CN-116861065-B - Recommendation processing method, device, equipment, storage medium and program product

Abstract

The application provides a recommendation processing method, a recommendation processing device, electronic equipment, a computer readable storage medium and a computer program product based on artificial intelligence; the method comprises the steps of obtaining interaction characteristics and object characteristics of a target object account, carrying out first attention processing on the interaction characteristics based on a first object interest pool to obtain interaction interest characteristics, carrying out second attention processing on the object characteristics based on a second object interest pool to obtain object interest characteristics, obtaining liveness characteristics of the target object account, carrying out selection processing on the interaction interest characteristics and the object interest characteristics based on the liveness characteristics to obtain recall characteristics of the target object account, obtaining target information from information to be recommended based on the recall characteristics and the information characteristics of the information to be recommended, and carrying out recommendation operation on the target object account based on the target information. According to the application, accurate personalized recommendation can be realized and recommendation processing efficiency is improved.

Inventors

• DAI WEI

Assignees

• 腾讯科技（深圳）有限公司

Dates

Publication Date

20260505

Application Date

20220325

Claims (20)

1. 1. An artificial intelligence based recommendation processing method, the method comprising: Acquiring at least one interaction characteristic and at least one object characteristic of a target object account; Acquiring target first interest features corresponding to the interaction features from at least one first interest feature in a first object interest pool; performing aggregation processing on target interaction features corresponding to the target first interest features to obtain interaction interest features corresponding to the target first interest features; Acquiring target second interest features corresponding to the object features from at least one second interest feature in a second object interest pool; performing aggregation processing on target object features corresponding to the target second interest features to obtain object interest features corresponding to the target second interest features; wherein, the object activity corresponding to the first object interest pool is higher than the object activity corresponding to the second object interest pool; Acquiring liveness characteristics of the target object account, mapping the liveness characteristics, the at least one interactive interest characteristic and the at least one object interest characteristic respectively, carrying out mixed descending order on the at least one interactive interest characteristic and the at least one object interest characteristic based on the obtained weights corresponding to each interactive interest characteristic and each object interest characteristic, and determining at least one characteristic with the earlier order as a recall characteristic of the target object account; And acquiring target information from the plurality of information to be recommended based on the at least one recall feature and the information features of the plurality of information to be recommended, and executing recommendation operation to the target object account based on the target information.
2. 2. The method of claim 1, wherein the obtaining the target first interest feature corresponding to the interaction feature from at least one first interest feature in the first object interest pool comprises: Acquiring at least one first interest feature from the first object interest pool; and acquiring target first interest features corresponding to the interaction features from the at least one first interest feature aiming at each interaction feature.
3. 3. The method of claim 2, wherein the obtaining the target first interest feature corresponding to the interaction feature from the at least one first interest feature comprises: performing first mapping processing on each first interest feature to obtain a first key feature corresponding to each first interest feature; Performing second mapping processing on the interaction characteristics to obtain first query characteristics corresponding to the interaction characteristics; Determining a first weight corresponding to each first interest feature based on a first query feature corresponding to the interaction feature and a first key feature corresponding to each first interest feature; And determining the first interest feature with the highest first weight as a target first interest feature corresponding to the interaction feature.
4. 4. The method according to claim 2, wherein the aggregating the target interaction feature corresponding to the target first interest feature to obtain the interaction interest feature corresponding to the target first interest feature includes: performing third mapping processing on the target first interest feature to obtain a second query feature corresponding to the target first interest feature; performing fourth mapping processing on each target interaction feature to obtain a second key feature corresponding to each target interaction feature; Acquiring a first value characteristic corresponding to each target interaction characteristic; Determining a second weight corresponding to each target interaction feature based on a second query feature corresponding to the target first interest feature and a second key feature corresponding to each target interaction feature; And carrying out weighted summation processing on the first value feature corresponding to each target interaction feature based on the second weight corresponding to each target interaction feature to obtain the interaction interest feature corresponding to the target first interest feature.
5. 5. The method of claim 1, wherein the obtaining a target second feature of interest corresponding to the object feature from at least one second feature of interest in a second object interest pool comprises: acquiring at least one second interest feature from the second object interest pool; For each object feature, acquiring a target second interest feature corresponding to the object feature from the at least one second interest feature.
6. 6. The method of claim 5, wherein the obtaining a target second feature of interest corresponding to the object feature from the at least one second feature of interest comprises: performing fifth mapping processing on each second interest feature to obtain a third key feature corresponding to each second interest feature; Performing sixth mapping processing on the object features to obtain third query features corresponding to the object features; Determining a third weight corresponding to each of the second features of interest based on a third query feature corresponding to the object feature and a third key feature corresponding to each of the second features of interest; and determining the second interest feature with the highest third weight as a target second interest feature corresponding to the object feature.
7. 7. The method of claim 5, wherein the aggregating the object features corresponding to the second object feature of interest to obtain the object feature of interest corresponding to the second object feature of interest comprises: Performing seventh mapping processing on the target second interest feature to obtain a fourth query feature corresponding to the target second interest feature; performing eighth mapping processing on each target object feature to obtain a fourth key feature corresponding to each target object feature; Acquiring a second value characteristic corresponding to each target object characteristic; determining a fourth weight corresponding to each target object feature based on a fourth query feature corresponding to the target second interest feature and a fourth key feature corresponding to each target object feature; And carrying out weighted summation processing on the second value characteristic corresponding to each target object characteristic based on the fourth weight corresponding to each target object characteristic to obtain the object interest characteristic corresponding to the target second interest characteristic.
8. 8. The method of claim 1, wherein the mapping the liveness feature, the at least one interactive interest feature, and the at least one object interest feature separately, and wherein the mixed descending order of the at least one interactive interest feature and the at least one object interest feature based on the obtained weights for each of the interactive interest feature and each of the object interest features comprises: performing ninth mapping processing on the liveness characteristic to obtain a fifth query characteristic of the liveness characteristic; Performing tenth mapping processing on each interactive interest feature to obtain a fifth key feature corresponding to each interactive interest feature, and performing tenth mapping processing on each object interest feature to obtain a fifth key feature corresponding to each object interest feature; determining a fifth weight corresponding to each interactive interest feature and a fifth weight corresponding to each object interest feature based on a fifth query feature corresponding to the liveness feature, a fifth key feature corresponding to each interactive interest feature, and a fifth key feature corresponding to each object interest feature; And based on the fifth weight, performing mixed descending order sorting processing on the at least one interactive interest feature and the at least one object interest feature.
9. 9. The method of claim 1, wherein the information characteristic is obtained through an information tower network comprising N cascaded fully connected layers, N being an integer greater than or equal to 2; The method further comprises, before the target information is acquired from the plurality of information to be recommended based on the at least one recall feature and the information features of the plurality of information to be recommended: the following processing is executed for each piece of information to be recommended: The input of the N th full-connection layer is subjected to full-connection processing through the N th full-connection layer in the N cascade full-connection layers, so that an N th full-connection result is obtained; transmitting the n-th full connection result to an n+1-th full connection layer to continue full connection processing; When the value of N is 1, the input of the N-th full-connection layer is attribute information of the information to be recommended, when the value of N is 2-1, the input of the N-th full-connection layer is the full-connection result of the N-1-th full-connection layer, and the output of the N-th full-connection layer is the information characteristic.
10. 10. The method of claim 1, wherein the obtaining target information from the plurality of information to be recommended based on the at least one recall feature and information features of the plurality of information to be recommended comprises: For each of the recall features, performing the following: Determining the similarity between the recall feature and the information feature of each piece of information to be recommended; And based on the similarity of each piece of information to be recommended, performing descending order sorting processing on the plurality of pieces of information to be recommended, and determining at least one piece of information to be recommended with the top sorting as target information aiming at the recall characteristic.
11. 11. The method according to claim 1, wherein the aggregating the target object features corresponding to the target second interest feature to obtain the object interest feature corresponding to the target second interest feature and the aggregating the target object features corresponding to the target second interest feature are performed by an object tower network, the information feature of the information to be recommended is obtained by an information tower network, the processing of obtaining the target information is performed by a similarity network, and the multiple-interest recall model is composed of the object tower network, the information tower network, and the similarity network, the method further comprises: forward propagating at least one interaction feature and at least one object feature of a sample object account in the object tower network to obtain recall features of the sample object account; forward spreading a plurality of sample information to be recommended on the information tower network to obtain information characteristics of each sample information to be recommended; Determining target sample information to be recommended corresponding to each recall feature through the similarity network; Determining loss of the corresponding recall feature based on the predicted similarity and the pre-marked similarity of the target sample to-be-recommended information of the corresponding recall feature; Updating parameters of the multiple interest recall model based on a minimum in the loss corresponding to at least one of the recall features.
12. 12. An artificial intelligence based recommendation processing device, the device comprising: the acquisition module is used for acquiring at least one interaction characteristic and at least one object characteristic of the target object account; The attention module is used for acquiring target first interest characteristics corresponding to the interaction characteristics from at least one first interest characteristic in a first object interest pool, carrying out aggregation treatment on target interaction characteristics corresponding to the target first interest characteristics to obtain interaction interest characteristics corresponding to the target first interest characteristics, acquiring target second interest characteristics corresponding to the object characteristics from at least one second interest characteristic in a second object interest pool, and carrying out aggregation treatment on target object characteristics corresponding to the target second interest characteristics to obtain object interest characteristics corresponding to the target second interest characteristics; wherein, the object activity corresponding to the first object interest pool is higher than the object activity corresponding to the second object interest pool; The selection module is used for acquiring the liveness characteristic of the target object account, mapping the liveness characteristic, the at least one interactive interest characteristic and the at least one object interest characteristic respectively, carrying out mixed descending order on the at least one interactive interest characteristic and the at least one object interest characteristic based on the obtained weights corresponding to each interactive interest characteristic and each object interest characteristic, and determining at least one characteristic with the earlier order as a recall characteristic of the target object account; And the recommending module is used for acquiring target information from the plurality of information to be recommended based on the at least one recall characteristic and the information characteristics of the plurality of information to be recommended, and executing recommending operation to the target object account based on the target information.
13. 13. The apparatus of claim 12, wherein the device comprises a plurality of sensors, The attention module is further used for acquiring at least one first interest feature from the first object interest pool, and acquiring target first interest features corresponding to the interaction features from the at least one first interest feature for each interaction feature.
14. 14. The apparatus of claim 13, wherein the device comprises a plurality of sensors, The attention module is further used for performing first mapping processing on each first interest feature to obtain first key features corresponding to each first interest feature, performing second mapping processing on the interaction features to obtain first query features corresponding to the interaction features, determining first weights corresponding to each first interest feature based on the first query features corresponding to the interaction features and the first key features corresponding to each first interest feature, and determining the first interest feature with the highest first weights as a target first interest feature corresponding to the interaction features.
15. 15. The apparatus of claim 13, wherein the device comprises a plurality of sensors, The attention module is further used for performing third mapping processing on the target first interest feature to obtain a second query feature corresponding to the target first interest feature, performing fourth mapping processing on each target interaction feature to obtain a second key feature corresponding to each target interaction feature, obtaining a first value feature corresponding to each target interaction feature, determining a second weight corresponding to each target interaction feature based on the second query feature corresponding to the target first interest feature and the second key feature corresponding to each target interaction feature, and performing weighted summation processing on the first value feature corresponding to each target interaction feature based on the second weight corresponding to each target interaction feature to obtain an interaction interest feature corresponding to the target first interest feature.
16. 16. The apparatus of claim 12, wherein the device comprises a plurality of sensors, The attention module is further used for acquiring at least one second interest feature from the second object interest pool, and acquiring target second interest features corresponding to the object features from the at least one second interest feature for each object feature.
17. 17. The apparatus of claim 16, wherein the device comprises a plurality of sensors, The attention module is further configured to perform fifth mapping processing on each second interest feature to obtain a third key feature corresponding to each second interest feature, perform sixth mapping processing on the object feature to obtain a third query feature corresponding to the object feature, determine a third weight corresponding to each second interest feature based on the third query feature corresponding to the object feature and the third key feature corresponding to each second interest feature, and determine a second interest feature with the highest third weight as a target second interest feature corresponding to the object feature.
18. 18. The apparatus of claim 16, wherein the device comprises a plurality of sensors, The attention module is further configured to perform seventh mapping processing on the target second interest feature to obtain a fourth query feature corresponding to the target second interest feature, perform eighth mapping processing on each target object feature to obtain a fourth key feature corresponding to each target object feature, obtain a second value feature corresponding to each target object feature, determine a fourth weight corresponding to each target object feature based on the fourth query feature corresponding to the target second interest feature and the fourth key feature corresponding to each target object feature, and perform weighted summation processing on the second value feature corresponding to each target object feature based on the fourth weight corresponding to each target object feature to obtain an object interest feature corresponding to the target second interest feature.
19. 19. The apparatus of claim 12, wherein the device comprises a plurality of sensors, The selection module is further configured to perform a ninth mapping process on the liveness feature to obtain a fifth query feature of the liveness feature, perform a tenth mapping process on each of the interactive interest features to obtain a fifth key feature of each of the interactive interest features, perform a tenth mapping process on each of the object interest features to obtain a fifth key feature of each of the object interest features, determine a fifth weight of each of the interactive interest features and a fifth weight of each of the object interest features based on the fifth query feature of the liveness feature, the fifth key feature of each of the interactive interest features, and the fifth key feature of each of the object interest features, and perform a mixed descending order sorting process on the at least one of the interactive interest features and the at least one of the object interest features based on the fifth weight.
20. 20. The apparatus of claim 12, wherein the information characteristic is obtained through an information tower network comprising N cascaded fully connected layers, N being an integer greater than or equal to 2; The recommending module is further used for executing the following processing on each piece of information to be recommended before the target information is acquired from the plurality of pieces of information to be recommended based on the at least one recall feature and the information features of the plurality of pieces of information to be recommended, wherein the N is an integer with the value increasing from 1, the value range of N is less than or equal to 1 and less than or equal to N-1, when the value of N is 1, the input of the N is the full connection result of the N-1 th full connection layer, and when the value of N is 2 and less than or equal to N-1, the input of the N-th full connection layer is the full connection result of the N-1 th full connection layer, and the output of the N-th full connection layer is the information feature.

Description

Recommendation processing method, device, equipment, storage medium and program product Technical Field The present application relates to artificial intelligence technology, and in particular, to an artificial intelligence based recommendation processing method, apparatus, electronic device, computer readable storage medium and computer program product. Background Artificial intelligence (AI, artificial Intelligence) is the theory, method and technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, sense the environment, acquire knowledge and use the knowledge to obtain optimal results. Recommendation processing is an important application of artificial intelligence, recall is used as the front end of a recommendation system, and the upper limit and the lower limit of the whole recommendation system are determined. With the development of deep learning, the large-scale identified deep learning network is widely popularized and applied in various stages of a recommendation system. In the related art, recall recommendation is performed to a user based on an information sequence interacted with the user, and vectors cannot be accurately represented for users with scarce interaction, so that interest features of the user are difficult to learn effectively and accurately, and the information recommended by the mode cannot be matched with the interests of the user effectively, so that bad experience is caused to the user. Disclosure of Invention The embodiment of the application provides a recommendation processing method, a device, electronic equipment, a computer readable storage medium and a computer program product based on artificial intelligence, which can map interaction characteristics and object characteristics to different interest pools, screen recall characteristics based on liveness so as to more accurately predict target information, thereby realizing accurate personalized recommendation and improving recommendation processing efficiency. The technical scheme of the embodiment of the application is realized as follows: the embodiment of the application provides a recommendation processing method based on artificial intelligence, which comprises the following steps: Acquiring at least one interaction characteristic and at least one object characteristic of a target object account; performing first attention processing on the at least one interactive feature based on the first object interest pool to obtain at least one interactive interest feature, and performing second attention processing on the at least one object feature based on the second object interest pool to obtain at least one object interest feature; wherein, the object activity corresponding to the first object interest pool is higher than the object activity corresponding to the second object interest pool; Acquiring liveness characteristics of the target object account, and selecting and processing the at least one interactive interest characteristic and the at least one object interest characteristic based on the liveness characteristics to obtain at least one recall characteristic of the target object account; And acquiring target information from the plurality of information to be recommended based on the at least one recall feature and the information features of the plurality of information to be recommended, and executing recommendation operation to the target object account based on the target information. The embodiment of the application provides a recommendation processing device based on artificial intelligence, which comprises: the acquisition module is used for acquiring at least one interaction characteristic and at least one object characteristic of the target object account; The attention module is used for carrying out first attention processing on the at least one interactive feature based on the first object interest pool to obtain at least one interactive interest feature, and carrying out second attention processing on the at least one object feature based on the second object interest pool to obtain at least one object interest feature; wherein, the object activity corresponding to the first object interest pool is higher than the object activity corresponding to the second object interest pool; The selection module is used for acquiring the liveness characteristic of the target object account, and selecting and processing the at least one interactive interest characteristic and the at least one object interest characteristic based on the liveness characteristic to obtain at least one recall characteristic of the target object account; And the recommending module is used for acquiring target information from the plurality of information to be recommended based on the at least one recall characteristic and the information characteristics of the plurality of information to be recommended, and executing recommending