extracting from each message a flow of digital signals comprises preserving the totality of the information, wherein the step of weighting probabilistically the digital signals comprises weighting each signal conditionally and probabilistic as the interrelations between its content of information, and the emerging auto-generated models from continuous learning, and delivering a probabilistic assessment of the propensity to communicate, in regard of relative frequencies of signals, knowing the previous probabilities of propensities to communicate, interact, reply on relative frequencies of signals, the method further comprising the step of displaying a hierarchical output of most probable interlocutors, or most probable signals or text that would foster attractiveness for an outgoing or incoming result, provided by probabilistic computation of a learning module, without reconfiguration, judgment, or human and manual assessment from interlocutors, operators or platform owners.

9. A method according to claim 1, further comprising the step of displaying an assessment of a likelihood of feed-back and interest of other interlocutors, to be viewed by the particular interlocutor before sending the message for allowing the particular interlocutor to adjust communication in real-time.

10. A method according to claim 1, wherein the step of weighting the digital signals comprises measuring, by an indicator, a relative frequency of appearance of each digital signal on an overall population of incoming and outgoing signals, to provide a self-designing management of a knowledge base, and wherein the step of auto-generating the Bayesian network comprises measuring a number of interrelated nodes and externalities in the Bayesian network with the node associated to said each digital signal, to assess a mortality factor of said digital signal.

11. A system for measuring relevance levels of incoming and outgoing messages associated with a particular interlocutor, the system comprising:

means for extracting from each message a flow of digital signals pertaining to transmission or reception context features, to content of the message, and to other interlocutors in relation with the particular interlocutor;

means for weighting probabilistically the digital signals using indicators of relative and interrelated frequencies of occurrences of the same digital signals extracted from previous messages and for auto-generating, from the probabilistic weights, a Bayesian network without an initial model, the Bayesian network providing a probabilistic prediction on an attractiveness of the incoming and outgoing messages, and most probably interested other interlocutors for each outgoing message, each node of the Bayesian network being associated with a respective digital signal;

means for displaying the probabilistic prediction on the attractiveness of the incoming and outgoing messages, and the most probably interested other interlocutors for each outgoing message.

12. A system according to claim 11, further comprising means for updating in real-time the relative frequencies and the shape of interrelations of the Bayesian network so as to increase performance of predictions and accurateness of interactions.

13. A system according to claim 11, further comprising means for sharing personal Bayesian communication behavioral networks with other interlocutors, so as to allow other interlocutors to interrogate remotely with electronic communication means, the levels of likelihood of interest of this particular interlocutor, even when the particular interlocutor is absent and therefore not available for an interaction, to provide other interlocutors a highest likelihood and propensity to communicate on a given subject even in the absence of the particular interlocutor.

14. A system according to claim 11, wherein the display means are arranged to display the most probable other interlocutors for given signals, and the most probable signals for given other interlocutors, with a full reversibility between contents and individuals provided by the means for auto-generating the Bayesian network.

15. A system according to claim 11, further comprising means for separating knowledge from inferences, and a learning module for extending the Bayesian network to multiple interactions, with multiple signals and multiple individuals to find a most probable receiver of a message or signal in a population of other interlocutors.

16. A computer program product in a computer-readable storage medium in which instructions are embodied for measuring relevance levels of incoming and outgoing messages associated with a particular interlocutor, the instructions when executed causing a computer to:

extract from each message a flow of digital signals pertaining to transmission or reception context features, to content of the message, and to other interlocutors in relation with the particular interlocutor;

weight probabilistically the digital signals using indicators of relative and interrelated frequencies of occurrences of the same digital signals extracted from previous messages and for auto-generating, from the probabilistic weights, a Bayesian network without an initial model, the Bayesian network providing a probabilistic prediction on an attractiveness of the incoming and outgoing messages, and most probably interested other interlocutors for each outgoing message, each node of the Bayesian network being associated with a respective digital signal;

and display, in a display device, the probabilistic prediction on the attractiveness and a list of the most probably interested other interlocutors.

17. A computer program product according to claim 16, wherein the instructions when executed cause the computer to further update in real-time the relative frequencies and a shape of interrelations of the Bayesian network so as to increase performance of predictions and accurateness of interactions.

18. A computer program product according to claim 16, wherein the instructions when executed cause the computer to further share personal Bayesian communication behavioral networks with other interlocutors, to find interlocutors with highest likelihood and propensity to communicate on a given subject.

19. A computer program product according to claim 16, wherein the instructions for displaying are designed to display the most probable other interlocutors for given signals, and the most probable signals for given other interlocutors, by means of a full reversibility between contents and individuals provided by the instructions for auto-generating the Bayesian network.

20. A computer program product according to claim 16, wherein the instructions when executed cause the computer to further separate knowledge from inferences, wherein the instructions for auto-generating the Bayesian network comprise instructions for extending the Bayesian network to multiple interactions, with multiple signals and multiple individuals to find a most probable receiver of a message or signal in a population of other interlocutors.