System and method for detecting quitting intention based on electronic-communication dynamics

A non-transitory computer-readable storage medium storing instructions that when executed by a message server cause the message server to perform a method for detecting an entity's intention to quit an organization, the method comprising:

• obtaining, from a storage device associated to the message server, recent electronic-communication information for the entity, wherein the entity comprises at least one person;
• extracting, from the obtained information, a plurality of recent communication features, comprising a sentiment of the entity, word statistics, and a pre-determined pattern indicating a writing style of the entity;
• training a quitting intention model based on a feature-control value:
• wherein the feature-control value indicates a difference of a mean value for a respective recent communication feature for a quitting group and a mean for the same feature for an overall group comprising current and previous employees;
• wherein the model includes a model weight proportional to the difference of means for the respective feature;
• and
• wherein training further comprises adjusting a variance for the means, across a plurality of time intervals, to follow a gamma distribution;
• detecting a change over time intervals in a behavior from earlier features;
• computing a quitting-intention value for the entity, using the model weight and based on the detected change, wherein the quitting-intention value indicates a likelihood that the change corresponds to an intention to quit the organization;
• and
• presenting, on a display, the quitting-intention value for an end user.

An apparatus for detecting an entity's intention to quit an organization, comprising:

• a message server computer comprising a processor;
• an associated storage device that stores electronic-communication information for a plurality of users;
• a display that presents a quitting-intention value for an end user;
• and
• a memory storing instructions that, when executed by the processor, cause the message server computer to implement:
• a data-obtaining module that obtains, from the storage device associated to the message server computer, recent electronic-communication information associated with the entity under investigation, wherein the entity comprises at least one person;
• a feature-extracting module that extracts, from the obtained information, a plurality of recent communication features, comprising a sentiment of the entity, word statistics, and a pre-determined pattern indicating a writing style of the entity;
• a model-training module that trains a quitting intention model based on a feature-control value:
• wherein the feature-control value indicates a difference of a mean value for a respective recent communication feature for a quitting group and a mean for the same feature for an overall group comprising current and previous employees;
• wherein the model includes a model weight proportional to the difference of means for the respective feature;
• and
• wherein training further comprises adjusting a variance for the means, across a plurality of time intervals, to follow a gamma distribution;
• a change-detecting module that detects a change over time intervals in a behavior from earlier features;
• and
• a quitting-intention estimating module that computes a quitting-intention value for the entity, using the model weight and based on the detected change, wherein the quitting-intention value indicates a likelihood that the change corresponds to an intention to quit the organization.