Information recommendation method and device, computer device, and storage medium
By adjusting the order of target information to meet recommendation rules and dynamically adjusting weights, the problem of poor recommendation performance caused by improper weight adjustment in existing technologies is solved, and efficient and rapid layout adjustment and user experience improvement of recommendation systems are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- MICRO DREAM TECHTRONIC NETWORK TECH CHINACO
- Filing Date
- 2022-09-28
- Publication Date
- 2026-05-19
AI Technical Summary
When existing recommendation systems adjust the weights of target information during the re-ranking stage, they are prone to insufficient or excessive weighting, resulting in poor recommendation performance, affecting user experience, and making it difficult to effectively adjust the layout of the recommendation domain.
By acquiring the first ranking and recommendation rules of the target information, the second ranking of the target information is adjusted to meet the recommendation rules of the target recommendation domain. The minimum value of the numerical range is set to ensure the probability of effective recommendation. The ranking score is then dynamically adjusted based on the difference between the actual and expected recommendation ratios.
While avoiding large-scale changes to the original ranking order, it improves the overall performance of the recommendation system, ensures effective recommendations in various target recommendation domains, and enhances the ecosystem traffic control and user experience of the recommendation system.
Smart Images
Figure CN115659009B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of Internet technology, and in particular to an information recommendation method, apparatus, computer device, and storage medium. Background Technology
[0002] In today's era of big data on the internet, the phenomenon of "information overload" permeates all aspects of human life. Recommendation systems are currently a very important technical solution to the problem of "information overload," helping users filter out the Top-K parts of information that interest them from massive amounts of data. Essentially, it is an information filtering system. Current recommendation systems are usually multi-stage, generally consisting of recall, coarse ranking, fine ranking, and re-ranking. Re-ranking, as the last stage in the recommendation system process, has a direct impact on recommendation results and user experience.
[0003] In related technologies, target information is usually rearranged by adjusting the weight of the target information, forcibly inserting it into a fixed position, or scattering it.
[0004] However, the above-mentioned reordering methods all have certain static and limitations. For example, by adjusting the target information, the exposure opportunities can be increased or decreased by changing the ranking score of the target information. However, the weighting is usually set manually, which makes it very easy for the weighting to be insufficient or excessive. If the weighting is too small, it will not achieve the expected business effect. If the weighting is too large, it will change the original fine ranking recommendation order too much, affecting the recommendation effect and causing a decline in user experience. Summary of the Invention
[0005] This application provides an information recommendation method, apparatus, device, and storage medium that can guarantee the effective recommendation probability of each target recommendation domain. While avoiding large-scale changes to the original ranking order, it can efficiently and quickly adjust the layout of the recommendation domains, improving the overall recommendation effect of the recommendation system. The technical solution is as follows:
[0006] On the one hand, an information recommendation method is provided, the method comprising:
[0007] Obtain a first sorting of each target information; the first sorting refers to the recommended order of each target information after information fine sorting.
[0008] Obtain the recommendation rules for the target recommendation domain to which each target information belongs; the recommendation rules are used to indicate the numerical range of the number of target information that can be recommended in the corresponding unit recommendation window for the target recommendation domain; the numerical range includes the minimum value;
[0009] Based on the recommendation rules of each of the target recommendation domains, the first ranking is adjusted to determine the second ranking of each of the target information;
[0010] Based on the second ranking, recommendations are made for each of the target information.
[0011] On the other hand, an information recommendation device is provided, the device comprising:
[0012] The sorting acquisition module is used to acquire the first sorting of each target information; the first sorting refers to the recommended order of each target information after information fine sorting.
[0013] The rule acquisition module is used to acquire the recommendation rules of the target recommendation domain to which each target information belongs; the recommendation rules are used to indicate the numerical range of the number of target information that can be recommended in the corresponding unit recommendation window of the target recommendation domain; the numerical range includes the minimum value;
[0014] The sorting adjustment module is used to adjust the first sorting based on the recommendation rules of each of the target recommendation domains to determine the second sorting of each of the target information.
[0015] The information recommendation module is used to recommend each of the target information based on the second ranking.
[0016] In one possible implementation, the sorting adjustment module includes:
[0017] The percentage acquisition submodule is used to acquire the actual recommendation percentage of the target recommendation domain to which each of the target information belongs; the actual recommendation percentage is used to indicate the percentage of the number of recommended messages in the target recommendation domain within the target duration compared to the number of already recommended messages;
[0018] The first sorting adjustment submodule is used to sort the difference between the actual recommendation ratio of each target recommendation domain and the expected recommendation ratio corresponding to each target recommendation domain, adjust the first sorting, and determine the third sorting of each target information.
[0019] The second sorting adjustment submodule is used to adjust the third sorting based on the recommendation rules of each of the target recommendation domains, and to determine the second sorting of each of the target information.
[0020] In one possible implementation, the first sorting adjustment submodule includes:
[0021] The parameter acquisition unit is used to acquire the weighting parameters corresponding to each of the target recommendation domains, wherein the weighting parameters are the ratio between the proportion difference of the target recommendation domain and the expected recommendation proportion of the target recommendation domain.
[0022] The score adjustment unit is used to adjust the first ranking score of each target information in each target recommendation domain based on the weighting parameters corresponding to each target recommendation domain, and to determine the second ranking score of each target information in each target recommendation domain.
[0023] The position determination unit is used to determine the ranking position of each of the target information in the third ranking based on the second ranking score of each of the target information in each of the target recommendation domains.
[0024] In one possible implementation, the recommendation rules have an application priority.
[0025] The second sorting adjustment submodule is used to determine the sorting determined by the numerical range of the first recommendation rule as the second sorting if the sorting determined by the numerical range of the first recommendation rule conflicts with the sorting determined by the numerical range of the second recommendation rule.
[0026] In this case, the application priority of the first recommendation rule is higher than the application priority of the second recommendation rule.
[0027] In one possible implementation, the second sorting adjustment submodule includes:
[0028] The first adjustment unit is used to adjust the third ranking based on the recommendation rules of each of the recommendation domains to determine a temporary ranking;
[0029] The sorting determination unit is configured to determine the temporary sorting as the second sorting if the number of target information in each target recommendation domain in each unit recommendation window corresponding to the temporary sorting satisfies the corresponding recommendation rule.
[0030] The second adjustment unit is configured to, if the number of target information in the first target recommendation field corresponding to the first recommendation window in the temporary sorting does not meet the minimum value in the numerical range indicated by the recommendation rule corresponding to the first target recommendation field, add the target information of the other target recommendation fields in the first recommendation window according to the order indicated by the temporary sorting, provided that the number of target information in the other target recommendation fields in the first recommendation window meets the corresponding recommendation rule; the first recommendation window is any one of the various unit recommendation windows; the other target recommendation fields are target recommendation fields other than the first target recommendation field in the target recommendation fields corresponding to the first recommendation window;
[0031] The sorting determination unit is used to determine the adjusted temporary sorting as the second sorting.
[0032] In one possible implementation, the numerical range further includes a maximum value; the rule acquisition module is used to acquire the recommendation rules for each of the target recommendation domains based on the expected recommendation ratio corresponding to each of the target recommendation domains.
[0033] Wherein, the ratio of the maximum number of target information indicated by the recommendation rule to the number of positions of the unit recommendation window is greater than or equal to the expected recommendation ratio.
[0034] In one possible implementation, the unit recommendation window may be of at least one type, including a sliding window and a header window.
[0035] On the other hand, a computer device is provided, the computer device including a processor and a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to implement the above-described information recommendation method.
[0036] On the other hand, a computer-readable storage medium is provided, wherein at least one computer program is stored in the computer program, which is loaded and executed by a processor to implement the above-described information recommendation method.
[0037] On the other hand, a computer program product is provided, the computer program product comprising at least one computer program, the computer program being loaded and executed by a processor to implement the information recommendation method provided in the various alternative implementations described above.
[0038] The technical solution provided in this application may include the following beneficial effects:
[0039] The information recommendation method provided in this application adjusts the first ranking of each target information based on the recommendation rules of the target recommendation domain to which each target information belongs, so that the ranking position of each target information satisfies the recommendation rules of each target recommendation domain. The recommendation rules are used to indicate the numerical range of the number of target information that can be recommended in the corresponding unit recommendation window of the target recommendation domain. By setting the minimum value of the numerical range, the effective recommendation probability of each target recommendation domain is guaranteed. While avoiding large-scale changes to the original fine ranking order, the layout of the recommendation domain can be adjusted efficiently and quickly, thereby improving the overall recommendation effect of the recommendation system.
[0040] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description
[0041] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0042] Figure 1 This illustration shows a schematic diagram of the information recommendation process corresponding to the information recommendation method provided in an exemplary embodiment of this application;
[0043] Figure 2 A flowchart illustrating an exemplary embodiment of this application is shown.
[0044] Figure 3 A flowchart of another information recommendation method provided by an exemplary embodiment of this application is shown;
[0045] Figure 4 This illustration shows a schematic diagram of a sorting adjustment provided by an exemplary embodiment of this application;
[0046] Figure 5 A schematic diagram illustrating another sorting adjustment provided by an exemplary embodiment of this application is shown;
[0047] Figure 6 A schematic diagram of the system framework corresponding to the information recommendation method provided in an exemplary embodiment of this application is shown;
[0048] Figure 7 A block diagram of an information recommendation apparatus provided in an exemplary embodiment of this application is shown;
[0049] Figure 8 This is a structural block diagram of a computer device according to an exemplary embodiment. Detailed Implementation
[0050] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.
[0051] Figure 1 This illustration shows a schematic diagram of the information recommendation process corresponding to the information recommendation method provided in an exemplary embodiment of this application, such as... Figure 1 As shown, the information recommendation process may include: information recall 110, information coarse ranking 120, information fine ranking 130, information re-ranking 140, and information recommendation 150.
[0052] The goal of information recall is to retrieve a few thousand pieces of information from tens of millions or even hundreds of millions of candidate pieces of information.
[0053] The ranking stage is used to sort the recalled results and use the Top-K results as the final output of the recommendation system. Among them, coarse information ranking is used to further reduce the amount of information sent to the next stage while ensuring a certain level of accuracy. Coarse information ranking is optional. Fine information ranking is used to personalize the results of coarse ranking as accurately as possible.
[0054] After fine-grained information ranking, the problem of "information cocoons" often arises. For users, the clustering and overlap of similar content in the recommendation results can easily lead to user fatigue and a decline in novelty. For recommendation system platforms, the highly homogenized content fails to showcase the core features of the product, potentially causing user churn in the long run. To address this issue, information re-ranking is introduced after fine-grained information ranking, primarily addressing two major aspects: user experience and ecosystem traffic control. The goal is to shift the recommendation system away from maximizing the value of individual pieces of information. Through information re-ranking, a greater focus can be placed on optimizing the overall ecosystem, making the entire business ecosystem more rational and distinctive, and paying closer attention to medium- and long-term value gains. The information recommendation method provided in this application is applied to the information re-ranking stage of the recommendation process to quickly and efficiently solve the problem of ecosystem traffic control layout while ensuring that user experience is not compromised.
[0055] Figure 2 This application illustrates a flowchart of an information recommendation method provided by an exemplary embodiment. This method can be executed by a computer device, which can be implemented as a terminal or server in a recommendation system, such as... Figure 2 As shown, the information recommendation method includes:
[0056] Step 210: Obtain the first ranking of each target information; the first ranking refers to the recommended order of each target information after information fine sorting.
[0057] The information recommendation method provided in this application is applied to the information rearrangement stage in the recommendation process. Therefore, the first sort is the sorting result after the information is finely sorted, and each target information has a corresponding first sorting position in the first sort.
[0058] The target information can be all or part of the information that has entered the information rearrangement stage after the initial ranking. Optionally, the target information is information with a rule identifier that has entered the rearrangement stage; the rule identifier is used to indicate that the ranking position of the target information needs to meet the corresponding recommendation rule.
[0059] Step 220: Obtain the recommendation rule for the target recommendation domain to which each target information belongs; the recommendation rule is used to indicate the numerical range of the number of target information that can be recommended in the corresponding unit recommendation window of the target recommendation domain; the numerical range includes the minimum value.
[0060] In this embodiment of the application, the target recommendation fields to which each target information belongs can be the same or different; that is, a target recommendation field can contain at least one target information; each target recommendation field can be set with at least one recommendation rule, which can be used to indicate the specifications of the unit recommendation window corresponding to the target recommendation field, that is, to indicate the number of information positions in the unit recommendation window, and the numerical range of the number of target information that the target recommendation field can recommend in the unit recommendation window; illustratively, the recommendation rule can indicate that target recommendation field A must recommend at least 1 target information and can recommend at most 2 target information in a unit recommendation window with a specification of 4 (the number of information positions is 4).
[0061] By setting the minimum number of output values within a numerical range, it can be ensured that the target recommendation domain has a sufficiently high probability of being recommended.
[0062] Optionally, a target information can belong to multiple target recommendation domains at the same time. For example, a target information can belong to both the animation domain and the film domain.
[0063] Step 230: Based on the recommendation rules of each target recommendation domain, adjust the first ranking to determine the second ranking of each target information.
[0064] The second ranking can be the result of superimposing and adjusting the recommendation rules of each target recommendation domain on the first ranking; illustratively, the recommendation rule for target recommendation domain A indicates that within a unit recommendation window with 4 information positions, at least 1 target information from that domain must be recommended, and at most 2 target information from that domain can be recommended; the recommendation rule for target recommendation domain B indicates that within a unit recommendation window with 5 information positions, at least 1 target information from that domain must be recommended, and at most 3 target information from that domain can be recommended; provided that the two recommendation rules do not conflict, the second ranking, adjusted according to the order of the target information indicated by the first ranking, satisfies the above two recommendation rules.
[0065] Step 240: Based on the second ranking, recommend each target information.
[0066] In summary, the information recommendation method provided in this application adjusts the first ranking of each target information based on the recommendation rules of the target recommendation domain to which each target information belongs, so that the ranking position of each target information satisfies the recommendation rules of each target recommendation domain. The recommendation rules indicate the numerical range of the number of target information that can be recommended in the corresponding unit recommendation window within the target recommendation domain. By setting the minimum value of the numerical range, the effective recommendation probability of each target recommendation domain is guaranteed. This avoids large-scale changes to the original ranking order while efficiently and quickly adjusting the layout of the recommendation domain, thus improving the overall recommendation effect of the recommendation system.
[0067] In this embodiment, before adjusting the order of each target information based on the recommendation rules of each target recommendation domain, the first ranking of each target information can be dynamically adjusted in real time to obtain a third ranking of each target information. Based on this third ranking, a second ranking is obtained by adjusting the recommendation rules of each target recommendation domain. Figure 3 This application illustrates a flowchart of another information recommendation method provided by an exemplary embodiment. This method can be executed by a computer device, which can be implemented as a terminal or server in a recommendation system, such as... Figure 3 As shown, the information recommendation method includes:
[0068] Step 310: Obtain the first ranking of each target information; the first ranking refers to the recommended order of each target information after information fine sorting.
[0069] Indicatively, the computer device can assign rule tags to information (i.e., target information) that meets specified conditions in the information input from the fine-ranking stage to the rearrangement stage. Indicatively, this rule tagging can be implemented by marking the target information with a rule identifier (i.e., tag_id) to indicate which recommendation rule needs to intervene in the target information. Optionally, a target information can be marked with multiple different rule identifiers to indicate that the target information can be intervened in by different recommendation rules at the same time to determine the final ranking position of the target information.
[0070] The specified condition can be set by relevant personnel based on actual needs. Alternatively, in another possible implementation, the computer device can count the amount of information contained in each recommendation domain in the first ranking, sort them from high to low based on the amount of information, and determine the recommendation domains at the bottom of the ranking as the target recommendation domains. The specified condition can be within the target recommendation domain. This application does not limit the way the specified condition is set.
[0071] Step 320: Obtain the actual recommendation percentage of the target recommendation domain to which each target information belongs; the actual recommendation percentage is used to indicate the proportion of the number of recommended messages in the target recommendation domain to the number of recommended messages within the target duration.
[0072] In this embodiment, the computer device can use a real-time computing platform to statistically analyze the actual recommendation ratio of each target recommendation domain in real time. Indicatively, the real-time computing platform can be a platform such as Flink. The target duration can be set by relevant personnel based on actual needs, and this application does not impose any restrictions. Indicatively, when the target duration is 30 minutes, the computer device uses the real-time computing platform to statistically analyze the number of recommendation messages in each target recommendation domain, as well as the total number of recommendation messages recommended within 30 minutes, to calculate the actual recommendation ratio of each target recommendation domain.
[0073] Step 330: Based on the difference between the actual recommendation ratio of each target recommendation domain and the expected recommendation ratio of each target recommendation domain, adjust the first ranking to determine the third ranking of each target information.
[0074] The expected recommendation ratio can be set based on business objectives.
[0075] In this embodiment, the computer device can determine weighting parameters for adjusting the ranking scores of each target information in each target recommendation domain based on the difference in the proportion of each target recommendation domain. After updating the ranking scores of each target information in each target recommendation domain according to the corresponding weighting parameters, the ranking position of each target information is re-determined based on the updated ranking scores of each target information in each target recommendation domain. This process can be implemented as follows:
[0076] Obtain the weighting parameters corresponding to each target recommendation domain. These weighting parameters are the ratio between the difference in the proportion of the target recommendation domain and the expected recommendation proportion of the target recommendation domain.
[0077] Based on the weighting parameters corresponding to each target recommendation domain, the first ranking score of each target information in each target recommendation domain is adjusted to determine the second ranking score of each target information in each target recommendation domain.
[0078] Based on the second ranking score of each target information in each target recommendation domain, the ranking position of each target information in the third ranking is determined.
[0079] Taking any recommendation domain among the target recommendation domains as an example, the above process can be implemented as follows:
[0080] Obtain the weighting parameter for the second target recommendation domain. This weighting parameter is the ratio between the proportion difference of the second target recommendation domain and the expected recommendation proportion of the second target recommendation domain; the second target recommendation domain is one of the various target recommendation domains.
[0081] The first ranking score of each first target information is adjusted based on the weighting parameters to determine the second ranking score of each first target information in the second target recommendation domain; the first target information is the target information in the second target recommendation domain.
[0082] Based on the second ranking score of each first target information, the ranking position of each first target information in the third ranking is determined.
[0083] In this embodiment, dynamic weighting coefficients corresponding to each target recommendation domain can be calculated based on the weighting parameters corresponding to each target recommendation domain. The first ranking score of each target information in each target recommendation domain can be adjusted according to the dynamic weighting coefficients corresponding to each target recommendation domain, thereby determining the second ranking score of each target information in each target recommendation domain.
[0084] The dynamic weighting parameter may include a first weighting coefficient and a second weighting coefficient.
[0085] As an illustration, the weighting parameters for each target recommendation domain can be calculated using the following formula:
[0086] delta = (Expected recommendation percentage - Actual recommendation percentage) / Expected recommendation percentage
[0087] The value of (expected recommendation percentage - actual recommendation percentage) is the difference in percentage within the target recommendation area.
[0088] The first adjustment coefficient can be calculated using the following formula:
[0089]
[0090] Where delta represents the weighting parameter, weight represents the first weighting coefficient, and m is the hyperparameter.
[0091] The second adjustment coefficient can be calculated using the following formula:
[0092] bias = delta × k
[0093] Where k is a hyperparameter.
[0094] The second ranking score for each target information can be calculated using the following formula:
[0095] score2 = score1 × weight + bias
[0096] Here, score2 represents the second ranking score, and score1 represents the first ranking score.
[0097] In one possible implementation, the method further includes, before adjusting the ranking of the information using dynamic weighting coefficients:
[0098] The first ranking score of each second objective information is adjusted by static weighting coefficients to obtain the third ranking score of each second objective information.
[0099] The second target information is information within the target recommendation domain that has a weighting indicator. This weighting indicator can be illustratively marked on the recommendation domain based on target business needs. For example, if the target business needs indicate a need to increase the number of recommended information in the video domain, a weighting indicator can be marked on the video domain so that when the computer device recognizes the weighting indicator, it can adjust the target information in the video domain using a static weighting coefficient. This static weighting coefficient can be set by relevant personnel.
[0100] As an illustration, adjusting the ranking score using static weighting coefficients can be achieved through the following formula:
[0101] score3 = score1 × weight1
[0102] Where score3 represents the third ranking score, score1 represents the first ranking score, and weight1 represents the static weighting coefficient.
[0103] If the second target information only needs to be adjusted for ranking based on static weighting coefficients, then the third ranking score is determined as the second ranking score of the second target information.
[0104] If the second target information needs to be adjusted for ranking score based on both static and dynamic weighting coefficients, then after obtaining the third ranking score of the second target information, the third ranking score is adjusted using dynamic weighting coefficients (including the first and second weighting coefficients) to obtain the second ranking score of the second target information.
[0105] Step 340: Obtain the recommendation rule for the target recommendation domain to which each target information belongs; the recommendation rule is used to indicate the numerical range of the number of target information that can be recommended in the corresponding unit recommendation window in the target recommendation domain; the numerical range includes the minimum value.
[0106] Optionally, the unit recommends that the window type include at least one of a sliding window and a header window (TOP window).
[0107] This numerical range includes a minimum and a maximum value. The minimum value ensures the probability of recommending information in the target recommendation domain, while the maximum value limits the recommendation frequency of information in the target recommendation domain, thus avoiding clustered recommendations of target information in the same target recommendation domain, which would negatively impact user experience.
[0108] The recommendation rule is pre-set based on actual needs. It indicates the specifications of the unit recommendation window within the target recommendation domain, the numerical range of the number of target information items that can be recommended within that unit recommendation window, and the type of recommendation window for that unit. Since multiple recommendation rules can be set for the same target recommendation domain, and recommendation rules from different domains may produce conflicting sorting results when applied to information rearrangement, in this embodiment, each recommendation rule has an application priority. Optionally, the recommendation rule also includes a priority identifier to indicate the application priority of the recommendation rule.
[0109] Based on the above description of recommendation rules, this application schematically illustrates a structural form of recommendation rules, as shown below. The rules in the rule base may include:
[0110] Rule 1: id=1, type=sliding window, tag_id=1000; priority identifier=1, window parameters=4,1,2; remarks=In a window with size 4, the maximum number of output messages for domain A is 2, and the minimum number of output messages is 1.
[0111] Rule 2: id=2, type=sliding window, tag_id=1001; priority identifier=2, window parameters=5,2,3; remarks=In the B domain, within a window of size 5, the maximum number of outputs is limited to 3, and the minimum number of outputs is 2;
[0112] Rule 3: id=3, type=TOP window, tag_id=1002; priority identifier=1, window parameters=5,1,3; remarks=C. Within a window of size 5, the maximum number of outputs is limited to 3, and the minimum number of outputs is 1.
[0113] ...
[0114] Rule n: id=n, type=sliding window, tag_id=xxxx; priority identifier=1, window parameters=9,3,5; remarks=xx domain within a window of size 9, the maximum number of outputs is 5, and the minimum number of outputs is 3.
[0115] Optionally, since each target information is marked with a rule identifier (i.e., tag_id), the computer device can obtain the recommendation rule for the target recommendation domain corresponding to the target information from the pre-set rules based on the rule identifier. For example, when the rule identifier marked on the target information is 1000, the computer device can obtain rule 1 from the rule base based on the rule identifier and determine rule 1 as the recommendation rule for the target recommendation domain corresponding to the target information.
[0116] Optionally, since multiple rule identifiers can be marked on the same target information, that is, multiple recommendation rules can correspond to the same target recommendation domain, in this case, the computer device can obtain the recommendation rules of each target recommendation domain based on the expected recommendation ratio corresponding to each target recommendation domain.
[0117] Among them, the ratio of the maximum number of target information indicated by the recommendation rule to the number of positions of the unit recommendation window is greater than or equal to the expected recommendation ratio.
[0118] For illustration, if the expected recommendation ratio of target recommendation domain 1 is 50%, it corresponds to two recommendation rules, namely recommendation rule 1: 4, 1, 2 and recommendation rule 2: 5, 1, 2. At this time, the ratio of the maximum number of target information in recommendation rule 1 to the number of positions of unit recommendation window is equal to 50%, while the ratio of the maximum number of target information in recommendation rule 2 to the number of positions of unit recommendation window is less than 50%. Therefore, the recommendation rule for target recommendation domain 1 at this time is determined to be recommendation rule 1.
[0119] In another scenario, if the ratio of the maximum number of target information to the number of positions in a unit recommendation window is greater than or equal to the expected recommendation ratio in both recommendation rules of the target recommendation domain, then one of the recommendation rules can be randomly selected as the recommendation rule for this target recommendation domain; alternatively, the recommendation rule that is traversed first can be selected as the recommendation rule for this target recommendation domain based on the order in which the computer traverses the two recommendation rules.
[0120] It should be noted that step 350 can also be executed before step 320, and this application does not restrict the execution order of step 350.
[0121] Step 350: Based on the recommendation rules of each target recommendation domain, adjust the third ranking to determine the second ranking of each target information.
[0122] Different target recommendation domains can correspond to different recommendation rules, and different recommendation rules can be combined and applied to the third ranking. Taking the type of unit recommendation window corresponding to the recommendation rule as a sliding window as an example, Figure 4 This illustration shows a schematic diagram of a sorting adjustment provided by an exemplary embodiment of this application, such as... Figure 4 As shown, the third sorting 410 contains 10 pieces of information, and these 10 pieces of information belong to 6 recommendation domains respectively. Information 2, information 3, and information 4 belong to target recommendation domain 1, and information 9 and information 10 belong to target recommendation domain 2. If the rule indicating the target recommendation domain 1 is in a sliding window of size 4, the number of pieces of information in target recommendation domain 1 is at least 1 and at most 2; if the rule indicating the target recommendation domain 2 is in a sliding window of size 4, the number of pieces of information in target recommendation domain 2 is at least 1 and at most 2. Since the first 4 pieces of information in the third sorting do not contain information from target recommendation domain 2, and the first 4 pieces of information contain 3 pieces of information from target recommendation domain 1, this does not meet the requirement. The rules corresponding to target recommendation domain 1 are as follows: Information 9 needs to be moved to the fourth position in the sliding window according to the order indicated by the third sorting instruction, and information 4 needs to be moved one position to the right. The adjusted information sorting is shown in sorting 420. When the sliding window slides, based on the order of sorting 420, it can be seen that the four pieces of information in the sliding window contain three pieces of information from target recommendation domain 1, which does not conform to the rules corresponding to target recommendation domain 1 and target recommendation domain 2 respectively. Therefore, according to the order indicated by sorting 420, information 4 is swapped with its adjacent information 5. The adjusted information sorting is shown in sorting 430. This process continues, and will not be elaborated further here. Finally, the information sorting is shown in sorting 440.
[0123] Taking the type of unit recommendation window corresponding to the recommendation rule as the header window as an example, Figure 5 This illustration shows another sorting adjustment provided by an exemplary embodiment of the present application, such as... Figure 5 As shown, the third sorting 510 contains 10 pieces of information, and these 10 pieces of information belong to 6 recommendation domains respectively. Information 2, information 3, and information 4 belong to target recommendation domain 1, and information 9 and information 10 belong to target recommendation domain 2. If the rule indication for target recommendation domain 1 is in a sliding window of size 5, the number of pieces of information in target recommendation domain 1 is at least 1 and at most 2. If the rule indication for target recommendation domain 2 is in a sliding window of size 5, the number of pieces of information in target recommendation domain 2 is at least 1 and at most 2. Since the first 5 pieces of information in the third sorting do not contain information from target recommendation domain 2, and the first 5 pieces of information contain 3 pieces of information from target recommendation domain 1, it does not conform to the rules corresponding to target recommendation domain 1 and target recommendation domain 2 respectively. Therefore, according to the order of the third sorting indication, information 9 needs to be moved to the sliding window, and information 4 needs to be moved to the sixth position. Since the sorting score of information 5 is higher than that of information 9, information 5 is moved to the fourth position, and information 9 is moved to the fifth position. The adjusted information sorting is shown in sorting 520.
[0124] Based on the above process of adjusting the sorting through sliding windows and header windows, it can be seen that when adjusting the information sorting, it is necessary to satisfy the recommendation rules while maintaining the information sorting obtained based on the sorting score as much as possible.
[0125] When adjusting the sorting using a sliding window, it is necessary to ensure that the information contained in the sliding window conforms to the recommendation rules during the sliding process; when adjusting the sorting using a header window, it is only necessary to ensure that the information in the header window conforms to the recommendation rules.
[0126] Optionally, if the sorting determined based on the numerical range of the first recommendation rule conflicts with the sorting determined based on the numerical range of the second recommendation rule, then the sorting determined based on the numerical range of the first recommendation rule shall be the second sorting.
[0127] The first recommendation rule has a higher application priority than the second recommendation rule.
[0128] The first recommendation rule and the second recommendation rule can be recommendation rules corresponding to the same target recommendation domain, or they can be recommendation rules corresponding to different target recommendation domains.
[0129] For illustration, if recommendation rule 1 for target recommendation domain 1 is 3, 2, 3, and recommendation rule 2 for target recommendation domain 2 is 4, 3, 4, and both recommendation rule 1 and recommendation rule 2 apply to the sliding window, there is a conflict between the two recommendation rules if there is no information belonging to neither target recommendation domain 1 nor target recommendation domain 2. That is, if recommendation rule 1 is satisfied, recommendation rule 2 is not satisfied, and if recommendation rule 2 is satisfied, recommendation rule 1 is not satisfied. In this case, the application priority of recommendation rule 1 and recommendation rule 2 can be obtained, and the recommendation rule with higher priority is satisfied first. The information in the sliding window should be made as close as possible to the recommendation rule with lower priority. For illustration, if the application priority of recommendation rule 1 is higher than the priority of recommendation rule 2, then when there are already 2 pieces of information from target recommendation domain 1 in the sliding window of size 4, 2 pieces of information from target recommendation domain 2 are arranged in the other two positions.
[0130] In one possible implementation, the process of adjusting the third ranking based on the recommendation rules of each target recommendation domain to determine the second ranking of each target information can be implemented as follows:
[0131] Based on the recommendation rules of each recommendation domain, the third ranking is adjusted once to determine a temporary ranking;
[0132] If the number of target information in each target recommendation domain in each unit recommendation window corresponding to the temporary ranking meets the corresponding recommendation rules, then the temporary ranking will be determined as the second ranking.
[0133] If the number of target information items in the first target recommendation domain corresponding to the first recommendation window in the temporary sorting does not meet the minimum value within the numerical range indicated by the recommendation rule for the first target recommendation domain, then, while ensuring that the number of target information items in other target recommendation domains within the first recommendation window meets the corresponding recommendation rule, target information items from other target recommendation domains are added to the first recommendation window according to the order indicated by the temporary sorting. This first recommendation window can be any one of the unit recommendation windows; other target recommendation domains are target recommendation domains within the target recommendation domains corresponding to the first recommendation window that are not part of the first target recommendation domain.
[0134] The adjusted temporary sort is designated as the second sort.
[0135] The first target recommendation domain is the target recommendation domain in which the number of target information in each target recommendation domain corresponding to the first recommendation window does not meet the minimum value in the numerical range indicated by the corresponding recommendation rule.
[0136] In other words, after adjusting the third ranking based on each recommendation rule, if the number of target information in each target recommendation domain in each unit recommendation window meets the corresponding recommendation rule, then the third ranking adjusted based on each recommendation rule will be determined as the second ranking.
[0137] If, after adjusting the third ranking based on each recommendation rule, the number of target information in the first target recommendation domain still does not meet the minimum value of its corresponding recommendation rule, it indicates that the number of information in the first target recommendation domain after information refinement is too small to meet the corresponding recommendation rule. In this case, to avoid recommendation gaps, under the premise of satisfying the recommendation rules of other target recommendation domains, the maximum value of the number of target information indicated by the recommendation rules of other target recommendation domains is used as a limit to add target information from other target recommendation domains to the corresponding unit recommendation window. For example, if the recommendation rule for target recommendation domain 1 is 4, 2, 3, and the recommendation rule for target recommendation domain 2 is 4, 2, 3, when both target recommendation domains satisfy their corresponding recommendation rules, a unit recommendation window should contain 2 pieces of information from target recommendation domain 1 and 2 pieces of information from target recommendation domain 2. However, at this time, there is only 1 piece of information left in target recommendation domain 2, which cannot meet the minimum value requirement of its corresponding recommendation rule. In this case, the computer device can add 1 more piece of information from target recommendation domain 1 to the unit recommendation window, with the maximum value of the recommendation rule for target recommendation domain 1 as a limit.
[0138] Step 360: Based on the second ranking, recommend information for each target.
[0139] In one possible implementation, after obtaining the second ranking of each target information, if there are other re-ranking rules, the second ranking is adjusted based on the other re-ranking rules, and the target information is recommended based on the adjusted second ranking.
[0140] For example, other rearrangement rules may include forced insertion rules, which involve inserting specific information at a fixed position to meet specific business needs; for example, after obtaining the second sort of content information, an advertisement is forcibly inserted at the 5th position of the second sort to meet specific business needs.
[0141] In summary, the information recommendation method provided in this application, while dispersing the ranking of each target information through recommendation rules with minimum output constraints, also incorporates a dynamic weighting coefficient obtained based on the difference between the actual recommendation ratio and the expected recommendation ratio. The ranking score of each target information is adjusted based on this dynamic weighting coefficient to fine-tune the ranking of each target information, gradually converging it to the expected recommendation ratio of the corresponding target recommendation domain. This ensures an effective recommendation probability for each target recommendation domain, avoids large-scale changes to the original fine-grained ranking, and makes the dynamic weighting coefficient more controllable, reducing the impact of manually set weights on the recommendation effect. Thus, it achieves efficient and rapid adjustment of the layout of recommendation domains, improving the overall recommendation effect of the recommendation system.
[0142] It should be noted that the information recommendation method provided in this application includes two ranking adjustment methods: ranking adjustment based on recommendation rules of the target recommendation domain, and ranking adjustment based on dynamic weighting coefficients obtained from the actual and expected recommendation proportions of the target recommendation domain; the two methods can be used individually or in combination. Figure 3 The illustrated embodiment corresponds to the case where both are used in combination. Figure 2 The illustrated embodiment corresponds to the case where ranking adjustments are made solely based on recommendation rules within the target recommendation domain; if ranking adjustments are made solely based on dynamic weighting coefficients, the following steps can be performed: Figure 3 In the corresponding embodiment, after obtaining the third sorting of each target information, the target information can be recommended according to the third sorting. This application will not elaborate further.
[0143] based on Figure 3 The content of the illustrated embodiment, Figure 6 A schematic diagram of the system framework corresponding to the information recommendation method provided in an exemplary embodiment of this application is shown, such as... Figure 6As shown, after receiving a front-end request sent by the user terminal 620, the recommendation system 610 starts the recommendation process. After information retrieval, coarse information ranking, and fine information ranking, it enters the information re-ranking stage. In the information re-ranking stage, the computer device can perform static weight adjustment on target information with adjustment indicators, dynamic weight adjustment on each target information, scatter target information with rule indicators by setting recommendation rules with minimum values, and sort and adjust the ranking based on other re-ranking rules. Then, it recommends each target information according to the adjusted ranking.
[0144] Among them, static weight adjustment refers to adjusting the ranking score through static weight adjustment coefficient, so as to sort based on the adjusted ranking score;
[0145] Dynamic weighting refers to calculating the dynamic weighting coefficient for each target recommendation domain based on the actual recommendation ratio and the expected recommendation ratio for each target recommendation domain, and then adjusting the ranking score of the target information in each target recommendation domain based on the dynamic weighting coefficient, so as to rank the information based on the adjusted ranking score.
[0146] The actual recommendation percentage can be obtained by statistically analyzing the recommendation log data from the user's end. For example, computer devices process the recommendation log data through a real-time computing platform to obtain the actual recommendation percentage for each target recommendation domain.
[0147] For details on how to break down target information with rule identifiers using recommendation rules with minimum values, please refer to [link / reference]. Figure 3 The relevant details in the illustrated embodiments will not be repeated here.
[0148] Table 1 shows a comparison between the recommendation results of the information recommendation method provided in this application and the recommendation results of the information recommendation method in related technologies. As shown in Table 1, the information recommendation method provided in this application can make significant adjustments to the recommendation ecosystem, thereby completing the ecosystem layout, making the recommended content more distinctive, in line with the long-term development direction of the product, and making the entire recommendation business ecosystem more reasonable and in line with long-term goal value.
[0149] Table 1
[0150]
[0151]
[0152] In summary, this application proposes a method for dynamic weighting based on the domain proportion of recommendation log data statistics. This method can dynamically change the weighting based on the difference between the expected recommendation proportion and the actual recommendation proportion, keeping the weights in an appropriate state. This avoids situations where the weights are too small to achieve the expected results, or where the weights are too large, thereby preventing excessive weights from disrupting the ranking order, causing insufficient accuracy in the recommendation results, and resulting in a significant decline in user experience.
[0153] This application proposes a window scattering method with minimum output rules (including sliding window scattering and head window scattering). Compared with the previous sliding window rules that only have the function of maximum output restriction, the method provided by this application will be more efficient in ecological field layout. This is because the traditional sliding window rules only have the function of maximum output restriction and can only scatter the head ecological fields to prevent them from clustering together. However, the window rules proposed in this application with minimum output function not only have the function of previous window scattering, but also allow scarce or long-tail fields in the ecosystem to obtain sufficient exposure opportunities under the scattering rules, so as to adjust the layout of the recommended ecological fields more efficiently and quickly.
[0154] When adjusting the recommendation ecosystem layout during the reordering stage, this application can comprehensively utilize a method of dynamically adjusting the weights based on the domain proportions statistically derived from recommendation log data, and a window scattering method with minimum output rules. The combination of these two methods can effectively and accurately adjust the domain proportions in the recommendation ecosystem. This is because the window scattering method with minimum output rules can define the fluctuation range of the exposure proportion of the corresponding domain. Combined with the method of dynamically adjusting the weights based on the domain proportions statistically derived from recommendation log data, the proportions are fine-tuned through dynamic weighting / reduction, gradually converging to the desired recommendation proportion. This also makes the weight adjustment weights more controllable, greatly reducing the occurrence of extreme phenomena such as excessively large weight adjustments and avoiding user experience loss.
[0155] Figure 7 A block diagram of an information recommendation apparatus provided in an exemplary embodiment of this application is shown, such as Figure 7 As shown, the information recommendation device includes:
[0156] The sorting acquisition module 710 is used to acquire the first sorting of each target information; the first sorting refers to the recommended order of each target information after information fine sorting.
[0157] The rule acquisition module 720 is used to acquire the recommendation rules of the target recommendation domain to which each target information belongs; the recommendation rules are used to indicate the numerical range of the number of target information that can be recommended in the corresponding unit recommendation window of the target recommendation domain; the numerical range includes the minimum value.
[0158] The sorting adjustment module 730 is used to adjust the first sorting based on the recommendation rules of each of the target recommendation domains to determine the second sorting of each of the target information.
[0159] The information recommendation module 740 is used to recommend each of the target information based on the second ranking.
[0160] In one possible implementation, the sorting adjustment module 730 includes:
[0161] The percentage acquisition submodule is used to acquire the actual recommendation percentage of the target recommendation domain to which each of the target information belongs; the actual recommendation percentage is used to indicate the percentage of the number of recommended messages in the target recommendation domain within the target duration compared to the number of already recommended messages;
[0162] The first sorting adjustment submodule is used to sort the difference between the actual recommendation ratio of each target recommendation domain and the expected recommendation ratio corresponding to each target recommendation domain, adjust the first sorting, and determine the third sorting of each target information.
[0163] The second sorting adjustment submodule is used to adjust the third sorting based on the recommendation rules of each of the target recommendation domains, and to determine the second sorting of each of the target information.
[0164] In one possible implementation, the first sorting adjustment submodule includes:
[0165] The parameter acquisition unit is used to acquire the weighting parameters corresponding to each of the target recommendation domains, wherein the weighting parameters are the ratio between the proportion difference of the target recommendation domain and the expected recommendation proportion of the target recommendation domain.
[0166] The score adjustment unit is used to adjust the first ranking score of each target information in each target recommendation domain based on the weighting parameters corresponding to each target recommendation domain, and to determine the second ranking score of each target information in each target recommendation domain.
[0167] The position determination unit is used to determine the ranking position of each of the target information in the third ranking based on the second ranking score of each of the target information in each of the target recommendation domains.
[0168] In one possible implementation, the recommendation rules have an application priority.
[0169] The second sorting adjustment submodule is used to determine the sorting determined by the numerical range of the first recommendation rule as the second sorting if the sorting determined by the numerical range of the first recommendation rule conflicts with the sorting determined by the numerical range of the second recommendation rule.
[0170] In this case, the application priority of the first recommendation rule is higher than the application priority of the second recommendation rule.
[0171] In one possible implementation, the second sorting adjustment submodule includes:
[0172] The first adjustment unit is used to adjust the third ranking based on the recommendation rules of each of the recommendation domains to determine a temporary ranking;
[0173] The sorting determination unit is configured to determine the temporary sorting as the second sorting if the number of target information in each target recommendation domain in each unit recommendation window corresponding to the temporary sorting satisfies the corresponding recommendation rule.
[0174] The second adjustment unit is configured to, if the number of target information in the first target recommendation field corresponding to the first recommendation window in the temporary sorting does not meet the minimum value in the numerical range indicated by the recommendation rule corresponding to the first target recommendation field, add the target information of the other target recommendation fields in the first recommendation window according to the order indicated by the temporary sorting, provided that the number of target information in the other target recommendation fields in the first recommendation window meets the corresponding recommendation rule; the first recommendation window is any one of the various unit recommendation windows; the other target recommendation fields are target recommendation fields other than the first target recommendation field in the target recommendation fields corresponding to the first recommendation window;
[0175] The sorting determination unit is used to determine the adjusted temporary sorting as the second sorting.
[0176] In one possible implementation, the numerical range further includes a maximum value; the rule acquisition module 720 is used to acquire the recommendation rules for each of the target recommendation domains based on the expected recommendation ratio corresponding to each of the target recommendation domains.
[0177] Wherein, the ratio of the maximum number of target information indicated by the recommendation rule to the number of positions of the unit recommendation window is greater than or equal to the expected recommendation ratio.
[0178] In one possible implementation, the unit recommendation window may be of at least one type, including a sliding window and a header window.
[0179] In summary, the information recommendation device provided in this application adjusts the first sorting of each target information based on the recommendation rules of the target recommendation domain to which each target information belongs, so that the sorting position of each target information satisfies the recommendation rules of each target recommendation domain. The recommendation rules indicate the numerical range of the number of target information that can be recommended in the corresponding unit recommendation window within the target recommendation domain. By setting the minimum value of the numerical range, the effective recommendation probability of each target recommendation domain is guaranteed. While avoiding large-scale changes to the original fine-ranking order, the layout of the recommendation domain can be adjusted efficiently and quickly, improving the overall recommendation effect of the recommendation system.
[0180] Figure 8 A structural block diagram of a computer device 800 illustrated in an exemplary embodiment of this application is shown. This computer device can be implemented as a server in the recommended system described above. The computer device 800 includes a Central Processing Unit (CPU) 801, a system memory 804 including Random Access Memory (RAM) 802 and Read-Only Memory (ROM) 803, and a system bus 805 connecting the system memory 804 and the CPU 801. The computer device 800 also includes a mass storage device 806 for storing an operating system 809, a client 810, and other program modules 811.
[0181] Without loss of generality, the computer-readable medium may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer storage media include RAM, ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other solid-state memory technologies, CD-ROM, digital versatile disc (DVD) or other optical storage, magnetic tape cassettes, magnetic tape, disk storage, or other magnetic storage devices. Of course, those skilled in the art will recognize that the computer storage media are not limited to the above-mentioned types. The system memory 804 and mass storage device 806 described above can be collectively referred to as memory.
[0182] According to various embodiments of this application, the computer device 800 can also be connected to a remote computer on a network, such as the Internet. That is, the computer device 800 can be connected to a network 808 via a network interface unit 807 connected to the system bus 805, or the network interface unit 807 can be used to connect to other types of networks or remote computer systems (not shown).
[0183] The memory further includes at least one instruction, at least one program, code set, or instruction set, which are stored in the memory. The central processing unit 801 executes the at least one instruction, at least one program, code set, or instruction set to implement all or part of the steps in the information recommendation method shown in the above embodiments.
[0184] In one exemplary embodiment, a computer-readable storage medium is also provided, which stores at least one computer program that is loaded and executed by a processor to implement all or part of the steps in the information recommendation method described above. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, or optical data storage device, etc.
[0185] In one exemplary embodiment, a computer program product is also provided, comprising at least one computer program that is loaded and executed by a processor. Figure 2 or Figure 3 All or part of the steps of the information recommendation method shown in any embodiment.
[0186] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.
[0187] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.
Claims
1. An information recommendation method, characterized in that, The method includes: Obtain a first ranking of each target information; the first ranking refers to the recommended order of each target information after information fine sorting. Obtain the recommendation rules for the target recommendation domain to which each target information belongs; the recommendation rules are used to indicate the numerical range of the number of target information that can be recommended in the corresponding unit recommendation window for the target recommendation domain; the numerical range includes the minimum value; Based on the recommendation rules of each of the target recommendation domains, the first ranking is adjusted to determine the second ranking of each of the target information; Based on the second ranking, recommendations are made for each of the target information; The step of adjusting the first ranking based on the recommendation rules of each of the target recommendation domains to determine the second ranking of each of the target information includes: Obtain the actual recommendation percentage of the target recommendation domain to which each of the target information belongs; the actual recommendation percentage is used to indicate the proportion of the number of recommended messages in the target recommendation domain to the number of recommended messages within the target duration; Based on the ratio difference between the actual recommendation ratio of each target recommendation domain and the expected recommendation ratio corresponding to each target recommendation domain, the first ranking is adjusted to determine the third ranking of each target information; the ratio difference is used to determine the weighting parameter for adjusting the ranking score of each target information in the target recommendation domain, and the weighting parameter is the ratio between the ratio difference of the target recommendation domain and the expected recommendation ratio of the target recommendation domain; Based on the recommendation rules of each of the target recommendation domains, the third ranking is adjusted to determine the second ranking of each of the target information.
2. The method according to claim 1, characterized in that, The method of adjusting the first ranking based on the difference between the actual recommendation ratio of each target recommendation domain and the expected recommendation ratio corresponding to each target recommendation domain, and determining the third ranking of each target information, includes: Obtain the weighting parameters corresponding to each of the target recommendation domains; Based on the weighting parameters corresponding to each of the target recommendation domains, the first ranking score of each of the target information in each of the target recommendation domains is adjusted to determine the second ranking score of each of the target information in each of the target recommendation domains; Based on the second ranking score of each target information in each target recommendation domain, the ranking position of each target information in the third ranking is determined.
3. The method according to claim 1, characterized in that, There is an application priority among the various recommendation rules; The process of adjusting the third ranking based on the recommendation rules of each of the target recommendation domains to determine the second ranking of each of the target information includes: If the sorting determined based on the numerical range of the first recommendation rule conflicts with the sorting determined based on the numerical range of the second recommendation rule, then the sorting determined based on the numerical range of the first recommendation rule shall be the second sorting. In this case, the application priority of the first recommendation rule is higher than the application priority of the second recommendation rule.
4. The method according to claim 1, characterized in that, The process of adjusting the third ranking based on the recommendation rules of each of the target recommendation domains to determine the second ranking of each of the target information includes: Based on the recommendation rules of each of the recommendation domains, the third ranking is adjusted once to determine a temporary ranking; If the number of target information in each target recommendation domain in each unit recommendation window corresponding to the temporary sorting satisfies the corresponding recommendation rule, then the temporary sorting is determined as the second sorting; If the number of target information in the first target recommendation field corresponding to the first recommendation window in the temporary sorting does not meet the minimum value in the numerical range indicated by the recommendation rule corresponding to the first target recommendation field, the target information of the other target recommendation fields in the first recommendation window is added to the first recommendation window according to the order indicated by the temporary sorting, provided that the number of target information in the other target recommendation fields in the first recommendation window meets the corresponding recommendation rule; the first recommendation window is any one of the various unit recommendation windows; the other target recommendation fields are target recommendation fields other than the first target recommendation field in the target recommendation fields corresponding to the first recommendation window; The adjusted temporary sort is determined as the second sort.
5. The method according to claim 1, characterized in that, The numerical range also includes a maximum value; the recommendation rules for obtaining the target recommendation domain to which each target information belongs include: Based on the expected recommendation ratio corresponding to each of the target recommendation domains, the recommendation rules for each of the target recommendation domains are obtained; Wherein, the ratio of the maximum number of target information indicated by the recommendation rule to the number of positions of the unit recommendation window is greater than or equal to the expected recommendation ratio.
6. The method according to claim 1, characterized in that, The unit recommendation window can be of at least one type, including a sliding window and a header window.
7. An information recommendation device, characterized in that, The device includes: The sorting acquisition module is used to acquire the first sorting of each target information; the first sorting refers to the recommended order of each target information after information fine sorting. The rule acquisition module is used to acquire the recommendation rules of the target recommendation domain to which each target information belongs; the recommendation rules are used to indicate the numerical range of the number of target information that can be recommended in the corresponding unit recommendation window of the target recommendation domain; the numerical range includes the minimum value; The sorting adjustment module is used to adjust the first sorting based on the recommendation rules of each of the target recommendation domains to determine the second sorting of each of the target information. The information recommendation module is used to recommend each of the target information based on the second ranking; The sorting adjustment module includes: The percentage acquisition submodule is used to acquire the actual recommendation percentage of the target recommendation domain to which each of the target information belongs; the actual recommendation percentage is used to indicate the percentage of the number of recommended messages in the target recommendation domain within the target duration compared to the number of already recommended messages; The first ranking adjustment submodule is used to adjust the first ranking based on the ratio difference between the actual recommendation ratio of each target recommendation domain and the expected recommendation ratio corresponding to each target recommendation domain, thereby determining the third ranking of each target information; the ratio difference is used to determine the weighting parameter for adjusting the ranking score of each target information in the target recommendation domain, and the weighting parameter is the ratio between the ratio difference of the target recommendation domain and the expected recommendation ratio of the target recommendation domain; The second sorting adjustment submodule is used to adjust the third sorting based on the recommendation rules of each of the target recommendation domains, and to determine the second sorting of each of the target information.
8. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing at least one computer program, which is loaded and executed by the processor to implement the information recommendation method as described in any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to implement the information recommendation method as described in any one of claims 1 to 6.