Computerized systems and methods for user action prediction
The disclosed systems and methods provide a novel action prediction framework that performs personalized action prediction. According to an embodiment, the disclosed framework is able to dynamically predict which action (if any) a user might perform in response to receiving a given message. In some embodiments, for a given message, the action prediction framework can determine the probability that a user (e.g., sender, recipient) associated with the message may perform an action or set of action actions (e.g., open, forward, delete, reply, archive) related to the message. In some embodiments, the framework may be used to suggest a predicted action to the user. In some embodiments, a computing device may use the predicted actions to automatically perform the action. According to an embodiment, the action prediction framework includes a multi-label or multi-class model using a neural network.
1 . A method comprising:
identifying a message related to a user;
analyzing the message to extract at least one feature;
applying an action prediction model to the at least one feature, the action prediction model including at least one fully connected layer corresponding to each of the plurality of feature sets, such that applying the action prediction model comprises:
processing the plurality of feature sets to generate a plurality of corresponding feature vectors;
propagating each of the feature vectors through the at least one corresponding fully connected layer to generate a plurality of layer outputs;
concatenating all the layer outputs into an output vector; and
applying an activation function to the output vector;
determining an output of the action prediction model, the output including at least one action in relation to the message; and
causing display of the message and the output within a user interface (UI), the display enabling automatic execution of the at least one action in relation to the message.
2 . The method of claim 1 , wherein the at least one feature is a feature chosen from a group of features consisting of a local feature, a global feature, a content feature, and a mail-user-term-vector (MUTV) feature.
3 . The method of claim 2 , wherein the at least one local feature is a feature associated with a recipient of the message; and the at least one global feature is a feature associated with a sender of the message.
4 . The method of claim 2 , wherein the at least one content feature is a feature associated with data and metadata of the message.
5 . The method of claim 1 , further comprising:
receiving a request to provide the message;
providing the message and the at least one action;
receiving a selection to perform the at least one action; and
performing the at least one action.
6 . The method of claim 1 , wherein the action prediction model is a trained action prediction model, and training the action prediction model comprises:
identifying a training message and at least one associated known action from a training dataset;
analyzing the training message to extract at least one training message feature;
applying the action prediction model to the at least one training message feature;
determining an output of the action prediction model, the output including at least one predicted action that may be performed in relation to the training message;
calculating a loss of the action prediction model based on the at least one predicted action and the at least one associated known action; and
generating the trained action prediction model by updating the action prediction model based on the calculated loss.
7 . The method of claim 6 , wherein the action prediction model comprises a plurality of layers associated with the at least one training message feature.
8 . A non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that when executed by a computing device, cause the computing device to:
identify a message related to a user;
analyze the message to extract at least one feature;
apply an action prediction model to the at least one feature, the action prediction model including at least one fully connected layer corresponding to each of the plurality of feature sets, such that applying the action prediction model comprises:
processing the plurality of feature sets to generate a plurality of corresponding feature vectors;
propagating each of the feature vectors through the at least one corresponding fully connected layer to generate a plurality of layer outputs;
concatenating all the layer outputs into an output vector; and
applying an activation function to the output vector;
determine an output of the action prediction model, the output including at least one action in relation to the message; and
cause display of the message and the output within a user interface (UI), the display enabling automatic execution of the at least one action in relation to the message.
9 . The computer-readable storage medium of claim 8 , wherein the at least one feature is a feature chosen from a group of features consisting of a local feature, a global feature, a content feature, and a MUTV feature.
10 . The computer-readable storage medium of claim 9 , wherein the at least one local feature is a feature associated with a recipient of the message; and the at least one global feature is a feature associated with a sender of the message.
11 . The computer-readable storage medium of claim 9 , wherein the at least one content feature is a feature associated with a data and metadata of the message.
12 . The computer-readable storage medium of claim 8 , wherein the instructions further cause the computing device to:
receive a request to provide the message;
provide the message and the at least one action;
receive a selection to perform the at least one action; and
perform the at least one action.
13 . The computer-readable storage medium of claim 8 , wherein the action prediction model is a trained action prediction model, and wherein the instructions further cause the computing device to:
identify a training message and at least one associated known action from a training dataset;
analyze the training message to extract at least one training message feature;
apply the action prediction model to the at least one training message feature;
determine an output of the action prediction model, the output including at least one predicted action that may be performed in relation to the training message;
calculate a loss of the action prediction model based on the at least one predicted action and the at least one associated known action; and
generate the trained action prediction model by updating the action prediction model based on the calculated loss.
14 . The computer-readable storage medium of claim 13 , wherein the action prediction model comprises a plurality of layers associated with the at least one training message feature.
15 . A computing device comprising:
a processor configured to:
identify a message related to a user;
analyze the message to extract at least one feature;
apply an action prediction model to the at least one feature, the action prediction model including at least one fully connected layer corresponding to each of the plurality of feature sets, such that applying the action prediction model comprises:
processing the plurality of feature sets to generate a plurality of corresponding feature vectors;
propagating each of the feature vectors through the at least one corresponding fully connected layer to generate a plurality of layer outputs;
concatenating all the layer outputs into an output vector; and
applying an activation function to the output vector;
determine an output of the action prediction model, the output including at least one action in relation to the message; and
cause display of the message and the output within a user interface (UI), the display enabling automatic execution of the at least one action in relation to the message.
16 . The computing device of claim 15 , wherein the at least one feature is a feature chosen from a group of features consisting of a local feature, a global feature, a content feature, and a MUTV feature.
17 . The computing device of claim 15 , the processor is further configured to:
receive a request to provide the message;
provide the message and the at least one action;
receive a selection to perform the at least one action; and
perform the at least one action.
18 . The computing device of claim 15 , wherein the action prediction model is a trained action prediction model, the processor further configured to:
identify a training message and at least one associated known action from a training dataset;
analyze the training message to extract at least one training message feature;
apply the action prediction model to the at least one training message feature;
determine an output of the action prediction model, the output including at least one predicted action that may be performed in relation to the training message;
calculate a loss of the action prediction model based on the at least one predicted action and the at least one associated known action; and
generate the trained action prediction model by updating the action prediction model based on the calculated loss.