Method, device, and computer program product for user behavior prediction
Embodiments of the present disclosure relate to a method, a device, and a computer program product for user behavior prediction. In some embodiments, at a client, a first user behavior embedding engine in the client generates behavior prediction information of a target user based on feature information of the target user. The client sends the behavior prediction information of the target user to a server, and receives information about a target item recommended for the target user from the server. Such method enables user privacy-related information to be processed only locally, thereby not only ensuring user privacy and security, but also significantly reducing overall resource overhead.
1. A method implemented at a client, comprising:
generating, by a first user behavior embedding engine in the client, behavior prediction information of a target user based on feature information of the target user, wherein the first user behavior embedding engine is implemented by a neural network, and wherein the neural network is compressed at the client;
mapping the feature information of the target user to the behavior prediction information of the target user;
sending the behavior prediction information of the target user to a server, the server comprising at least a second neural network; and
receiving information about a target item recommended for the target user from the server.
2. The method according to claim 1 , further comprising:
jointly training the first user behavior embedding engine and a first item embedding engine in the client by using a training data set, wherein the first item embedding engine is used to generate item recommendation information based on feature information of an item.
3. The method according to claim 2 , wherein jointly training the first user behavior embedding engine and the first item embedding engine comprises:
generating, by the first user behavior embedding engine, behavior prediction information of a reference user based on feature information of the reference user in the training data set;
generating, by the first item embedding engine, recommendation information of a reference item based on feature information of the reference item in the training data set;
determining a degree of matching between the reference user and the reference item based on the behavior prediction information of the reference user and the recommendation information of the reference item; and
updating the first user behavior embedding engine and the first item embedding engine based on comparison of the determined degree of matching with a reference degree of matching between the reference user and the target item in the training data set.
4. The method according to claim 2 , wherein the first user behavior embedding engine and the first item embedding engine are implemented by a deep semantic similarity model (DSSM), and the method further comprises:
sending a first group of parameters of the DSSM to the server after the training.
5. The method according to claim 4 , further comprising:
receiving a second group of parameters of the DSSM from the server; and
updating the DSSM based on the second group of parameters.
6. The method according to claim 2 , wherein the neural network comprises multiple branches, and each branch of the multiple branches has a weight associated with an influence of the branch on user behavior prediction.
7. The method according to claim 6 , further comprising:
updating the trained first user behavior embedding engine by deleting, based on multiple weights of the multiple branches, branches whose influences on the user behavior prediction are lower than a threshold influence from the multiple branches.
8. The method according to claim 7 , further comprising:
updating the trained first user behavior embedding engine by quantizing values of the multiple weights of the multiple branches.
9. A device implemented at a client, comprising:
a processor; and
a memory having computer-executable instructions stored therein, wherein the computer-executable instructions, when executed by the processor, cause the device to perform actions comprising:
generating, by a first user behavior embedding engine in the client, behavior prediction information of a target user based on feature information of the target user, wherein the first user behavior embedding engine is implemented by a neural network, and wherein the neural network is compressed at the client;
mapping the feature information of the target user to the behavior prediction information of the target user;
sending the behavior prediction information of the target user to a server, the server comprising at least a second neural network; and
receiving information about a target item recommended for the target user from the server.
10. The device according to claim 9 , wherein the actions further comprise:
jointly training the first user behavior embedding engine and a first item embedding engine in the client by using a training data set, wherein the first item embedding engine is used to generate item recommendation information based on feature information of an item.
11. The device according to claim 10 , wherein jointly training the first user behavior embedding engine and the first item embedding engine comprises:
generating, by the first user behavior embedding engine, behavior prediction information of a reference user based on feature information of the reference user in the training data set;
generating, by the first item embedding engine, recommendation information of a reference item based on feature information of the reference item in the training data set;
determining a matching degree of the reference user and the reference item based on the behavior prediction information of the reference user and the recommendation information of the reference item; and
updating the first user behavior embedding engine and the first item embedding engine based on comparison of the determined matching degree with a reference matching degree of the reference user and the target item in the training data set.
12. The device according to claim 10 , wherein the first user behavior embedding engine and the first item embedding engine are implemented by a deep semantic similarity model (DSSM), and the actions further comprise:
sending a first group of parameters of the DSSM to the server after the training.
13. A computer program product tangibly stored on a non-transitory computer-readable medium and comprising machine-executable instructions, wherein the machine-executable instructions, when executed, cause a machine to perform the method according to claim 1 .