Mining email inboxes for suggesting actions
View Patent ↗Embodiments are directed towards automatically learning user behavioral patterns when interacting with messages and based on the learned patterns, suggesting one or more predicted actions that a user might take in response to receiving subsequent message. One or more classifiers are trained and employed to predict one or more actions that a user might take in response to receiving the message. In one embodiment, the one or more predicted actions are provided suggested to the user as an action the user might take on the message. Messages may be rank ordered within a given suggested action based on a confidence level of the prediction.
1. A method comprising:
training, by a processor, at least three learning classifiers, which comprise a horizontal classifier, a vertical general classifier, and a vertical pair-wise classifier,
the horizontal classifier is trained across a plurality of users' inboxes and based on actions upon messages within said inboxes to determine a suggested response,
the vertical general classifier is trained on a particular user's inbox and the particular user's actions upon messages within said inbox to predict a user-specific response to a given message for the particular user, and
the vertical pair-wise classifier is trained on the particular user's inbox and the particular user's actions upon messages within said inbox from a particular sender in order to predict a user-sender pair specific response to a given message;
receiving, by the processor, a first message directed towards a first user;
selecting, by the processor, one or more learning classifiers of the at least three learning classifiers, said selection comprising:
determining whether the first user is a new user,
responsive to the determination the first user is a new user, selecting only the horizontal classifier, and
responsive to the determination the first user is not a new user, further determining whether the first user has received and acted upon at least a second message from a sender of the said received message, and
responsive to determining that the user is not a new user and determining the first user has not received and acted upon at least the second message from the sender of the said received message, selecting only the vertical general classifier and the horizontal classifier, and
responsive to determining the first user is not a new user and determining the first user has received and acted upon at least a second message from the sender of the said received message, selecting the vertical pair-wise classifier, the vertical general classifier and the horizontal classifier;
predicting, for each of the selected classifiers, one or more actions the user is likely to perform on said received message by employing the selected classifiers and analyzing each of one or more actions, and based on said analysis, determining a weight for each of the one or more actions, wherein, when more than one classifier is selected, combining the predicted actions;
identifying, by the processor, one or more suggested actions from said one or more predicted actions based on said one more predicted actions having an associated weight satisfying a threshold;
transmitting, by the processor for display to the first user, said received message being transmitted with the one or more suggested actions for the first user to take on said received message; and
retraining, by the processor, one or more of the learning classifiers based on an action actually taken by the first user on said received message.
2. The or method of claim 1 , wherein, a plurality of historical actions are available to train at least one of the vertical classifiers.
3. The method of claim 1 , wherein at least one of the predicted actions is directed towards preserving a conversation associated with the message.
4. The method of claim 1 , wherein a plurality of messages, including said received message, having a same suggested action are rank ordered within the same suggested action based on a confidence value associated with a confidence of a prediction in the suggested action.
5. A network device, comprising:
a memory; and
a processor configured to;
train at least three learning classifiers, which comprise a horizontal classifier, a vertical general classifier, and a vertical pair-wise classifier,
the horizontal classifier being trained across a plurality of users' inboxes and based on actions upon messages within said inboxes to determine a suggested response,
the vertical general classifier being trained on a particular user's inbox and the particular user's actions upon messages within said inbox to predict a user-specific response to a given message for the particular user, and
the vertical pair-wise classifier being trained on the particular user's inbox and the particular user's actions upon messages within said inbox from a particular sender in order to predict a user-sender pair specific response to a given message;
receive a message directed towards a first user;
select one or more learning classifiers of the at least three learning classifiers, said selection comprising:
determining whether the first user is a new user,
responsive to the determination the first user is a new user, selecting only the horizontal classifier,
responsive to the determination the first user is not a new user, further determining whether the first user has received and acted upon at least a second message from a sender of the said received message, and
responsive to determining that the user is not a new user and determining the first user has not received and acted upon at least the second message from the sender of the said received message, selecting only the vertical general classifier and the horizontal classifier, and
responsive to determining the first user is not a new user and determining the first user has received and acted upon at least a second message from the sender of the said received message, selecting the vertical pair-wise classifier, the vertical general classifier and the horizontal classifier;
predict, for each of the selected classifiers, one or more actions the user is likely to perform on said received message by employing the selected classifiers and analyzing each of one or more actions, and based on said analysis, determining a weight for each of the one or more actions, wherein, when more than one classifier is selected, combining the predicted actions;
identify one or more suggested actions from said one or more predicted actions based on said one or more predicted actions having an associated weight satisfying a threshold;
transmit said received message for display to the first user, said received message being transmitted with the one or more suggested actions for the first user to take on said received message; and
retrain one or more of the learning classifiers based on an action actually taken by the first user on said received message.
6. The network device of claim 5 , wherein, a plurality of historical actions are available to train at least one of the vertical classifiers.
7. The network device of claim 5 , wherein at least one of the predicted actions is directed towards preserving a conversation associated with said message.
8. The network device of claim 5 , wherein the processor is further configured to:
determine a confidence value in the one or more predicted action for said message; and
rank, based on confidence value, said message within a plurality of other messages having a same suggested action.
9. A non-transitory computer readable storage medium tangibly storing thereon computer program instructions that, when executed by a processor of a network device, cause the network device to perform a method comprising:
training, by the processor, at least three learning classifiers, which comprise a horizontal classifier, a vertical general classifier, and a vertical pair-wise classifier,
the horizontal classifier is trained across a plurality of users' inboxes and based on actions upon messages within said inboxes to determine a suggested response,
the vertical general classifier is trained on a particular user's inbox and the particular user's actions upon messages within said inbox to predict a user-specific response to a given message for the particular user, and
the vertical pair-wise classifier is trained on the particular user's inbox and the particular user's actions upon messages within said inbox from a particular sender in order to predict a user-sender pair specific response to a given message;
receiving, by the processor, a first message directed towards a first user;
selecting, by the processor, one or more learning classifiers of the at least three learning classifiers, said selection comprising:
determining whether the first user is a new user,
responsive to the determination the first user is a new user, selecting only the horizontal classifier, and
responsive to the determination the first user is not a new user, further determining whether the first user has received and acted upon at least a second message from a sender of the said received message, and
responsive to determining that the user is not a new user and determining the first user has not received and acted upon at least the second message from the sender of the said received message, selecting only the vertical general classifier and the horizontal classifier, and
responsive to determining the first user is not a new user and determining the first user has received and acted upon at least a second message from the sender of the said received message, selecting the vertical pair-wise classifier, the vertical general classifier and the horizontal classifier;
predicting, for each of the selected classifiers, one or more actions the user is likely to perform on said received message by employing the selected classifiers and analyzing each of one or more actions, and based on said analysis, determining a weight for each of the one or more actions, wherein, when more than one classifier is selected, combining the predicted actions;
identifying, by the processor, one or more suggested actions from said one or more predicted actions based on said one or more predicted actions having an associated weight satisfying a threshold;
transmitting, by the processor for display to the first user, said received message being transmitted with the one or more suggested actions for the first user to take on said received message; and
retraining, by the processor, one or more of the learning classifiers based on an action actually taken by the first user on said received message.