Target user directing method and apparatus and computer storage medium

1. A method comprising:

extracting, by at least one processor as a positive example feature for training a similarity model, a user feature of a seed user as a first positive example user;

extracting, by the at least one processor as a negative example feature for training the similarity model, a user feature of a first negative example user; and

training, by the at least one processor, the similarity model using the first positive example feature and the first negative example feature;

determining, by the at least one processor, a similarity between each of a plurality of candidate users and a seed user using the trained similarity model;

pushing, by the at least one processor, a digital advertisement to a social application interface of candidate users whose similarity with the seed user is greater than a threshold similarity;

extracting, by the at least one processor as a second positive example user, according to data of the pushed digital advertisement, a user performing a click operation, an attention operation, or a purchase operation on the pushed digital advertisement through the social application interface;

extracting, by the at least one processor as a second negative example user, according to data of the pushed digital advertisement, a user not performing the click operation, the attention operation and the purchase operation on the pushed digital advertisement through the social application interface; and

training, by the at least one processor, a conversion prediction model using an information feature of the pushed digital advertisement, a first user feature of the second positive example user and a second user feature of the second negative example user;

predicting, by the at least one processor using the trained conversion prediction model, a probability that each of the plurality of candidate users will perform a click operation, an attention operation, or a purchase operation on a to-be-delivered digital advertisement;

determining a first weight of the similarity and a second weight corresponding to the probability for each of the plurality of candidate users; and

calculating a plurality of directing scores based on the similarity, the first weight, the probability, the second weight, and the function relationship between the similarity and the probability; and

selecting, as one or more target users by the at least one processor, candidate users whose directing score satisfies a directing condition, from among the plurality of candidate users; and

pushing the to-be-delivered digital advertisement to a social application interface of the one or more target users.

2. The method according to claim 1 , wherein

the method further comprises:

outputting, using the similarity model, a core feature for determining the similarity, wherein the core feature is a same feature or a similar feature among a plurality of seed users,

wherein the one or more target users are further selected based on the core feature.

3. The method according to claim 1 , wherein the predicting comprises:

extracting an information feature of the to-be-delivered digital advertisement;

extracting a plurality of user features of the plurality of candidate users; and

inputting the information feature and the plurality of user features to the conversion prediction model, to predict the probability.

4. An apparatus comprising:

at least one memory configured to store computer program code; and

at least one processor configured to access the at least one memory and operate according to the computer program code, the computer program code including:

first training code configured to cause at least one of the at least one processor to:

extract, as a positive example feature for training a similarity model, a user feature of a seed user as a first positive example user;

extract, as a negative example feature for training the similarity model, a user feature of a first negative example user; and

train the similarity model using the positive example feature and the negative example feature;

determining code configured to cause at least one of the at least one processor to determine a similarity between each of a plurality of candidate users and a seed user by using the trained similarity model;

pushing code configured to cause at least one of the at least one processor to push a digital advertisement to a social application interface of candidate users whose similarity with the seed user is greater than a threshold similarity;

second training code configured to cause at least one of the at least one processor to:

extract, as a second positive example user, according to data of the pushed digital advertisement, a user performing a click operation, an attention operation or a purchase operation on the pushed digital advertisement through the social application interface;

extract, as a second negative example user, according to data of the pushed digital advertisement, a user not performing the click operation, the attention operation or the purchase operation on the pushed digital advertisement through the social application interface; and

train a conversion prediction model using an information feature of the pushed digital advertisement, a first user feature of the second positive example user and a second user feature of the second negative example user;

prediction code configured to cause at least one of the at least one processor to predict, by using the trained conversion prediction model, a probability that each of the plurality of candidate users performs a click operation, an attention operation or a purchase operation on a to-be-delivered digital advertisement;

selection code configured to cause at least one of the at least one processor to:

determine a first weight of the similarity and a second weight corresponding to the probability for each of the plurality of candidate users;

calculate a plurality of directing scores by using the similarity, the first weight, the probability, the second weight, and a function relationship between the similarity and the probability; and

select, as the one or more target users, candidate users whose directing score satisfies a directing condition, from among the plurality of candidate users; and

transmitting code configured to cause at least one of the at least one processor to push the to-be-delivered digital advertisement to a social application interface of the one or more target users.

5. The apparatus according to claim 4 , wherein the computer program code further comprises output code configured to cause at least one of the at least one processor to output, by using the similarity model, a core feature for determining the similarity, wherein the core feature is a same feature or a similar feature among a plurality of seed users,

wherein the one or more target users are further selected based on the core feature.

6. The apparatus according to claim 4 , wherein the prediction code is further configured to cause at least one of the at least one processor to:

extract an information feature of the to-be digital advertisement;

extract a plurality of user features of the plurality of candidate users; and

input the information feature and the plurality of user features to the conversion prediction model, to predict the probability.

7. A non-transitory computer readable storage medium storing a computer program which, when executed by a computer, performs the following operations:

determining a similarity between each of a plurality of candidate users and a seed user using a similarity model;

predicting, using a conversion prediction model, a probability that each of the plurality of candidate users will perform a click operation, an attention operation, or a purchase operation on a to-be-delivered digital advertisement;

selecting one or more target users for the to-be-delivered digital advertisement from the plurality of candidate users, according to the similarity that is determined and the probability that is predicted for each of the plurality of candidate users; and

pushing the to-be-delivered digital advertisement to a social application interface of the one or more target users,

wherein the selecting comprises:

calculating a plurality of directing scores of the plurality of candidate users using the similarity, the probability, and a function relationship between the similarity and the probability; and

selecting, as the one or more target users, candidate users whose directing score satisfies a directing condition, from among the plurality of candidate users,

wherein the calculating comprises:

determining a first weight of the similarity and a second weight corresponding to the probability for each of the plurality of candidate users; and

calculating the directing scores based on the similarity, the first weight, the probability, the second weight, and the function relationship,

wherein before determining the similarity, the method comprises:

extracting, as a positive example feature for training the similarity model, a user feature of the seed user as a first positive example user;

extracting, as a negative example feature for training the similarity model, a user feature of a first negative example user; and

training a similarity model using the positive example feature and the negative example feature, and

wherein, before predicting the probability, the method comprises:

pushing a digital advertisement to a social application interface of candidate users whose similarity with the seed user is greater than a threshold similarity;

extracting, as a second positive example user, according to data of the pushed digital advertisement, a user performing the click operation, the attention operation or the purchase operation on the pushed digital advertisement through the social application interface;

extracting, as a second negative example user, according to data of pushed digital advertisement, a user not performing the click operation, the attention operation or the purchase operation on the pushed digital advertisement through the social application interface; and

training a conversion prediction model using an information feature of the pushed digital advertisement, a first user feature of the second positive example user and a second user feature of the second negative example user.

8. The non-transitory computer readable storage medium according to claim 7 , wherein the computer program, when executed by the computer, further performs:

outputting, using the similarity model, a core feature for determining the similarity, wherein the core feature is a same feature or a similar feature among a plurality of seed users,

wherein the one or more target users are further selected based on the core feature.