Content recommendation method, computer storage medium and electronic equipment

Abstract

The embodiment of the invention provides a content recommendation method, a computer storage medium and electronic equipment. The method comprises the following steps: acquiring a plurality of candidate recommendation contents including candidate interest contents and candidate service contents; arranging the candidate interest contents in the first arrangement position range, and arranging the candidate service contents in the second arrangement position range to obtain a plurality of candidate recommendation sequences; the first arrangement position range in the candidate recommendation sequence is before the second arrangement position range; evaluating each candidate recommendation sequence to obtain an evaluation result corresponding to each candidate recommendation sequence; and according to the evaluation result corresponding to each candidate recommendation sequence, determining a target recommendation sequence from the plurality of candidate recommendation sequences. According to the method, the influence of the relative arrangement of the candidate recommendation contents in the recommendation sequence on the whole sequence is considered, and the candidate interest contents are ranked in front of the candidate service contents, so that the content ranking of the recommendation sequence starts from the interest of the user, and the use experience of the user is improved.

Description

Technical Field

[0001] Embodiments of the present application relate to the field of computer technologies, and in particular, to a content recommendation method, a computer storage medium, and an electronic device. Background Art

[0002] With the continuous development of Internet technologies, network platforms are increasingly widely used in users' daily work and life. To ensure the attractiveness of the platform and improve users' experience, network platforms often perform personalized content recommendations for different users.

[0003] In the prior art, the content recommendation method for different users by network platforms has a poor recommendation effect, resulting in a poor user experience, easy user loss, and further having an adverse impact on the operation of network platforms. Summary of the Invention

[0004] In view of this, embodiments of the present application provide a content recommendation solution to at least partially solve the above problems.

[0005] According to a first aspect of embodiments of the present application, a content recommendation method is provided, including: obtaining a plurality of candidate recommended contents; the plurality of candidate recommended contents include candidate interest contents and candidate service contents; within a first arrangement position range, performing permutations and combinations on the candidate interest contents, and within a second arrangement position range, performing permutations and combinations on the candidate service contents to obtain a plurality of candidate recommended sequences; wherein, in the candidate recommended sequences, the first arrangement position range is before the second arrangement position range; respectively evaluating each candidate recommended sequence to obtain an evaluation result corresponding to each candidate recommended sequence; the evaluation result corresponding to the candidate recommended sequence represents the evaluation result at the sequence level of the entire candidate recommended sequence; determining a target recommended sequence from the plurality of candidate recommended sequences according to the evaluation results corresponding to each candidate recommended sequence, so as to perform content recommendation based on the target recommended sequence.

[0006] According to a second aspect of embodiments of the present application, a content recommendation method is provided, which is applied to a user terminal and includes: in response to a first operation of a user on the user terminal, sending a content recommendation request to a server; receiving the target recommended sequence sent by the server based on the content recommendation request, where the target recommended sequence is determined by the server through respectively evaluating a plurality of candidate recommended sequences obtained by permutations and combinations of a plurality of candidate recommended contents and according to the evaluation results from the plurality of candidate recommended sequences; wherein, the plurality of candidate recommended contents include candidate interest contents and candidate service contents; in the candidate recommended sequence, the first arrangement position range where the candidate interest content is located is before the second arrangement position range where the candidate service content is located; and performing content display on the user terminal according to the target recommended sequence.

[0007] According to a third aspect of an embodiment of the present application, a content recommendation device is provided, including: an acquisition module, configured to acquire a plurality of candidate recommendation contents; the plurality of candidate recommendation contents include candidate interest contents and candidate service contents; a sorting module, configured to perform permutations and combinations on the candidate interest contents within a first permutation position range, and perform permutations and combinations on the candidate service contents within a second permutation position range, to obtain a plurality of candidate recommendation sequences; wherein, in the candidate recommendation sequences, the first permutation position range is before the second permutation position range; an evaluation module, configured to evaluate each candidate recommendation sequence respectively, to obtain an evaluation result corresponding to each candidate recommendation sequence; the evaluation result corresponding to the candidate recommendation sequence represents the sequence-level evaluation result of the overall candidate recommendation sequence; a selection module, configured to determine a target recommendation sequence from the plurality of candidate recommendation sequences according to the evaluation results corresponding to the candidate recommendation sequences, so as to perform content recommendation based on the target recommendation sequence.

[0008] According to a fourth aspect of an embodiment of the present application, a content recommendation device is provided, which is applied to a user terminal and includes: a sending module, configured to send a content recommendation request to a server in response to a first operation of a user on the user terminal; a receiving module, configured to receive a target recommendation sequence sent by the server based on the content recommendation request, the target recommendation sequence being determined by the server through evaluating a plurality of candidate recommendation sequences obtained by permutations and combinations of a plurality of candidate recommendation contents respectively, and according to the evaluation results; wherein, the plurality of candidate recommendation contents include candidate interest contents and candidate service contents; in the candidate recommendation sequence, the first permutation position range where the candidate interest content is located is before the second permutation position range where the candidate service content is located; a display module, configured to perform content display on the user terminal according to the target recommendation sequence.

[0009] According to a fifth aspect of an embodiment of the present application, an electronic device is provided, including: a processor, a communication interface, a memory, and a communication bus, wherein, the processor, the communication interface, and the memory complete communication with each other through the communication bus; the memory is configured to store a computer program; the processor is configured to execute the content recommendation method described in the foregoing first aspect or second aspect by running the computer program stored on the memory.

[0010] According to a sixth aspect of an embodiment of the present application, a computer storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the content recommendation method described in the first aspect or second aspect is implemented.

[0011] According to a seventh aspect of an embodiment of the present application, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the content recommendation method described in the first aspect or second aspect is implemented.

[0012] According to the content recommendation solution provided by the embodiments of the present application, first, a plurality of candidate recommendation contents are obtained; the plurality of candidate recommendation contents include candidate interest contents and candidate service contents; within the first arrangement position range, the candidate interest contents are arranged and combined, and within the second arrangement position range, the candidate service contents are arranged and combined to obtain a plurality of candidate recommendation sequences; wherein, in the candidate recommendation sequence, the first arrangement position range is before the second arrangement position range; then, each candidate recommendation sequence is comprehensively evaluated to obtain an evaluation result corresponding to each candidate recommendation sequence; finally, according to the evaluation results corresponding to each candidate recommendation sequence, a target recommendation sequence is determined from the plurality of candidate recommendation sequences. The content recommendation method of this embodiment does not perform a single-point evaluation on a single candidate recommendation content, but adopts a method of mixed arrangement of candidate recommendation contents to generate candidate recommendation sequences that simultaneously include a plurality of candidate recommendation contents. Furthermore, the candidate recommendation sequences are used as a whole evaluation sample for evaluating the quality of the entire sequence. Then, according to the evaluation results at the sequence level, an overall sequence recommendation is made to the user. The solution provided by the embodiments of the present application can select a target recommendation sequence with a better overall evaluation result for content recommendation. Since the target recommendation sequence is obtained by comprehensively evaluating the candidate recommendation sequences, the evaluation process considers the influence of the relative arrangement of each candidate recommendation content within the recommendation sequence on the overall value of the sequence, and the evaluation dimension is more comprehensive. Therefore, the evaluation result can be more accurate, and further, the accuracy of the recommendation result can be improved, and the user experience can be enhanced. Further, in the candidate recommendation sequence, arranging the position of the candidate interest content before the candidate service content can make the content sorting of each candidate recommendation sequence start from the user's interests, which can enhance the user experience, increase the user retention rate, and further increase the traffic obtained by subsequent service contents, ensuring the overall recommendation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0014] Figure 1 FIG. is a schematic diagram of a content recommendation system according to an embodiment of the present application;

[0015] Figure 2 FIG. is a flowchart of the steps of a content recommendation method according to an embodiment of the present application;

[0016] Figure 3A and 3B FIG. is a schematic diagram of the process of a content recommendation method according to an embodiment of the present application;

[0017] Figure 4 A flowchart of the steps of a content recommendation method according to another exemplary embodiment of the present application;

[0018] Figure 5 A schematic diagram of a scenario of an example of the content recommendation method according to an embodiment of the present application;

[0019] Figure 6 A structural block diagram of a content recommendation device according to an embodiment of the present application;

[0020] Figure 7 A structural block diagram of a content recommendation device according to another exemplary embodiment of the present application;

[0021] Figure 8 A schematic diagram of the structure of an electronic device according to an embodiment of the present application. Detailed implementation manners

[0022] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art shall fall within the protection scope of the embodiments of the present application.

[0023] The following further illustrates the specific implementation of the embodiments of the present application with reference to the accompanying drawings of the embodiments of the present application.

[0024] Figure 1 An exemplary system applicable to the solution of the embodiment of the present application is shown. As Figure 1 shown, the system 100 may include a server 102, a communication network 104, and / or one or more user terminals 106, Figure 1 exemplified as multiple user terminals 106 in, and an application for displaying content is provided on the user terminal 106.

[0025] The server 102 can be any suitable device for storing information, data, programs, and / or any other suitable type of content, including but not limited to distributed storage system devices, server clusters, computing server clusters, etc. In some embodiments, the server 102 can perform any suitable function. For example, when implementing the solution of the embodiments of the present application by the server 102, in some embodiments, the server 102 can be used to execute a content recommendation method. As an optional example, in some embodiments, the server 102 can first obtain a plurality of candidate recommended contents, where the plurality of candidate recommended contents include candidate interest contents and candidate service contents; within a first arrangement position range, perform permutations and combinations on the candidate interest contents, and within a second arrangement position range, perform permutations and combinations on the candidate service contents to obtain a plurality of candidate recommended sequences; where, in the candidate recommended sequence, the first arrangement position range is before the second arrangement position range; further, respectively evaluate each candidate recommended sequence to obtain an evaluation result corresponding to each candidate recommended sequence; the evaluation result corresponding to the candidate recommended sequence represents the sequence-level evaluation result of the entire candidate recommended sequence; finally, determine a target recommended sequence from the plurality of candidate recommended sequences according to the evaluation results corresponding to each candidate recommended sequence, so as to perform content recommendation based on the target recommended sequence. In some embodiments, the server 102 can receive a content recommendation request sent by the user terminal 106, and after generating the target recommended sequence in the foregoing manner, send the target recommended sequence to the user terminal 106, so that the user terminal 106 can display content according to the target recommended sequence.

[0026] In some embodiments, the communication network 104 can be any suitable combination of one or more wired and / or wireless networks. For example, the communication network 104 can include any one or more of the following: the Internet, an intranet, a Wide Area Network (WAN), a Local Area Network (LAN), a wireless network, a Digital Subscriber Line (DSL) network, a Frame Relay network, an Asynchronous Transfer Mode (ATM) network, a Virtual Private Network (VPN), and / or any other suitable communication network. The user terminal 106 can be connected to the communication network 104 through one or more communication links (for example, the communication link 112), and the communication network 104 can be linked to the server 102 through one or more communication links (for example, the communication link 114). The communication link can be any communication link suitable for transmitting data between the user terminal 106 and the server 102, such as a network link, a dial-up link, a wireless link, a hard-wired link, any other suitable communication link, or any suitable combination of such links.

[0027] Optionally, the user terminal 106 may be provided with an application for presenting content. The user terminal 106 may include any one or more user terminals suitable for presenting content, interacting with users, etc. In some embodiments, the user terminal 106 may include any suitable type of device. For example, in some embodiments, the user terminal 106 may include a mobile device, a tablet computer, a laptop computer, a desktop computer, and / or any other suitable type of user terminal.

[0028] Based on the above system, embodiments of the present application provide a content recommendation solution, which will be described below through multiple embodiments.

[0029] Figure 2 It is a flowchart of the steps of a content recommendation method according to an embodiment of the present application. According to the first aspect in the embodiments of the present application, a content recommendation method is provided, and this method can be Figure 1 executed by the server 102 in the system shown. Referring to Figure 2 shown, this method includes steps S202, S204, and S206. Specifically:

[0030] S202. Obtain multiple candidate recommended contents.

[0031] Among them, the multiple candidate recommended contents include candidate interest contents and candidate service contents;

[0032] The candidate recommended contents in this embodiment may be the contents to be recommended to the user. Here, neither the data type of the candidate recommended contents nor the content category to which they belong is limited. Exemplarily, for a single candidate recommended content, its data type may be any one of the following: text type, image type, video type, audio type, etc., and this embodiment does not limit this. For a single candidate recommended content, the content category to which it belongs includes an interest category and a service category, and the service category to which it belongs may be any one of the following: live broadcast service, goods selling service, short drama video service, storytelling service, and so on. Further, to improve the user experience, for the same candidate recommendation sequence, the candidate recommended contents included therein may respectively belong to different data types, and / or respectively belong to the interest category and different service categories, so as to realize the multi-modal and multi-service of the candidate recommendation sequence content.

[0033] The content information included in the candidate recommended contents may be determined by the content category operated by the network platform of the recommended contents. The content category may include an interest category and a service category. Among them, the content belonging to the interest category may refer to the content that does not include a consumption nature, and the content belonging to the service category may refer to the content that includes a consumption nature.

[0034] Based on the attribute information of the user to be recommended, multiple candidate recommended contents can be matched and obtained from the content library of the network platform for the user to be recommended. Among them, the attribute information of the user to be recommended can include the basic information and preference information of the user. The basic information can include age, gender, region where the user is located, etc., and the preference information can include historical browsing information, historical operation information, etc. The multiple candidate recommended contents obtained can belong to different content categories. Optionally, at least two of the multiple candidate recommended contents belong to different content categories. Optionally, the multiple candidate recommended contents include candidate interest contents and candidate service contents.

[0035] S204. Within the first arrangement position range, perform permutations and combinations on the candidate interest contents, and within the second arrangement position range, perform permutations and combinations on the candidate service contents to obtain multiple candidate recommended sequences.

[0036] Among them, in the candidate recommended sequence, the first arrangement position range is before the second arrangement position range.

[0037] In this embodiment, the positions in a single candidate recommended sequence can be divided in advance according to the sequence length of the single candidate recommended sequence, and the first arrangement position range corresponding to the candidate interest contents and the second arrangement position range corresponding to the candidate service contents can be obtained. Here, the first arrangement position range corresponding to the candidate interest contents is before the second arrangement position range corresponding to the candidate service contents. Then, based on the determination of the first arrangement position range and the second arrangement position range, the candidate interest contents and the candidate service contents are permuted and combined within the corresponding position ranges, and multiple candidate recommended sequences can be obtained. By arranging the positions of the candidate interest contents before the candidate service contents, the content sorting of each candidate recommended sequence can start from the user's interests, which can improve the user experience and increase the user retention rate. Furthermore, the traffic obtained by subsequent service contents can be increased, ensuring the overall recommendation efficiency.

[0038] The sequence lengths of the multiple candidate recommended sequences in this embodiment can be equal. The sequence length refers to the number of candidate recommended contents included in the candidate recommended sequence, and the sequence length can be flexibly set by those skilled in the art according to the actual situation, and this embodiment does not limit it. Among the multiple candidate recommended sequences, the multiple candidate recommended contents included in any two candidate recommended sequences are different, or the multiple candidate recommended contents included in any two candidate recommended sequences can be the same, but the positions of the multiple candidate recommended contents in the candidate recommended sequence are different.

[0039] In some alternative embodiments, within the first arrangement position range, permutations and combinations of candidate interest contents are performed, and within the second arrangement position range, permutations and combinations of candidate service contents are performed to obtain a plurality of candidate recommendation sequences, including: selecting target interest contents from the candidate interest contents and target service contents from the candidate service contents according to the attribute information of the user to be recommended; performing permutations and combinations of the target interest contents within the first arrangement position range and performing permutations and combinations of the target service contents within the second arrangement position range to obtain a plurality of candidate recommendation sequences; wherein, in the candidate recommendation sequences, the positions of the target interest contents are before the target service contents.

[0040] Exemplarily, the plurality of candidate recommendation contents may include a plurality of candidate interest contents and a plurality of candidate service contents. The matching degrees of the plurality of candidate interest contents and the plurality of candidate service contents may be sorted in advance based on the attribute information of the user. Then, according to the order of the matching degree sorting, a first preset number of candidate interest contents may be selected from the candidate interest contents as the target interest contents, and a second preset number of candidate service contents may be selected from the candidate service contents as the target service contents. Here, the first preset number corresponds to the first arrangement position range, the second preset number corresponds to the second arrangement position range, and the sum of the first preset number and the second preset number is equal to the sequence length of the candidate recommendation sequence. Finally, within the first arrangement position range, permutations and combinations of the target interest contents are performed, and within the second arrangement position range, permutations and combinations of the target service contents are performed to obtain a plurality of candidate recommendation sequences; a plurality of candidate recommendation sequences may be obtained. In addition to ensuring that the positions of the target interest contents are before the positions of the target service contents, this method may also reduce the number of permutations and combinations and improve the processing efficiency.

[0041] In this embodiment, by separately selecting target interest contents and target service contents from the candidate interest contents and the candidate service contents, the target interest contents and the target service contents may be determined according to the attribute information of the user, thereby improving the overall satisfaction of the user with the candidate recommendation sequences and reducing the number of candidate recommendation contents for permutations and combinations. By performing mixed permutation combinations on the selected target interest contents and target service contents to obtain each candidate recommendation sequence, the number of candidate recommendation sequences for subsequent evaluation may be reduced, and the efficiency of subsequent evaluation processing may be improved. In addition, by arranging the positions of the target interest contents before the target service contents, the content sorting of each candidate recommendation sequence may start from the user's interests, improving the user experience, increasing the user retention rate, and then increasing the traffic obtained by the service contents, ensuring the overall recommendation efficiency.

[0042] In some alternative embodiments, selecting target service content from candidate service content includes: for a preset plurality of service categories, respectively determining the candidate service content corresponding to each service category; determining a plurality of target service categories from the plurality of service categories, and determining target service content from the candidate service content corresponding to the target service categories; wherein, each target service content included in the same candidate recommendation sequence corresponds to a different service category.

[0043] Exemplarily, the preset plurality of service categories may be a plurality of service categories matched from the service categories of the network platform according to the attribute information of the user to be recommended. For example, referring to Figure 3A , a schematic diagram of the process of selecting target service content from candidate service content is shown. The service categories of the candidate service content may include live broadcast service, short drama video service, storytelling service, and goods selling service, and the preset plurality of service categories may be live broadcast service, short drama video service, and storytelling service, or at least one of the plurality of service categories. Then, based on the attribute information of the user to be recommended, from the content library of the network platform, the candidate service content corresponding to the preset plurality of service categories is respectively obtained. For example, it may be sorted according to the matching degree, and a preset number of candidate service content is obtained for each service category. Then, for a single candidate recommendation queue, each target service content included therein corresponds to a different service category. Therefore, the position range corresponding to the target service content in the candidate recommendation queue can be determined in advance. Based on this position range, at least one target service category can be determined from the plurality of service categories. For example, referring to Figure 3A , the plurality of service categories may be live broadcast service, short drama video service, and storytelling service, the sequence length of a single candidate recommendation queue is 4, and the pre-determined position range corresponding to the target service content is the 3rd position and the 4th position. Then, two of them can be determined from live broadcast service, short drama video service, and storytelling service as the target service categories. Further, the target service content is determined from the candidate service content corresponding to the target service categories. The target service content may be any one of the candidate service content corresponding to the target service category, or, according to the matching degree sorting of the candidate service content corresponding to the target service category, any one within the preset sorting range can be selected. For example, referring to Figure 3A , the target service categories include live broadcast service and short drama video service. The candidate service content corresponding to the live broadcast service is live broadcast content 1, live broadcast content 2, and live broadcast content 3, and the candidate service content corresponding to the short drama video service is short drama content 1, short drama content 2, and short drama content 3. Then, live broadcast content 1 and short drama content 1 can be selected as the target service content, or, live broadcast content 1 and short drama content 2 can be selected as the target service content, and so on. Finally, the selected target service content is sorted and combined within the second sorting position range corresponding to the target service content.

[0044] In this embodiment, when selecting target service content from candidate service content, by determining multiple target service categories from multiple service categories and respectively determining target service content from the candidate service content corresponding to the target service categories, each target service content included in the same candidate recommendation sequence corresponds to a different service category, so as to avoid including target service content of the same service category in the obtained candidate recommendation sequence and improve the diversity of the recommended content in the same candidate recommendation sequence.

[0045] In some alternative embodiments, there are multiple candidate interest contents. Selecting target interest content from the candidate interest contents includes: inputting the attribute information of the user to be recommended and each candidate interest content into a multi-objective learning model for evaluation to obtain an initial evaluation result corresponding to each candidate interest content; sorting each candidate interest content based on the initial evaluation result corresponding to each candidate interest content to obtain an initial sorting result corresponding to each candidate interest content; and selecting target interest content from each candidate interest content based on the initial sorting result corresponding to each candidate interest content.

[0046] Exemplarily, for the first arrangement position range corresponding to the interest category in a single candidate recommendation queue, target interest content can be determined from the candidate interest content corresponding to the interest category according to the quantity requirement of the first arrangement position range and the order of the matching degree sorting based on the attribute information of the user to be recommended.

[0047] Furthermore, the attribute information of the user to be recommended and each candidate interest content can be input into a pre-trained multi-objective learning model for multi-objective evaluation to output a multi-objective initial evaluation result corresponding to each candidate interest content; then, based on the sorting of the multi-objective initial evaluation result corresponding to each candidate interest content in the entire multi-objective initial evaluation result, the multi-objective initial evaluation result corresponding to each candidate interest content is normalized to obtain a processed multi-objective initial evaluation result; linear interpolation processing is performed on the processed multi-objective initial evaluation result to obtain an initial evaluation result corresponding to each candidate interest content. After that, each candidate interest content is sorted according to the initial evaluation result corresponding to each candidate interest content to obtain an initial sorting result corresponding to each candidate interest content; then, target interest content is selected from each candidate interest content according to the initial sorting result corresponding to each candidate interest content. For example, referring to Figure 3A , interest category content 1 and interest category content 2 can be selected as target interest content. Then, the selected target interest content is sorted and combined within the first sorting position range corresponding to the target interest content. Optionally, it can be set that the target interest content is sorted within the first sorting position range according to its corresponding initial sorting result.

[0048] In this embodiment, by inputting the attribute information of the user to be recommended and each candidate interest content into a multi-objective learning model for evaluation, an initial evaluation result corresponding to each candidate interest content is obtained; then, based on the initial evaluation results corresponding to each candidate interest content, each candidate interest content is sorted to obtain an initial sorting result corresponding to each candidate interest content; and by selecting the target interest content according to the initial sorting result corresponding to each candidate interest content, the matching degree between the interest content included in each candidate recommendation sequence and the user to be recommended can be improved, and the satisfaction of the user with the final target recommendation sequence can be enhanced.

[0049] In some alternative embodiments, after step 202 and before step 204, the method of this embodiment further includes: based on a sequence screening strategy, performing sequence screening processing on a plurality of candidate recommendation sequences to obtain a plurality of screened sequences; respectively evaluating each candidate recommendation sequence to obtain an evaluation result corresponding to each candidate recommendation sequence, including: respectively evaluating each screened sequence to obtain an evaluation result corresponding to each screened sequence; and determining a target recommendation sequence from the plurality of candidate recommendation sequences according to the evaluation results corresponding to each candidate recommendation sequence, including: determining a target recommendation sequence from the plurality of screened sequences according to the evaluation results corresponding to each screened sequence.

[0050] Exemplarily, the sequence screening strategy is used to perform screening processing on a plurality of candidate recommendation sequences to reduce the number of candidate recommendation sequences for subsequent evaluation. For example, screening processing can be performed on a plurality of candidate recommendation sequences according to the situation of the recommendation data of each service category. Alternatively, screening processing can also be performed on a plurality of candidate recommendation sequences based on the consideration of the diversity of the service categories corresponding to the recommended content, and so on. This embodiment does not limit this.

[0051] In this embodiment, after obtaining a plurality of candidate recommendation sequences, based on the sequence screening strategy, performing sequence screening processing on the plurality of candidate recommendation sequences to obtain a plurality of screened sequences realizes the pruning process of the plurality of candidate recommendation sequences. Then, by respectively evaluating each screened sequence, the amount of data in the evaluation process is reduced, the efficiency of the evaluation process is improved, and thus the efficiency of the entire content recommendation process is improved.

[0052] In some alternative embodiments, performing sequence screening on a plurality of candidate recommendation sequences based on the sequence screening strategy to obtain a plurality of screened sequences may include: determining an over-service category from a plurality of service categories, where the over-service category is a service category for which the recommendation data reaches a preset recommendation threshold; determining an over-candidate recommendation sequence from each candidate recommendation sequence, where the over-candidate recommendation sequence includes candidate recommendation content belonging to the over-service category; and deleting the over-candidate recommendation sequences in the plurality of candidate recommendation sequences to obtain a plurality of screened sequences.

[0053] Exemplarily, the recommendation data of multiple service categories can be counted. If the recommendation data corresponding to a certain service category reaches a preset recommendation threshold, then this service category is determined as an over-service category. Here, the preset recommendation threshold can be flexibly set by those skilled in the art according to the actual situation, and this embodiment does not limit it. Then, from the obtained multiple candidate recommendation sequences, the candidate recommendation sequences containing candidate recommendation content belonging to the over-service category are screened out, and this candidate recommendation sequence is determined as an over-candidate recommendation sequence. Then, from the multiple candidate recommendation sequences, the determined over-candidate recommendation sequence is deleted, and multiple screened sequences can be obtained.

[0054] In this embodiment, after the recommendation data of the service category reaches the preset recommendation threshold, this service category is determined as an over-service category, and then the candidate recommendation sequences containing candidate recommendation content belonging to the over-service category are deleted from the multiple candidate recommendation sequences, so as to balance the recommendation data of each service category, avoid the situation that the recommendation data of some service categories is difficult to reach the target expectation, and ensure that each service category can obtain stable traffic.

[0055] In some alternative embodiments, based on the sequence screening strategy, when performing sequence screening on multiple candidate recommendation sequences to obtain multiple screened sequences, it may further include: determining similar service category pairs from multiple service categories, where the similar service category pairs include multiple service categories with similar service contents; determining similar sequences from each candidate recommendation sequence, where the similar sequences include multiple candidate recommendation contents belonging to the same similar service category pair; and deleting the similar sequences from the multiple candidate recommendation sequences to obtain multiple screened sequences.

[0056] Exemplarily, those skilled in the art can preset similar service categories as similar service category pairs in advance. After obtaining multiple candidate recommendation sequences, the candidate recommendation sequences containing multiple candidate recommendation contents belonging to the same similar service category pair are screened out, and this candidate recommendation sequence is determined as a similar sequence. For example, service category B and service category C are similar service category pairs. Service category B includes candidate recommendation contents B1, B2, B3, etc., and service category C includes candidate recommendation contents C1, C2, C3, etc. If a certain candidate recommendation sequence includes a combination of candidate recommendation content B1 (or B2, B3) and candidate recommendation content C2 (or C1, C3), it means that this candidate recommendation sequence is a similar sequence. Then, from the multiple candidate recommendation sequences, the determined similar sequence is deleted, and multiple screened sequences can be obtained.

[0057] In this embodiment, by determining a candidate recommendation sequence containing multiple candidate recommendation contents belonging to the same similar service category pair as a similar sequence and deleting the similar sequence from multiple candidate recommendation sequences, it can be ensured that the candidate recommendation sequences recommended to the user include diverse service contents, avoid the repeated occurrence of similar service contents, and improve the user experience.

[0058] It should be noted that for the above two sequence screening strategies, the processing methods for screening multiple candidate recommendation sequences can be selected for separate use or combined use. This embodiment does not limit this. In addition, if combined use is adopted, the order of use can also be flexibly set by those skilled in the art according to the actual situation. This embodiment does not limit this.

[0059] S206. Evaluate each candidate recommendation sequence respectively to obtain the evaluation result corresponding to each candidate recommendation sequence.

[0060] Among them, the evaluation result corresponding to the candidate recommendation sequence represents the evaluation result at the sequence level of the candidate recommendation sequence as a whole.

[0061] Exemplarily, when evaluating each candidate recommendation sequence, it is possible to first evaluate the multiple candidate recommendation contents included in a single candidate recommendation sequence respectively, then evaluate the positions of the candidate recommendation contents in a single candidate recommendation sequence, and finally, by fusing the evaluation results of the two, the comprehensive evaluation result corresponding to each candidate recommendation content can be obtained. Then, by fusing the comprehensive evaluation results corresponding to the candidate recommendation contents in a single candidate recommendation sequence, the evaluation result corresponding to the single candidate recommendation sequence can be obtained. Further, when comprehensively evaluating each candidate recommendation content, the preference information of the user to be recommended for each service category can also be combined to obtain the final comprehensive evaluation result corresponding to each candidate recommendation content, and so on. The evaluation result corresponding to the candidate recommendation sequence represents the evaluation result at the sequence level of the candidate recommendation sequence as a whole, that is, the evaluation of the candidate recommendation sequence is a comprehensive evaluation of the candidate recommendation sequence as a whole. The evaluation result indicates the sequence level obtained by comprehensively evaluating each candidate recommendation sequence as a whole. The presentation form of the evaluation result corresponding to the candidate recommendation sequence can be the evaluation score of the candidate recommendation sequence, and so on.

[0062] In some alternative embodiments, evaluating each candidate recommendation sequence separately to obtain the evaluation results corresponding to each candidate recommendation sequence may include: for a single candidate recommendation sequence among multiple candidate recommendation sequences, using a multi-objective learning model to evaluate to obtain the basic evaluation results corresponding to each candidate recommendation content in the single candidate recommendation sequence; obtaining the position evaluation results of each candidate recommendation content in the single candidate recommendation sequence, where the position evaluation results represent the exposure information of the positions where each candidate recommendation content is located in the single candidate recommendation sequence; fusing the basic evaluation results and the position evaluation results to obtain the comprehensive evaluation results corresponding to each candidate recommendation content; and fusing the comprehensive evaluation results corresponding to each candidate recommendation content to obtain the evaluation result corresponding to the single candidate recommendation sequence.

[0063] Exemplarily, a multi-objective learning model for multi-objective evaluation of each candidate recommendation content may be pre-trained. For example, the multi-objective learning model may be a PLE (Progressive Layered Extraction model) model, and the PLE model may be composed of multiple layers of CGC (Customized Gate Control) networks. Additionally, the corresponding position evaluation results may be determined in advance for each position in a single candidate recommendation sequence. Here, the position evaluation results may represent the exposure information of each position in the single candidate recommendation sequence. For example, the exposure probabilities of different positions may be analyzed based on the position sorting of historical recommendation content and the corresponding exposure data, and the corresponding position evaluation results may be set for each position in the single candidate recommendation sequence based on this. It can be understood that for each candidate recommendation sequence, the position evaluation results of the same position are the same.

[0064] After obtaining multiple candidate recommendation sequences, each candidate recommendation content in a single candidate recommendation sequence can be input into a multi-objective learning model. The multi-objective learning model performs multi-objective evaluation on each candidate recommendation content. The evaluation metrics of the multi-objective learning model can include the 5-second complete playback rate, click-through rate, total complete playback rate, playback duration, interaction data, 3-second short playback rate, etc. Among them, the 3-second short playback rate refers to the probability that the playback duration is less than or equal to 3 seconds. After obtaining the evaluation results (such as scores) corresponding to each evaluation metric, linear interpolation is used to comprehensively calculate the evaluation results corresponding to multiple evaluation metrics, and finally the basic evaluation results corresponding to each candidate recommendation content can be output. Then, the position evaluation results of each candidate recommendation content in a single candidate recommendation sequence are obtained. For a single candidate recommendation content, the basic evaluation result corresponding to this candidate recommendation content and the position evaluation result are fused to obtain the comprehensive evaluation result corresponding to each candidate recommendation content. The fusion method can adopt mathematical calculation methods such as multiplication and addition, and this embodiment does not limit this. Then, the comprehensive evaluation results corresponding to each candidate recommendation content included in a single candidate recommendation sequence are fused to obtain the evaluation result corresponding to the single candidate recommendation sequence. The fusion method can adopt mathematical calculation methods such as multiplication and addition, and this embodiment does not limit this.

[0065] In this embodiment, considering that users' acceptance of different recommendation positions is different, resulting in different exposure rates for different positions, therefore, when evaluating candidate recommendation sequences, based on the basic evaluation results obtained from content evaluation of each candidate recommendation content, the position evaluation results of each candidate recommendation content in a single candidate recommendation sequence are further fused, that is, the evaluation factor of exposure information of different recommended content arranged in different recommended positions is introduced, which improves the comprehensiveness of the evaluation results corresponding to each candidate recommendation sequence. Further, based on the evaluation results obtained, for the target recommendation sequence, after recommending it to the user, from the feedback results of the user on the target recommendation sequence, the influence of non-user preference factors on the feedback results of the target recommendation sequence can be excluded, so as to better and more accurately feedback the user's preference information for the target recommendation sequence, which is convenient for subsequent iteration of user preference factors.

[0066] In some alternative embodiments, the basic evaluation results corresponding to each candidate recommendation content in a single candidate recommendation sequence are obtained through multi-objective learning model evaluation, including: inputting the attribute information of the user to be recommended and each candidate recommendation content in each candidate recommendation sequence into a pre-trained multi-objective learning model, and outputting the multi-objective evaluation results corresponding to each candidate recommendation content; normalizing the multi-objective evaluation results corresponding to each candidate recommendation content based on the sorting of the multi-objective evaluation results corresponding to each candidate recommendation content to obtain the processed multi-objective evaluation results; performing linear interpolation processing on the processed multi-objective evaluation results corresponding to each candidate recommendation content in a single candidate recommendation sequence to obtain the basic evaluation results corresponding to each candidate recommendation content in the single candidate recommendation sequence.

[0067] Exemplarily, inputting the attribute information of the user to be recommended and each candidate recommendation content in each candidate recommendation sequence into a pre-trained multi-objective learning model, and outputting the multi-objective evaluation results corresponding to each candidate recommendation content, such as the evaluation results corresponding to multiple objectives such as the 5-second complete play rate, click-through rate, total complete play rate, play duration, interaction data, and 3-second short play rate. Then, for a single objective, determine the sorting of the evaluation results of each candidate recommendation content corresponding to this objective in the entire evaluation results corresponding to this objective, and then normalize the multi-objective evaluation results corresponding to each candidate recommendation content according to the sorting of the multi-objective evaluation results corresponding to each candidate recommendation content, that is, perform sequential normalization processing, to obtain the processed multi-objective evaluation results. The processed multi-objective evaluation results can be values between 0 and 1; finally, perform linear interpolation processing on the processed multi-objective evaluation results corresponding to each candidate recommendation content in a single candidate recommendation sequence to obtain the basic evaluation results corresponding to each candidate recommendation content in the single candidate recommendation sequence.

[0068] In this embodiment, the attribute information of the user to be recommended and each candidate recommendation content in each candidate recommendation sequence are processed through a multi-objective learning model to output the multi-objective evaluation results corresponding to each candidate recommendation content; then, sequential normalization processing is performed on the multi-objective evaluation results corresponding to each candidate recommendation content to obtain the processed multi-objective evaluation results. After that, linear interpolation processing is performed on the processed multi-objective evaluation results corresponding to each candidate recommendation content in a single candidate recommendation sequence to calculate the basic evaluation results corresponding to each candidate recommendation content, realizing multi-objective comprehensive evaluation of candidate recommendation content, and the calculation process is simple, which can improve data processing efficiency.

[0069] In some alternative embodiments, the multi-objective learning model includes a first expert network, a first gating network, a second expert network, a second gating network, and a tower network; inputting the attribute information of the user to be recommended and each candidate recommendation content in each candidate recommendation sequence into a pre-trained multi-objective learning model, and outputting the multi-objective evaluation results corresponding to each candidate recommendation content, including:

[0070] For each candidate recommendation content in each candidate recommendation sequence, the attribute information of the user to be recommended and a single candidate recommendation content are input into the first shared expert network and the first target expert networks corresponding to each target in the first expert network for feature extraction, and the first features corresponding to each target and the first shared feature are output;

[0071] The attribute information of the user to be recommended, a single candidate recommendation content, and the first features corresponding to each target and the first shared feature are respectively input into the first gating networks corresponding to each target for feature fusion, and the intermediate features corresponding to each target are output;

[0072] The intermediate features corresponding to each target are input into the second shared expert network and the second target expert networks corresponding to each target in the second expert network for feature extraction, and the second features corresponding to each target and the second shared feature are output;

[0073] The intermediate features corresponding to each target, and the second features corresponding to each target and the second shared feature are respectively input into the second gating networks corresponding to each target for feature fusion, and the target features corresponding to each target are output;

[0074] The target features corresponding to each target are input into the tower networks corresponding to each target, and the multi-target evaluation results corresponding to the candidate recommendation content are output.

[0075] Exemplarily, the pre-trained multi-target learning model may include a first expert network, a first gating network, a second expert network, a second gating network, and a tower network, wherein the first expert network includes a first shared expert network and the first target expert networks corresponding to each target, the second expert network includes a second shared expert network and the second target expert networks corresponding to each target, and the first gating network, the second gating network, and the tower network all correspond to the number of targets.

[0076] For a single candidate recommended content, the attribute information of the user to be recommended and the single candidate recommended content can be respectively input into the first shared expert network and the first target expert network corresponding to each target for feature extraction, and the first shared feature and the first feature corresponding to each target can be output. Subsequently, the first feature corresponding to each target can be respectively input into the first gating network corresponding to each target, and at the same time, the attribute information of the user to be recommended, the single candidate recommended content, and the first shared feature are all input into the first gating network corresponding to each target. Through the first gating network for feature fusion, the intermediate feature corresponding to each target can be output. Then, the intermediate feature corresponding to each target can be respectively input into the second shared expert network and the second target expert network corresponding to each target for feature extraction again, and the second shared feature and the second feature corresponding to each target can be output. Furthermore, the intermediate feature and the second feature corresponding to each target are respectively input into the second gating network corresponding to each target, and at the same time, the second shared feature is also input into the second gating network corresponding to each target. Through the second gating network for feature fusion, the target feature corresponding to each target can be output. Finally, the target feature corresponding to each target is input into the tower network corresponding to each target, and the multi-target evaluation result corresponding to the single candidate recommended content can be output.

[0077] In this embodiment, by setting that the multi-target learning model includes the first gating network and the second gating network for feature fusion processing, the accuracy of the inference result of the multi-target learning model can be improved while ensuring fewer model parameters.

[0078] In some alternative embodiments, before fusing the basic evaluation result and the position evaluation result, the method of this embodiment further includes: determining the acceptance evaluation result of each candidate recommended content, where the acceptance evaluation result represents the degree of acceptance of the candidate recommended content by the user to be recommended; fusing the basic evaluation result and the position evaluation result to obtain the comprehensive evaluation result corresponding to each candidate recommended content, including: fusing the basic evaluation result, the position evaluation result, and the acceptance evaluation result to obtain the comprehensive evaluation result corresponding to each candidate recommended content.

[0079] Exemplarily, the preference information of the user to be recommended can be obtained, such as historical browsing information, historical operation information, etc. Here, "historical" refers to before the current moment. Based on the preference information of the user to be recommended, the acceptance evaluation results of the user to be recommended for the interest categories and each service category are analyzed, that is, the preference acceptance degrees of the user to be recommended for the interest categories and each service category. For the interest categories, further sub-category subdivision can be performed according to the interest content, so that the preference acceptance degrees of the user to be recommended for each sub-category can be analyzed based on the preference information of the user to be recommended, and the acceptance evaluation results of each sub-category can be determined. Then, based on the acceptance evaluation results of the user to be recommended for the interest categories and each service category, and the attribution relationships between each candidate recommendation content in the candidate recommendation sequence and the interest categories and each service category, the acceptance evaluation results of the user to be recommended for each candidate recommendation content in the candidate recommendation sequence can be determined. For the attribution relationships between each candidate recommendation content and the interest categories, they can be subdivided into the attribution relationships between each candidate recommendation content in the candidate recommendation sequence and each sub-category of the interest categories. Furthermore, for each candidate recommendation content, by integrating its corresponding basic evaluation result, position evaluation result, and acceptance evaluation result, the comprehensive evaluation result corresponding to each candidate recommendation content can be obtained.

[0080] Referring to Figure 3B , an exemplary process for evaluating each candidate recommendation sequence is shown. As shown in the figure, assume that the obtained candidate recommendation sequence A includes 4 candidate recommendation contents. Then, the 4 candidate recommendation contents are respectively input into the multi-objective learning model for evaluation, and the basic evaluation results corresponding to the candidate recommendation contents are output by the multi-objective learning model. Furthermore, based on the basic evaluation results corresponding to the candidate recommendation contents, the position evaluation results of the candidate recommendation contents in a single candidate recommendation sequence are integrated. Then, based on the above evaluation results, the acceptance evaluation results of the user to be recommended for the candidate recommendation contents are further integrated, and the comprehensive evaluation results corresponding to the candidate recommendation contents can be obtained. Finally, the comprehensive evaluation results of the 4 candidate recommendation contents in a single candidate recommendation sequence can be summed to obtain the evaluation result corresponding to the candidate recommendation sequence A.

[0081] In this embodiment, when evaluating the candidate recommendation sequence, based on the basic evaluation results obtained by content evaluation of each candidate recommendation content, after integrating the position evaluation results of each candidate recommendation content in a single candidate recommendation sequence, the acceptance evaluation results of the user to be recommended for the candidate recommendation contents are further integrated, so that the evaluation results of the candidate recommendation sequence can be adjusted in a timely manner according to the user's preference information, making the final target recommendation sequence closer to the user's preference and improving the user experience.

[0082] In some alternative embodiments, determining the acceptance evaluation results of each candidate recommended content includes: obtaining the acceptance information of the user to be recommended for the historical recommendation sequence; the acceptance information is used to represent the degree of acceptance and preference of the user to be recommended for each target recommended content in the historical recommendation sequence; generating the acceptance evaluation results of the service categories to which the target recommended contents belong according to the acceptance information; and determining the acceptance evaluation results of the candidate recommended contents according to the service categories to which the candidate recommended contents belong.

[0083] Exemplarily, in the process of content recommendation for the user to be recommended based on the target recommendation sequence, the acceptance information of the user to be recommended for the historical recommendation sequence can be obtained in real time; the acceptance information is used to represent the degree of acceptance and preference of the user to be recommended for each target recommended content in the historical recommendation sequence. For example, the acceptance information of the user to be recommended for the historical recommendation sequence can be obtained according to the user's like operation, report operation, block operation, non-recommendation operation, etc. on the target recommended content. According to the acceptance information corresponding to each target recommended content in the historical recommendation sequence, the acceptance evaluation results of the service categories to which the target recommended contents belong can be generated. Then, when determining the acceptance evaluation results of each candidate recommended content, the acceptance evaluation results of the service categories to which the candidate recommended contents belong can be determined as the acceptance evaluation results of the candidate recommended contents.

[0084] In this embodiment, the acceptance information of the user to be recommended for each target recommended content in the historical recommendation sequence can be obtained in a timely manner, and the acceptance evaluation results of the service categories to which the target recommended contents belong can be generated and updated. Thus, the acceptance evaluation results of each candidate recommended content can be determined according to the service categories to which the candidate recommended contents belong. This method can adjust the evaluation results of the subsequent candidate recommendation sequences in a timely manner, so that the target recommendation sequence recommended to the user to be recommended next time can be closer to the user's preferences and improve the user experience.

[0085] S208. Determine a target recommendation sequence from multiple candidate recommendation sequences according to the evaluation results corresponding to each candidate recommendation sequence, so as to perform content recommendation based on the target recommendation sequence.

[0086] Exemplarily, assuming that the evaluation results are presented in the form of evaluation scores, the candidate recommendation sequence with the highest evaluation score can be determined as the target recommendation sequence from multiple candidate recommendation sequences according to the magnitude relationship between the evaluation scores corresponding to each candidate recommendation sequence. After determining the target recommendation sequence, the target recommendation sequence can be sent to the user terminal, and the target recommendation sequence can include multiple target recommended contents sorted according to the target positions.

[0087] In some alternative embodiments, the evaluation results are presented in the form of evaluation scores; after obtaining the evaluation results corresponding to each candidate recommendation sequence, the method of this embodiment further includes: determining a hit service category from a preset plurality of service categories, where there is a preset association relationship between the hit service category and the user to be recommended; determining a hit candidate recommendation sequence from each candidate recommendation sequence, where the hit candidate recommendation sequence contains candidate recommendation content belonging to the hit service category; performing a correction process on the evaluation score corresponding to the hit candidate recommendation sequence to obtain a corrected evaluation score for the hit candidate recommendation sequence; where the corrected evaluation score is higher than the evaluation score before correction; determining a target recommendation sequence from a plurality of candidate recommendation sequences according to the evaluation results corresponding to each candidate recommendation sequence, including: determining a target recommendation sequence from a plurality of candidate recommendation sequences according to the magnitude relationship between the evaluation scores corresponding to each candidate recommendation sequence.

[0088] Exemplarily, after obtaining the evaluation results corresponding to each candidate recommendation sequence, the preset association relationship of the user to be recommended can be obtained through an inquiry interface. The preset association relationship includes a specific association relationship between the user to be recommended and at least one service category, and the preset association relationship can indicate that the user to be recommended is an important user of this service category. According to the preset association relationship of the user to be recommended, a hit service category having this preset association relationship with the user to be recommended is determined from a preset plurality of service categories. Then, a hit candidate recommendation sequence containing candidate recommendation content belonging to the hit service category is determined from each candidate recommendation sequence; and a boosting process is performed on the evaluation score corresponding to the hit candidate recommendation sequence to obtain a corrected evaluation score for the hit candidate recommendation sequence after the correction process. For example, a certain score value can be added to the basis of the evaluation score corresponding to the hit candidate recommendation sequence, or if there is only one hit candidate recommendation sequence, this hit candidate recommendation sequence can be used as the target recommendation sequence, and so on.

[0089] In this embodiment, after obtaining the evaluation results corresponding to each candidate recommendation sequence, further determining a hit service category having a preset association relationship with the user to be recommended, and performing an evaluation score boosting process on the hit candidate recommendation sequence containing candidate recommendation content belonging to the hit service category can improve the recommendation probability of the service content corresponding to the service category having a preset association relationship with the user, thereby avoiding the loss of important users of each service category.

[0090] In summary, in the content recommendation solution provided in the embodiments of the present application, first, a plurality of candidate recommended contents are obtained, and the plurality of candidate recommended contents are arranged and combined to obtain a plurality of candidate recommended sequences; then, each candidate recommended sequence is evaluated as a whole to obtain an evaluation result corresponding to each candidate recommended sequence; finally, according to the evaluation results corresponding to each candidate recommended sequence, a target recommended sequence is determined from the plurality of candidate recommended sequences. The content recommendation method of this embodiment does not perform a single-point evaluation on a single candidate recommended content, but adopts a method of mixing and arranging the candidate recommended contents to generate candidate recommended sequences that simultaneously include a plurality of candidate recommended contents. Furthermore, the candidate recommended sequences are used as a whole evaluation sample to evaluate the pros and cons of the sequences as a whole. Then, according to the evaluation results at the sequence level, the sequence as a whole is recommended to the user. The solution provided in the embodiments of the present application can select a target recommended sequence with a better overall evaluation result for content recommendation. Since the target recommended sequence is obtained by evaluating the candidate recommended sequences as a whole, the evaluation process considers the influence of the relative arrangement of each candidate recommended content within the recommended sequence on the overall value of the sequence, and the evaluation dimension is more comprehensive. Therefore, the evaluation result can be made more accurate, and furthermore, the accuracy of the recommendation result can be improved, and the user experience can be enhanced. Further, in the candidate recommended sequence, arranging the position of the candidate interest content before the candidate service content can make the content sorting of each candidate recommended sequence start from the user's interests, which can enhance the user experience, increase the user retention rate, and further increase the traffic obtained by the subsequent service content, ensuring the overall recommendation efficiency.

[0091] Figure 4 FIG. is a flowchart of steps of a content recommendation method according to another exemplary embodiment of the present application. According to a second aspect in the embodiments of the present application, a content recommendation method is provided, which is applied to a user terminal. Referring to Figure 4 as shown, the method includes steps S402, S404, and S406, specifically:

[0092] S402. In response to a first operation of the user on the user terminal, send a content recommendation request to the server.

[0093] Exemplarily, the first operation is used to send a content recommendation request to the server. For example, the first operation may be an operation for the user to enter a network platform that can recommend content to the user, or it may also be an operation for the user to obtain new recommended content during the process of browsing the recommended content, and so on. The content recommendation request sent to the server may include the user's attribute information, etc., and this embodiment does not limit this.

[0094] S404. Receive the target recommendation sequence sent by the server based on the content recommendation request. The target recommendation sequence is determined by the server through evaluating multiple candidate recommendation sequences obtained by permuting and combining multiple candidate recommendation contents, and selecting from the multiple candidate recommendation sequences according to the evaluation results.

[0095] Exemplarily, after sending the content recommendation request to the server, the server can obtain multiple candidate recommendation contents based on the content recommendation request, and then perform permutation and combination on the multiple candidate recommendation contents to obtain multiple candidate recommendation sequences. Among them, the multiple candidate recommendation contents include candidate interest contents and candidate service contents. In the candidate recommendation sequence, the range of the first permutation position where the candidate interest content is located is before the range of the second permutation position where the candidate service content is located. Then, evaluate each candidate recommendation sequence respectively to obtain the evaluation results corresponding to each candidate recommendation sequence. Finally, according to the evaluation results corresponding to each candidate recommendation sequence, determine the target recommendation sequence from the multiple candidate recommendation sequences and send the target recommendation sequence to the user terminal.

[0096] S406. Display the content on the user terminal according to the target recommendation sequence.

[0097] Exemplarily, after receiving the target recommendation sequence, the user terminal can display the target recommendation contents on the interface of the user terminal one by one according to the target position sorting of each target recommendation content in the target recommendation sequence.

[0098] In some alternative embodiments, during the process of displaying the content on the user terminal according to the target recommendation sequence, the method of this embodiment further includes: in response to a second operation of the user on the user terminal, obtain the acceptance information of the user for each target recommendation content in the target recommendation sequence. The acceptance information is used to represent the degree of acceptance and preference of the user for the target recommendation content. Send the acceptance information to the server so that the server can update the acceptance evaluation results of the candidate recommendation contents belonging to the same service category as the target recommendation content according to the acceptance information.

[0099] Exemplarily, the second operation may be an operation for feedback on the user's preference for the target recommended content. For example, the second operation may be the user's like operation, report operation, block operation, not recommend operation, etc. for the target recommended content. Such operations can directly reflect the acceptance information of the user for the target recommended content. Additionally, the second operation may also be the user's browsing operation, such as the close operation, swipe-away operation, etc. Such operations can indirectly reflect the acceptance information of the user for the target recommended content by calculating the browsing duration. In response to the second operation of the user on the user terminal, obtain the acceptance information of the user for each target recommended content in the target recommendation sequence. The acceptance information can feedback the degree of acceptance of the user for the target recommended content. Then, the user terminal may send the acceptance information of the user for each target recommended content in the target recommendation sequence to the server. The server may update the acceptance evaluation result of the user for the interest category or service category to which the target recommended content belongs according to the received acceptance information of each target recommended content. Furthermore, according to the updated acceptance evaluation result of this service category, determine the next target recommendation sequence for this user.

[0100] In this embodiment, through the second operation of the user on the user terminal, the acceptance information of the user for each target recommended content in the target recommendation sequence can be obtained in a timely manner, and then the acceptance information of the user for each target recommended content in the target recommendation sequence is sent to the server, so that the evaluation result of the server for the subsequent candidate recommendation sequence can be adjusted in a timely manner, so that the next target recommendation sequence recommended to the user can be closer to the user's preferences and improve the user experience.

[0101] In some alternative embodiments, before obtaining the acceptance information of the user for each target recommended content in the target recommendation sequence in response to the second operation of the user on the user terminal, the method of this embodiment further includes: displaying preference options for each target recommended content; detecting the user's selection operation for the preference options as the second operation.

[0102] Exemplarily, when displaying content on the user terminal according to the target recommendation sequence, preference options may be displayed for each target recommended content on the display interface. The preference options may include options such as like, average, dislike, etc. When detecting the user's selection operation for the preference options, such as the click operation for the "like" option, determine this operation as the second operation.

[0103] In this embodiment, by providing preference options to the user during the process of presenting the target recommended content to the user, the preference situation of the user for the currently presented target recommended content can be collected in a timely manner, so that the acceptance information of the user for the currently presented target recommended content can be fed back to the server in a timely manner to improve the accuracy of the next target recommendation sequence recommended by the server to the user.

[0104] In some alternative embodiments, after content is displayed on the user terminal according to the target recommendation sequence, the method of this embodiment further includes: if the currently displayed target recommendation content corresponds to the target position in the target recommendation sequence, sending a next content recommendation request to the server, so that the server determines the next target recommendation sequence based on the next content recommendation request.

[0105] Exemplarily, according to the sequence length of the target recommendation sequence, the target position in the target recommendation sequence can be preset as the trigger position for sending the next content recommendation request to the server. For example, when the sequence length of the target recommendation sequence is 6, the 4th position in the target recommendation sequence can be set as the target position. When it is monitored that the currently displayed target recommendation content corresponds to the target position in the target recommendation sequence, a next content recommendation request is sent to the server. Here, the next content recommendation request can be directly sent to the server, or, after it is monitored that the currently displayed target recommendation content corresponds to the target position in the target recommendation sequence, in response to an operation of the user to obtain new recommended content, such as a swipe operation, a refresh operation, etc., a next content recommendation request is sent to the server, and this embodiment does not limit this. After receiving the next content recommendation request, the server determines and sends the next target recommendation sequence for the user.

[0106] In this embodiment, when the currently displayed target recommendation content corresponds to the target position in the target recommendation sequence, sending a next content recommendation request to the server, so that the server determines the next target recommendation sequence based on the next content recommendation request, can ensure that after the user browses the content of the current target recommendation sequence, the content of the next target recommendation sequence can be seamlessly connected. In addition, the recommendation evaluation of the next target recommendation sequence can also be adjusted based on the obtained acceptance information of the user for the content of the current target recommendation sequence, so as to improve the accuracy of the next target recommendation sequence.

[0107] In summary, the content recommendation solution provided in the embodiments of the present application is applied to a user terminal. First, in response to a first operation of the user on the user terminal, a content recommendation request is sent to the server; then, the server is received to evaluate multiple candidate recommendation sequences respectively based on the content recommendation request, and a target recommendation sequence is determined from the multiple candidate recommendation sequences according to the evaluation results; finally, content is displayed on the user terminal according to the target recommendation sequence. The content recommendation method of this embodiment does not perform a single-point evaluation of a single candidate recommended content, but adopts a way of mixing and arranging the candidate recommended contents to generate a candidate recommendation sequence that simultaneously includes multiple candidate recommended contents. Then, the candidate recommendation sequence as a whole is used as an evaluation sample to evaluate the pros and cons of the sequence as a whole. After that, according to the evaluation results at the sequence level, the sequence as a whole is recommended to the user. The solution provided in the embodiments of the present application can select a target recommendation sequence with a better overall evaluation result for content recommendation. Since the target recommendation sequence is obtained by overall evaluation of the candidate recommendation sequences, the evaluation process takes into account the influence of the relative arrangement of each candidate recommended content within the recommendation sequence on the overall value of the sequence, and the evaluation dimension is more comprehensive. Therefore, the evaluation result can be more accurate, and thus the accuracy of the recommendation result can also be improved, enhancing the user experience. Further, in the candidate recommendation sequence, arranging the position of the candidate interest content before the candidate service content can make the content sorting of each candidate recommendation sequence start from the user's interest, which can enhance the user experience, increase the user retention rate, and thus increase the traffic obtained by subsequent service content, ensuring the overall recommendation efficiency.

[0108] Next, with reference to Figure 5 the following scenario schematic diagram, an exemplary description of an overall implementation process of the content recommendation solution of the embodiments of the present application is given. This overall implementation process can be understood by substituting it into the scenario of video recommendation. As Figure 5As shown, the server can provide candidate recommended content for interest categories and service categories. Among them, service categories include live streaming, short drama, story, and e-commerce. First, the user terminal sends a content recommendation request to the server in response to the user's first operation on the user terminal. Then, based on the content recommendation request, the server obtains multiple candidate recommended content from the content of interest categories and service categories. Among them, the multiple candidate recommended content includes candidate interest content and candidate service content. Then, within the first arrangement position range, the candidate interest content is arranged and combined, and within the second arrangement position range, the candidate service content is arranged and combined to obtain multiple candidate recommended sequences; among them, in the candidate recommended sequence, the first arrangement position range is before the second arrangement position range. After that, each obtained candidate recommended sequence is evaluated to obtain the evaluation result corresponding to each candidate recommended sequence; finally, according to the evaluation results corresponding to each candidate recommended sequence, the target recommended sequence is determined from the multiple candidate recommended sequences and sent to the user terminal. Finally, the user terminal displays the target recommended content on the interface of the user terminal one by one according to the target position sorting of each target recommended content in the target recommended sequence.

[0109] It should be understood that Figure 5 more details of the overall implementation process shown can also be understood in combination with the previous embodiments, and Figure 5 the overall implementation process shown is only some examples for easy understanding of the embodiments of the present application and does not impose any limitation on the embodiments of the present application.

[0110] It can be understood that the foregoing description of the content recommendation method is only some exemplary descriptions of the embodiments of the present application and does not impose any limitation on the embodiments of the present application.

[0111] Referring to Figure 6 , a structural block diagram of a content recommendation device according to an exemplary embodiment of the present application is shown.

[0112] The content recommendation device in this embodiment includes an acquisition module 602, a sorting module 604, an evaluation module 606, and a selection module 608.

[0113] Among them, the obtaining module 602 is used to obtain a plurality of candidate recommended contents; the plurality of candidate recommended contents include candidate interest contents and candidate service contents; the sorting module 604 is used to perform permutations and combinations on the candidate interest contents within the first permutation position range and perform permutations and combinations on the candidate service contents within the second permutation position range to obtain a plurality of candidate recommended sequences; wherein, in the candidate recommended sequences, the first permutation position range is before the second permutation position range; the evaluation module 606 is used to evaluate each candidate recommended sequence respectively to obtain the evaluation result corresponding to each candidate recommended sequence; the evaluation result corresponding to the candidate recommended sequence represents the sequence-level evaluation result of the overall candidate recommended sequence; the selection module 608 is used to determine the target recommended sequence from the plurality of candidate recommended sequences according to the evaluation results corresponding to each candidate recommended sequence, so as to perform content recommendation based on the target recommended sequence.

[0114] In some alternative embodiments, the obtaining module 602 is further used to: select target interest contents from the candidate interest contents and select target service contents from the candidate service contents according to the attribute information of the user to be recommended; the sorting module 604 is further used to: perform permutations and combinations on the target interest contents within the first permutation position range and perform permutations and combinations on the target service contents within the second permutation position range to obtain a plurality of candidate recommended sequences; wherein, in the candidate recommended sequences, the positions of the target interest contents are before the target service contents.

[0115] In some alternative embodiments, the obtaining module 602 is further used to: respectively determine the candidate service contents corresponding to each preset service category; determine a plurality of target service categories from the plurality of service categories and determine target service contents from the candidate service contents corresponding to the target service categories; wherein, each of the target service contents included in the same candidate recommended sequence corresponds to a different service category.

[0116] In some alternative embodiments, the evaluation module 606 is further used to: for a single candidate recommended sequence among the plurality of candidate recommended sequences, use a multi-objective learning model to evaluate and obtain the basic evaluation results corresponding to the candidate recommended contents in the single candidate recommended sequence; obtain the position evaluation results of the candidate recommended contents in the single candidate recommended sequence, and the position evaluation results represent the exposure information of the positions where the candidate recommended contents are located in the single candidate recommended sequence; fuse the basic evaluation results and the position evaluation results to obtain the comprehensive evaluation results corresponding to the candidate recommended contents; fuse the comprehensive evaluation results corresponding to the candidate recommended contents to obtain the evaluation result corresponding to the single candidate recommended sequence.

[0117] In some alternative embodiments, the evaluation module 606 is further configured to: before fusing the basic evaluation result and the location evaluation result, determine the acceptance evaluation result of each candidate recommendation content, where the acceptance evaluation result represents the degree of acceptance and preference of the user to be recommended for the candidate recommendation content; fuse the basic evaluation result, the location evaluation result, and the acceptance evaluation result to obtain the comprehensive evaluation result corresponding to each candidate recommendation content.

[0118] In some alternative embodiments, the evaluation result is presented in the form of an evaluation score; the selection module 608 is further configured to: after obtaining the evaluation results corresponding to each candidate recommendation sequence, determine the hit service category from a preset plurality of service categories, where there is a preset association relationship between the hit service category and the user to be recommended; determine the hit candidate recommendation sequence from each candidate recommendation sequence, where the hit candidate recommendation sequence includes candidate recommendation content belonging to the hit service category; perform a correction process on the evaluation score corresponding to the hit candidate recommendation sequence to obtain the corrected evaluation score of the hit candidate recommendation sequence; wherein, the corrected evaluation score is higher than the evaluation score before correction; and determine the target recommendation sequence from the plurality of candidate recommendation sequences according to the magnitude relationship between the evaluation scores corresponding to each candidate recommendation sequence.

[0119] In some alternative embodiments, the apparatus of this embodiment further includes a screening module, configured to: before evaluating each candidate recommendation sequence respectively to obtain the evaluation result corresponding to each candidate recommendation sequence, perform a sequence screening process on the plurality of candidate recommendation sequences based on a sequence screening strategy to obtain a plurality of screened sequences; the evaluation module 606 is further configured to: evaluate each of the screened sequences respectively to obtain the evaluation result corresponding to each screened sequence; the selection module 608 is further configured to: determine the target recommendation sequence from the plurality of screened sequences according to the evaluation results corresponding to each screened sequence.

[0120] In some alternative embodiments, the screening module is further configured to: determine the overage service category from the plurality of service categories, where the overage service category is a service category for which the recommended data reaches a preset recommendation threshold; determine the overage candidate recommendation sequence from each candidate recommendation sequence, where the overage candidate recommendation sequence includes candidate recommendation content belonging to the overage service category; delete the overage candidate recommendation sequences in the plurality of candidate recommendation sequences to obtain a plurality of screened sequences.

[0121] In some alternative embodiments, the screening module is further configured to: determine pairs of similar service categories from the plurality of service categories, where each pair of similar service categories includes a plurality of service categories with similar service contents; determine the similar sequences from each candidate recommendation sequence, where each similar sequence includes a plurality of candidate recommendation contents belonging to the same pair of similar service categories; delete the similar sequences in the plurality of candidate recommendation sequences to obtain a plurality of screened sequences.

[0122] In some alternative embodiments, the evaluation module 606 is further configured to: input the attribute information of the user to be recommended and each candidate recommendation content in each candidate recommendation sequence into a pre-trained multi-objective learning model, and output the multi-objective evaluation results corresponding to each candidate recommendation content; perform normalization processing on the multi-objective evaluation results corresponding to each candidate recommendation content based on the sorting corresponding to the multi-objective evaluation results of each candidate recommendation content, to obtain the processed multi-objective evaluation results; perform linear interpolation processing on the processed multi-objective evaluation results corresponding to each candidate recommendation content in a single candidate recommendation sequence, to obtain the basic evaluation results corresponding to each candidate recommendation content in the single candidate recommendation sequence.

[0123] In some alternative embodiments, the multi-objective learning model includes a first expert network, a first gating network, a second expert network, a second gating network, and a tower network; the evaluation module 606 is further configured to: for each candidate recommendation content in each candidate recommendation sequence, input the attribute information of the user to be recommended and a single candidate recommendation content into the first shared expert network and the first target expert networks corresponding to each target in the first expert network for feature extraction, and output the first features and the first shared feature corresponding to each target; input the attribute information of the user to be recommended and a single candidate recommendation content, as well as the first features and the first shared feature corresponding to each target, into the first gating networks corresponding to each target for feature fusion, and output the intermediate features corresponding to each target; input the intermediate features corresponding to each target into the second shared expert network and the second target expert networks corresponding to each target in the second expert network for feature extraction, and output the second features and the second shared feature corresponding to each target; input the intermediate features corresponding to each target, as well as the second features and the second shared feature corresponding to each target, into the second gating networks corresponding to each target for feature fusion, and output the target features corresponding to each target; input the target features corresponding to each target into the tower networks corresponding to each target, and output the multi-objective evaluation results corresponding to the candidate recommendation content.

[0124] In some alternative embodiments, there are multiple candidate interest contents, and the acquisition module 602 is further configured to: input the attribute information of the user to be recommended and each candidate interest content into the multi-objective learning model for evaluation, to obtain the initial evaluation results corresponding to each candidate interest content; sort each candidate interest content based on the initial evaluation results corresponding to each candidate interest content, to obtain the initial sorting results corresponding to each candidate interest content; select the target interest content from each candidate interest content based on the initial sorting results corresponding to each candidate interest content.

[0125] In some alternative embodiments, the evaluation module 606 is further configured to: obtain acceptance information of the user to be recommended for the historical recommendation sequence; the acceptance information is used to characterize the degree of acceptance and preference of the user to be recommended for each target recommended content in the historical recommendation sequence; generate an acceptance evaluation result for the service category to which each target recommended content belongs according to the acceptance information; determine the acceptance evaluation result of the candidate recommended content according to the service category to which the candidate recommended content belongs.

[0126] The content recommendation device of this embodiment is used to implement the corresponding content recommendation methods in multiple method embodiments of the foregoing first aspect, and has the beneficial effects of the corresponding method embodiments, which will not be elaborated here. In addition, the functions of each module in the content recommendation device of this embodiment can be referred to the descriptions of the corresponding parts in the foregoing method embodiments, which will not be elaborated here either.

[0127] Refer to Figure 7 , which shows a structural block diagram of a content recommendation device according to an exemplary embodiment of the present application.

[0128] The content recommendation device of this embodiment is applied to a user terminal and includes a sending module 702, a receiving module 704, and a display module 706.

[0129] Among them, the sending module 702 is configured to send a content recommendation request to the server in response to a first operation of the user on the user terminal; the receiving module 704 is configured to receive a target recommendation sequence sent by the server based on the content recommendation request, and the target recommendation sequence is determined by the server through evaluating multiple candidate recommendation sequences obtained by permuting and combining multiple candidate recommended contents, and according to the evaluation results; among the multiple candidate recommended contents, candidate interest contents and candidate service contents are included; in the candidate recommendation sequence, the first arrangement position range where the candidate interest content is located is before the second arrangement position range where the candidate service content is located; the display module 706 is configured to perform content display on the user terminal according to the target recommendation sequence.

[0130] In some alternative embodiments, during the process of performing content display on the user terminal according to the target recommendation sequence, the sending module 702 is further configured to: obtain acceptance information of the user for each target recommended content in the target recommendation sequence in response to a second operation of the user on the user terminal; the acceptance information is used to characterize the degree of acceptance and preference of the user for the target recommended content; send the acceptance information to the server so that the server can update the acceptance evaluation result of the candidate recommended content belonging to the same service category as the target recommended content according to the acceptance information.

[0131] In some alternative embodiments, before obtaining the acceptance information of the user for each target recommendation content in the target recommendation sequence in response to a second operation of the user on the user terminal, the display module 706 is further configured to: display preference options for each target recommendation content; detect a selection operation of the user for the preference options as the second operation.

[0132] In some alternative embodiments, after content is displayed on the user terminal according to the target recommendation sequence, the sending module 702 is further configured to: if the currently displayed target recommendation content corresponds to a target position in the target recommendation sequence, send a next content recommendation request to the server, so that the server determines a next target recommendation sequence based on the next content recommendation request.

[0133] The content recommendation device of this embodiment is used to implement the corresponding content recommendation methods in multiple method embodiments of the foregoing second aspect, and has the beneficial effects of the corresponding method embodiments, which will not be elaborated here. In addition, the functions of each module in the content recommendation device of this embodiment can be referred to the corresponding parts in the foregoing method embodiments, which will not be elaborated here either.

[0134] According to a fifth aspect of the embodiments of the present application, there is provided an electronic device, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory complete communication with each other through the communication bus; the memory is used to store a computer program; the processor is configured to execute the content recommendation method described in the foregoing first aspect by running the computer program stored on the memory.

[0135] Figure 8 The structural block diagram of an alternative electronic device in the embodiments of the present application is shown. The embodiments of the present application do not limit the specific implementation of the electronic device 800. Exemplarily, referring to Figure 8 , the electronic device 800 provided by the embodiments of the present application includes: a processor 802, a communication interface 804, a memory 806, and a communication bus 808. Among them:

[0136] The processor 802, the communication interface 804, and the memory 806 complete communication with each other through the communication bus 808.

[0137] The communication interface 804 is used to communicate with other electronic devices or servers.

[0138] The processor 802 is configured to execute the computer program 810, and specifically may execute relevant steps in any of the foregoing content recommendation method embodiments.

[0139] Specifically, the computer program 810 may include program code, which includes computer operation instructions.

[0140] The processor 802 may be a CPU, or a GPU (Graphic Processing Unit), or an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. One or more processors included in the intelligent device may be of the same type of processor, such as one or more CPUs; or may be of different types of processors, such as one or more CPUs and one or more ASICs.

[0141] The memory 806 is used to store the computer program 810. The memory 806 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk memory.

[0142] The computer program 810 may specifically be used to cause the processor 802 to execute the content recommendation method in any of the foregoing embodiments.

[0143] For the specific implementation of each step in the computer program 810, reference may be made to the corresponding steps and units in any of the foregoing content recommendation method embodiments, which will not be elaborated herein. Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described devices and modules may refer to the corresponding process descriptions in the foregoing method embodiments, which will not be elaborated herein.

[0144] The electronic device 800 in the embodiments of the present application has been described in detail in the foregoing content recommendation method embodiments. Therefore, the relevant content and beneficial effects thereof may be understood with reference to the foregoing method embodiments, and will not be elaborated herein.

[0145] According to the sixth aspect of the embodiments of the present application, the embodiments of the present application further provide a computer storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the content recommendation method described in any of the foregoing method embodiments. The computer storage medium includes, but is not limited to: Compact Disc Read-Only Memory (CD-ROM), Random Access Memory (RAM), floppy disk, hard disk, or magneto-optical disk, etc.

[0146] According to the seventh aspect of the embodiments of the present application, the embodiments of the present application further provide a computer program product, including a computer program, and the computer program, when executed by a processor, implements the content recommendation method described in any one of the above-mentioned multiple method embodiments.

[0147] The embodiments of the electronic device 800 / computer storage medium / computer program product in the embodiments of the present application have been described in detail in the foregoing content recommendation method embodiments. Therefore, the relevant content and beneficial effects can be understood with reference to the above-mentioned method embodiments and will not be elaborated herein.

[0148] In addition, it should be noted that the information related to users (including but not limited to user device information, user personal information, etc.) and data (including but not limited to sample data for training the model, data for analysis, stored data, displayed data, etc.) involved in the embodiments of the present application are all information and data authorized by the users or fully authorized by all parties. And the collection, use, and processing of the relevant data need to comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or refuse.

[0149] It should be pointed out that according to the needs of implementation, each component / step described in the embodiments of the present application can be split into more components / steps, or two or more components / steps or partial operations of the components / steps can be combined into new components / steps to achieve the purpose of the embodiments of the present application.

[0150] The method according to the embodiments of the present application can be implemented in hardware, firmware, or be implemented as software or computer code that can be stored in a recording medium (such as a CD-ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or be implemented as computer code originally stored in a remote recording medium or a non-transitory machine-readable medium and to be downloaded through a network and stored in a local recording medium, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an Application Specific Integrated Circuit (ASIC) or a Field Programmable Gate Array (FPGA)). It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component (such as a Random Access Memory (RAM), a Read-Only Memory (ROM), a flash memory, etc.) that can store or receive software or computer code, and when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the method shown herein, the execution of the code converts the general-purpose computer into a dedicated computer for executing the method shown herein.

[0151] Those of ordinary skill in the art can realize that the units and method steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for a specific application, but such implementation should not be considered to exceed the scope of the embodiments of the present application.

[0152] The term "including" and its variations used herein are open-ended, that is, "including but not limited to". The term "based on" is "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". It should be noted that the concepts such as "first" and "second" mentioned in the embodiments of the present application are only used to distinguish different devices, modules, or units, and are not used to limit the order or interdependence relationship of the functions executed by these devices, modules, or units. It should be noted that the modifications of "one" and "multiple" mentioned in the embodiments of the present application are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise clearly stated in the context, it should be understood as "one or more".

[0153] The above embodiments are only used to illustrate the embodiments of the present application, rather than to limit the embodiments of the present application. Those of ordinary skill in the relevant technical field can also make various changes and modifications without departing from the spirit and scope of the embodiments of the present application. Therefore, all equivalent technical solutions also belong to the scope of the embodiments of the present application. The patent protection scope of the embodiments of the present application shall be defined by the claims.

Claims

1. A content recommendation method, comprising: Obtaining a plurality of candidate recommended contents, where the plurality of candidate recommended contents include candidate interest contents and candidate service contents; Performing permutations and combinations on the candidate interest contents within a first arrangement position range, and performing permutations and combinations on the candidate service contents within a second arrangement position range to obtain a plurality of candidate recommendation sequences; wherein, in the candidate recommendation sequences, the first arrangement position range is before the second arrangement position range; Evaluating each candidate recommendation sequence respectively to obtain an evaluation result corresponding to each candidate recommendation sequence; the evaluation result corresponding to a candidate recommendation sequence represents the evaluation result at the sequence level of the entire candidate recommendation sequence; Determining a target recommendation sequence from the plurality of candidate recommendation sequences according to the evaluation results corresponding to each candidate recommendation sequence, so as to perform content recommendation based on the target recommendation sequence.

2. The method according to claim 1, wherein, Performing permutations and combinations on the candidate interest contents within a first arrangement position range, and performing permutations and combinations on the candidate service contents within a second arrangement position range to obtain a plurality of candidate recommendation sequences, including: Selecting target interest contents from the candidate interest contents and selecting target service contents from the candidate service contents according to the attribute information of the user to be recommended; Performing permutations and combinations on the target interest contents within a first arrangement position range, and performing permutations and combinations on the target service contents within a second arrangement position range to obtain a plurality of candidate recommendation sequences; Wherein, in the candidate recommendation sequences, the positions of the target interest contents are before the positions of the target service contents.

3. The method according to claim 2, wherein The selecting target service contents from the candidate service contents includes: For a plurality of preset service categories, respectively determining the candidate service contents corresponding to each service category; Determining at least one target service category from the plurality of service categories, and determining target service contents from the candidate service contents corresponding to the target service category; Wherein, each of the target service contents included in the same candidate recommendation sequence corresponds to a different service category.

4. The method according to claim 3, wherein, The evaluating each candidate recommendation sequence respectively to obtain an evaluation result corresponding to each candidate recommendation sequence includes: For a single candidate recommendation sequence among the plurality of candidate recommendation sequences, using a multi-objective learning model to evaluate to obtain a basic evaluation result corresponding to each candidate recommended content in the single candidate recommendation sequence; Obtaining a position evaluation result of each candidate recommended content in the single candidate recommendation sequence, where the position evaluation result represents the exposure information of the position where each candidate recommended content is located in the single candidate recommendation sequence; Fusing the basic evaluation result and the position evaluation result to obtain a comprehensive evaluation result corresponding to each candidate recommended content; Fusing the comprehensive evaluation results corresponding to each candidate recommended content to obtain the evaluation result corresponding to the single candidate recommendation sequence.

5. The method according to claim 4, wherein Before the fusing the basic evaluation result and the position evaluation result, the method further includes: Determining an acceptance evaluation result of each candidate recommended content, where the acceptance evaluation result represents the degree of acceptance and preference of the user to be recommended for the candidate recommended content; Fusing the basic evaluation result and the location evaluation result to obtain a comprehensive evaluation result corresponding to each candidate recommendation content, including: Fusing the basic evaluation result, the location evaluation result, and the acceptance evaluation result to obtain a comprehensive evaluation result corresponding to each candidate recommendation content.

6. The method according to any one of claims 3-5, wherein, The evaluation result is presented in the form of an evaluation score; After obtaining the evaluation results corresponding to each candidate recommendation sequence, the method further includes: Determining a hit service category from the preset multiple service categories, where there is a preset association relationship between the hit service category and the user to be recommended; Determining a hit candidate recommendation sequence from each candidate recommendation sequence, where the hit candidate recommendation sequence contains candidate recommendation content belonging to the hit service category; Performing a correction process on the evaluation score corresponding to the hit candidate recommendation sequence to obtain a corrected evaluation score for the hit candidate recommendation sequence; wherein, the corrected evaluation score is higher than the evaluation score before correction; Determining a target recommendation sequence from the multiple candidate recommendation sequences according to the evaluation results corresponding to each candidate recommendation sequence, including: Determining a target recommendation sequence from the multiple candidate recommendation sequences according to the magnitude relationship between the evaluation scores corresponding to each candidate recommendation sequence.

7. The method according to any one of claims 1-5, wherein Before respectively evaluating each candidate recommendation sequence to obtain the evaluation results corresponding to each candidate recommendation sequence, the method further includes: Performing a sequence screening process on the multiple candidate recommendation sequences based on a sequence screening strategy to obtain multiple screened sequences; Respectively evaluating each candidate recommendation sequence to obtain the evaluation results corresponding to each candidate recommendation sequence, including: Respectively evaluating each screened sequence to obtain the evaluation results corresponding to each screened sequence; Determining a target recommendation sequence from the multiple candidate recommendation sequences according to the evaluation results corresponding to each candidate recommendation sequence, including: Determining a target recommendation sequence from the multiple screened sequences according to the evaluation results corresponding to each screened sequence.

8. The method according to claim 7, wherein Performing a sequence screening on the multiple candidate recommendation sequences based on a sequence screening strategy to obtain multiple screened sequences, including: Determining an overage service category from multiple service categories, where the overage service category is a service category for which the recommended data reaches a preset recommendation threshold; Determining an overage candidate recommendation sequence from each candidate recommendation sequence, where the overage candidate recommendation sequence contains candidate recommendation content belonging to the overage service category; Deleting the overage candidate recommendation sequences in the multiple candidate recommendation sequences to obtain multiple screened sequences.

9. The method according to claim 7, wherein Performing a sequence screening on the multiple candidate recommendation sequences based on a sequence screening strategy to obtain multiple screened sequences, including: Determining pairs of similar service categories from multiple service categories, where each pair of similar service categories contains multiple service categories with similar service contents; Determining similar sequences from each candidate recommendation sequence, where each similar sequence contains multiple candidate recommendation content belonging to the same pair of similar service categories; Deleting the similar sequences in the multiple candidate recommendation sequences to obtain multiple screened sequences.

10. The method according to claim 4, wherein, The basic evaluation results corresponding to each candidate recommendation content in the single candidate recommendation sequence obtained by using the multi-objective learning model evaluation include: Input the attribute information of the user to be recommended and each candidate recommendation content in each candidate recommendation sequence into a pre-trained multi-objective learning model, and output the multi-objective evaluation results corresponding to each candidate recommendation content; Based on the sorting corresponding to the multi-objective evaluation results of each candidate recommendation content, perform normalization processing on the multi-objective evaluation results corresponding to each candidate recommendation content to obtain the processed multi-objective evaluation results; Perform linear interpolation processing on the processed multi-objective evaluation results corresponding to each candidate recommendation content in the single candidate recommendation sequence to obtain the basic evaluation results corresponding to each candidate recommendation content in the single candidate recommendation sequence.

11. The method according to claim 10, wherein, The multi-objective learning model includes a first expert network, a first gating network, a second expert network, a second gating network, and a tower network; the step of inputting the attribute information of the user to be recommended and each candidate recommendation content in each candidate recommendation sequence into a pre-trained multi-objective learning model and outputting the multi-objective evaluation results corresponding to each candidate recommendation content includes: For each candidate recommendation content in each candidate recommendation sequence, input the attribute information of the user to be recommended and a single candidate recommendation content into the first shared expert network and the first target expert network corresponding to each target in the first expert network for feature extraction, and output the first feature and the first shared feature corresponding to each target; Input the attribute information of the user to be recommended, a single candidate recommendation content, and the first features and the first shared features corresponding to each target into the first gating network corresponding to each target for feature fusion, and output the intermediate features corresponding to each target; Input the intermediate features corresponding to each target into the second shared expert network and the second target expert network corresponding to each target in the second expert network for feature extraction, and output the second features and the second shared features corresponding to each target; Input the intermediate features corresponding to each target, and the second features and the second shared features corresponding to each target into the second gating network corresponding to each target for feature fusion, and output the target features corresponding to each target; Input the target features corresponding to each target into the tower network corresponding to each target, and output the multi-objective evaluation results corresponding to the candidate recommendation content.

12. The method according to claim 2, wherein, There are multiple candidate interest contents, and the step of selecting target interest contents from the candidate interest contents includes: Input the attribute information of the user to be recommended and each candidate interest content into a multi-objective learning model for evaluation, and obtain the initial evaluation results corresponding to each candidate interest content; Based on the initial evaluation results corresponding to each candidate interest content, sort each candidate interest content to obtain the initial sorting results corresponding to each candidate interest content; Based on the initial sorting results corresponding to each candidate interest content, select target interest contents from each candidate interest content.

13. The method according to claim 5, wherein, Determine the acceptance evaluation results of each candidate recommendation content, including: Obtain the acceptance information of the to-be-recommended user for the historical recommendation sequence; the acceptance information is used to characterize the degree of acceptance and preference of the to-be-recommended user for each target recommendation content in the historical recommendation sequence; Generate an acceptance evaluation result for the service category to which each target recommendation content belongs according to the acceptance information; Determine the acceptance evaluation result of the candidate recommendation content according to the service category to which the candidate recommendation content belongs.

14. A content recommendation method, applied to a user terminal, includes: In response to a first operation of the user on the user terminal, send a content recommendation request to the server; Receive a target recommendation sequence sent by the server based on the content recommendation request, where the target recommendation sequence is obtained by the server evaluating multiple candidate recommendation sequences obtained by permuting and combining a plurality of candidate recommendation contents respectively, and determining from the multiple candidate recommendation sequences according to the evaluation results; wherein, the plurality of candidate recommendation contents include candidate interest contents and candidate service contents; in the candidate recommendation sequence, the first arrangement position range where the candidate interest content is located is before the second arrangement position range where the candidate service content is located; Perform content display on the user terminal according to the target recommendation sequence.

15. The method according to claim 14, wherein, During the process of performing content display on the user terminal according to the target recommendation sequence, the method further includes: In response to a second operation of the user on the user terminal, obtain the acceptance information of the user for each target recommendation content in the target recommendation sequence; the acceptance information is used to characterize the degree of acceptance and preference of the user for the target recommendation content; Send the acceptance information to the server, so that the server updates the acceptance evaluation result of the candidate recommendation content belonging to the same service category as the target recommendation content according to the acceptance information.

16. The method according to claim 15, wherein, Before the step of, in response to a second operation of the user on the user terminal, obtaining the acceptance information of the user for each target recommendation content in the target recommendation sequence, the method further includes: Display preference options for each of the target recommendation contents; Detect a selection operation of the user for the preference options as the second operation.

17. The method according to claim 16, wherein, After performing content display on the user terminal according to the target recommendation sequence, the method further includes: If the currently displayed target recommendation content corresponds to a target position in the target recommendation sequence, send a next content recommendation request to the server, so that the server determines a next target recommendation sequence based on the next content recommendation request.

18. An electronic device, comprising: A processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory complete communication with each other through the communication bus; The memory is used to store a computer program; The processor is configured to execute the method according to any one of claims 1-13, or execute the method according to any one of claims 13-17 by running the computer program stored on the memory.

19. A computer storage medium having a computer program stored thereon, which when executed by a processor implements the method according to any one of claims 1-13, or implements the method according to any one of claims 13-17.

20. A computer program product comprising computer instructions that direct a computing device to perform operations corresponding to the method according to any one of claims 1-13, or to perform operations corresponding to the method according to any one of claims 13-17.

Images