An information recommendation method, device, equipment and storage medium

Abstract

The application relates to the technical field of information recommendation, and particularly relates to an information recommendation method, device and equipment and a storage medium, the method comprising the following steps: in response to an information recommendation request, using multiple different recall strategies to determine respective corresponding recall information sets from a preset information library; obtaining historical data of information recommendation, determining correction parameters of each recall strategy based on the historical data; determining actual recall gears corresponding to each recall strategy according to the correction parameters of each recall strategy and preset recall gears; for each recall strategy, determining a candidate recommendation information set from the corresponding recall information set based on a preset sorting model and the actual recall gears; and determining a preset number of information from the candidate recommendation information sets corresponding to each recall strategy and recommending the information to a target user. The information recommendation method can guarantee the effect of information recommendation and improve the diversity of recommended information.

Description

Technical Field

[0001] This invention relates to the field of information recommendation technology, and in particular to an information recommendation method, apparatus, device, and storage medium. Background Technology

[0002] With the advent of the big data era, recommender systems have become the best solution for the public when faced with the problem of filtering massive amounts of internet information. The core of a recommender system is the recommender engine, which aims to filter useful information from large amounts of data and provide users with the information they need. Recommender engines can mine sample patterns in datasets by storing and analyzing historical information such as user behavior logs and item ratings, further analyzing user characteristics, and generating recommendations that are relevant to user needs and interests. These recommender systems not only provide users with convenient information filtering services but also enrich their online lives.

[0003] Existing recommendation systems typically employ multiple recall strategies for multi-path recall, retrieving relevant candidate information from various dimensions before determining the final recommendations. Two methods exist for determining the final recommendations: one involves aggregating the information retrieved by multiple strategies, ranking it using a ranking model, and then selecting a certain number of recommendations for the user. However, different ranking models have different biases—some favor popular information, while others prefer newly emerging information. If the retrieved information is aggregated and then ranked using a ranking model, the bias of the ranking model itself can weaken the effectiveness of different recall algorithms. This can lead to some recall algorithms failing to perform as intended and failing to meet the user's needs from multiple perspectives.

[0004] Another approach is to avoid aggregating information retrieved by multiple recall strategies. Instead, a ranking model is used to sort the information retrieved by each strategy separately, truncating it according to a fixed percentage or quantity, and then selecting a certain number of items from the truncated information to recommend to the user. This method treats all recall algorithms equally, ensuring the diversity of recommended information, but it fails to reflect the strengths and weaknesses of each recall strategy, thus affecting the effectiveness of information recommendation. Summary of the Invention

[0005] In view of the above-mentioned problems in the prior art, the purpose of the present invention is to provide an information recommendation method, apparatus, device and storage medium that can not only ensure the effectiveness of information recommendation, but also improve the diversity of recommended information.

[0006] To address the above problems, this invention provides an information recommendation method, comprising:

[0007] In response to an information recommendation request, multiple different recall strategies are used to determine the corresponding recall information set from a preset information database, wherein the recall information set includes at least one piece of information;

[0008] Obtain historical data on information recommendation, and determine the correction parameters for each of the recall strategies based on the historical data;

[0009] The actual recall tier corresponding to each recall strategy is determined based on the correction parameters of each recall strategy and the preset recall tier.

[0010] For each of the recall strategies, a set of candidate recommendation information is determined from the corresponding recall information set based on a preset ranking model and the actual recall tier.

[0011] A preset number of information items are determined from the candidate recommendation information set corresponding to each of the aforementioned recall strategies and recommended to the target user.

[0012] Another aspect of the present invention provides an information recommendation device, comprising:

[0013] The recall information set determination module is used to respond to an information recommendation request and determine the corresponding recall information set from a preset information database using multiple different recall strategies. The recall information set includes at least one piece of information.

[0014] The correction parameter determination module is used to acquire historical data of information recommendation and determine the correction parameters of each of the recall strategies based on the historical data.

[0015] The actual recall tier determination module is used to determine the actual recall tier corresponding to each of the recall strategies based on the correction parameters of each recall strategy and the preset recall tier.

[0016] The candidate recommendation information set determination module is used to determine the candidate recommendation information set from the corresponding recall information set for each recall strategy based on a preset ranking model and the actual recall tier.

[0017] The recommendation module is used to determine a preset number of information from the candidate recommendation information set corresponding to each of the recall strategies and recommend it to the target user.

[0018] In another aspect, the present invention provides an electronic device including a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the above-described information recommendation method.

[0019] In another aspect, the present invention provides a computer-readable storage medium storing at least one instruction or at least one program, wherein the at least one instruction or the at least one program is loaded and executed by a processor to implement the information recommendation method as described above.

[0020] The information recommendation method of this invention utilizes multiple recall strategies for information retrieval, determines the correction parameters for each recall strategy based on historical data from the entire network, and adjusts the preset recall tiers according to the correction parameters to obtain the actual recall tiers for each recall strategy. Finally, the recalled information is truncated according to its actual recall tier. This approach ensures both the effectiveness of information recommendation and enhances the diversity of recommended information, avoiding the underperformance of certain recall strategies due to biases in the ranking model. Furthermore, the amount of information retained by each recall strategy is determined by its performance in historical data; the better the recall strategy, the more information is retained, preventing the sacrifice of final recommendation effectiveness for the sake of diversity. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the present invention, the accompanying drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0022] Figure 1 This is a schematic diagram of the implementation environment provided in one embodiment of the present invention;

[0023] Figure 2 This is a flowchart of an information recommendation method provided in one embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram of the correction parameters of the recall strategy provided in one embodiment of the present invention;

[0025] Figure 4 This is a flowchart of an information recommendation method provided in another embodiment of the present invention;

[0026] Figure 5 This is a schematic diagram showing the distribution of recall strategies according to an embodiment of the present invention.

[0027] Figure 6 This is a schematic diagram of an information recommendation process provided in a specific embodiment of the present invention;

[0028] Figure 7 This is a schematic diagram of an interface for information recommendation provided in a specific embodiment of the present invention;

[0029] Figure 8 This is a schematic diagram of the structure of an information recommendation device provided in one embodiment of the present invention;

[0030] Figure 9 This is a schematic diagram of the structure of a server provided in one embodiment of the present invention. Detailed Implementation

[0031] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0033] To make the objectives, technical solutions, and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. First, the following concepts are explained in the embodiments of the present invention:

[0034] Recall: From a large number of candidate items, a recall algorithm is used to quickly reduce the set of candidate items to the thousands or even hundreds.

[0035] Recall Algorithms: The algorithms used in the recall process are called recall algorithms. Commonly used recall algorithms include profile recall, collaborative recall, and Deep Structured Semantic Model (DSSM) recall, etc.

[0036] Sorting: The sorting model is used to score and sort the recall results of the recall algorithm. There may be multiple sorting layers, such as coarse sorting and fine sorting (and possibly more sorting layers). After each sorting layer, the number of items in the next sorting layer will be truncated to reduce the number of items in the next sorting layer.

[0037] Please refer to the attached instruction manual. Figure 1 The illustration shows a schematic diagram of an implementation environment provided by an embodiment of the present invention, which may include a terminal 110 and a server 120. The terminal 110 and the server 120 may be directly or indirectly connected via wired or wireless communication, and this application does not impose any limitations on this connection.

[0038] The terminal 110 may be a smartphone, tablet computer, laptop computer, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal 110 may run one or more clients, which may be clients of the same type or clients of different types.

[0039] The server 120 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0040] In practical applications, the client can be a client for a news application. Users can browse information pages after opening, logging in, or clicking on the client. When a user browses an information page on the terminal 110, an information recommendation request is generated, and the terminal 110 sends this request to the server 120. The information recommendation request can be generated when the user browses the news application client through the terminal 110, or it can be generated periodically by the news application, for example, a request to recommend relevant information to the user every day at 3:00 PM.

[0041] The information recommendation device that implements the information recommendation method provided in this embodiment of the invention can be installed as a server program on the server 120. The server 120 will identify the information recommendation request sent by the terminal 110, and retrieve the information in the information database corresponding to the information recommendation request using the information recommendation method provided in this embodiment of the invention, generate an information recommendation list, and return the information recommendation list to the terminal 110 so that the terminal 110 can output the information recommendation list, thereby achieving the purpose of recommending information to the user of the terminal.

[0042] It should be noted that, Figure 1 This is just one example.

[0043] Please refer to the attached instruction manual. Figure 2 This illustrates the flow of an information recommendation method provided by an embodiment of the present invention, which can be applied to... Figure 1 The server in the middle. Specifically, such as... Figure 2 As shown, the method may include the following steps:

[0044] S210: In response to an information recommendation request, using multiple different recall strategies, determine the corresponding recall information set from a preset information database, wherein the recall information set includes at least one piece of information.

[0045] In this embodiment of the invention, a recommendation request is generated when a target user browses information on a news application client on the terminal; or, when a target user uses a news application on the terminal, a recommendation request is generated within a fixed time period; or, if an application on the terminal has a function to receive recommendation information, a recommendation request will be generated within a fixed time period even if the target user does not use or browse the application. After the news application generates a recommendation request, the terminal sends the recommendation request to the server so that the server can receive the recommendation request.

[0046] When the server receives the information recommendation request, it can determine the preset information database corresponding to the information recommendation request and use different recall strategies to retrieve the information, resulting in multiple recall queues. Each recall queue is a set of retrieved information. These multiple recall strategies can employ different recall algorithms, such as profile-based recall, collaborative recall, DSSM recall, etc., and this embodiment of the invention does not impose any limitations on this. Since different recall algorithms have different biases—for example, some recall algorithms prefer to recall popular information, while others prefer to recall information that the target user is interested in, and still others prefer to recall information that the target user's user group likes—using different recall strategies can improve the diversity of the retrieved information and satisfy the interests of target users from different perspectives.

[0047] S220: Obtain historical data on information recommendation, and determine the correction parameters for each of the recall strategies based on the historical data.

[0048] In this embodiment of the invention, the performance of each recall strategy can be determined based on the historical data of the information application. The truncation level for information in the corresponding recall information set is dynamically determined based on the effectiveness of each recall strategy. Specifically, information will be truncated when its predicted click-through rate (PCTR) obtained by the ranking model is lower than the threshold of that level. The correction parameters may include user click-through rate or click duration, and these parameters reflect the performance of the recall strategy.

[0049] In one possible embodiment, obtaining historical data for information recommendations and determining correction parameters for each of the recall strategies based on the historical data may include:

[0050] The information recommendation request is parsed to determine the target user;

[0051] Obtain the characteristic information of the target user, and determine the target user group to which the target user belongs based on the characteristic information;

[0052] Obtain a set of historical recall information for each of the aforementioned recall strategies targeting the target user group from the historical data;

[0053] For each of the recall strategies, the click data of the target user group on the information in the historical recall information set is determined as the correction parameter of the recall strategy.

[0054] In this embodiment of the invention, users of the information application can be pre-divided into multiple user groups. For example, users can be classified according to their gender and age group, first dividing them into males and females, and then further dividing male and female users into multiple different groups based on their age. Exemplarily, refer to the appendix to the specification. Figure 3 Women aged 12-15, 16-18, and 19-23 can be categorized as different user groups. Then, the gender and age of the target users can be obtained, and the target user group can be determined based on the gender and age.

[0055] In one possible embodiment, the user group can be further segmented based on gender, age, and other information such as the user's location. This embodiment of the invention does not limit this.

[0056] In this embodiment of the invention, the correction parameters for each recall strategy for the target user can be determined based on the performance of each recall strategy in the target user group. The click data may include the user's click-through rate or click duration, that is, the click-through rate or click duration of each recall strategy in the target user group can be used as its correction parameters.

[0057] In one possible embodiment, determining the click data of the target user group on the information in the historical recall information set for each of the recall strategies, as a correction parameter for the recall strategy, may include:

[0058] Obtain the information type of the information in the historical recall information set;

[0059] The information in the historical recall information set is classified according to the information type to obtain multiple sub-historical recall information sets;

[0060] The click data of the target user group on the information in each sub-historical recall information set is determined as the correction parameter of the recall strategy in each information type.

[0061] In this embodiment of the invention, click data for each recall strategy under different types of information within the target user group can be determined separately, and the click data can be used as correction parameters for the recall strategy in each information type. For example, refer to the appendix to the specification. Figure 3 The information types can include entertainment, sports, and animation, etc. Assuming the target user is a 22-year-old woman, the target user group can be determined to be women aged 19-23. Then, based on historical information recommendation data, the click-through rate (CTR) of each recall strategy under different information types within the 19-23-year-old female group is determined. Specifically, recall strategy 1 has a CTR of 8% under the entertainment category, recall strategy 2 has a CTR of 10% under the entertainment category, and recall strategy 3 has a CTR of 15% under the entertainment category. Therefore, the adjustment parameters for recall strategies 1, 2, and 3 under the entertainment category are 8%, 10%, and 15%, respectively.

[0062] In one possible embodiment, more detailed classifications can be made based on the various information types. For example, the information can be refined to secondary or tertiary categories based on its content. This embodiment of the invention does not limit this. For example, sports can be further divided into track and field, ball sports, and water sports, etc., and ball sports can be further divided into badminton, softball, basketball, and football, etc.

[0063] S230: Determine the actual recall level corresponding to each of the recall strategies based on the correction parameters of each recall strategy and the preset recall level.

[0064] In this embodiment of the invention, the correction parameters of the recall strategy may include the correction parameters of the recall strategy in multiple information types, and the actual recall tier corresponding to the recall strategy may also include the actual recall tier of the recall strategy in multiple information types; for different information types, the actual recall tier of each recall strategy in the information type can be determined based on a preset recall tier, so as to truncate information of different information types in the recall information set.

[0065] In one possible embodiment, determining the actual recall tier corresponding to each of the recall strategies based on the correction parameters of each recall strategy and the preset recall tier may include:

[0066] For each information type, calculate the average value of the correction parameters of each recall strategy for that information type;

[0067] The actual recall level of each recall strategy in the information type is determined based on the correction parameters of each recall strategy in the information type, the average value, and the preset recall level.

[0068] Specifically, the preset recall tiers can be set according to actual conditions, and the actual recall tiers of the recall strategy in each information type can be calculated using the following formula:

[0069]

[0070] In one example, suppose we use three different recall strategies, A, B, and C, to recall information from a pre-defined database. The total number of articles recalled is 1000. The expectation is that after truncation by a ranking model, only 100 articles will remain, representing the top 10% of articles. Assume that the click-through rates for entertainment-related articles in the target user group using recall strategies A, B, and C are 8%, 10%, and 15%, respectively. Therefore, the adjustment parameters for recall strategies A, B, and C are 8%, 10%, and 15%, respectively.

[0071] Assuming a total of 100 recall tiers, the truncated preset recall tier is 10 tiers (10 tiers / 100 tiers = 10%). The average correction parameter for the three recall strategies A, B, and C is 11%. The actual recall tiers for entertainment types under recall strategies A, B, and C are calculated as follows:

[0072] Recall Strategy A: Rounded to the nearest 7;

[0073] Recall Strategy B: Rounded to the nearest 9;

[0074] Recall Strategy C: Rounded to 14.

[0075] It should be noted that the formula for adjusting the recall level can also have different variations. For example, a Gaussian distribution can be used to calculate the actual recall level. This embodiment of the invention does not limit this.

[0076] This invention introduces the concept of tiers to measure the score of information. The tier distribution of each recall strategy can be predetermined. The preset recall tier is determined based on the amount of information remaining after the expected sorting model. The tier is then adjusted based on the correction parameters of each recall strategy to determine the final actual recall tier.

[0077] In one possible embodiment, the method may further include a step of determining the tier distribution corresponding to each of the recall strategies, wherein the tier distribution includes multiple tiers and corresponding tier thresholds. In practical applications, the corresponding tier distribution can be determined based on the recall records of each recall strategy, for classifications based on different information types for different user groups. Specifically, as... Figure 4 As shown, determining the distribution of tiers corresponding to each of the recall strategies may include:

[0078] S410: For each of the recall strategies, obtain the recall information record of the recall strategy, wherein the recall information record includes multiple pieces of information and the corresponding user group.

[0079] S420: Classify the information in the recall information record according to the user group corresponding to the information and the information type of the information to obtain multiple sub-recall information sets; wherein, at least one of the information type and user group corresponding to different sub-recall information sets is different.

[0080] Specifically, each category corresponds to an information type and a user group. The sub-recall information sets of different categories have different information types or correspond to different user groups, or have different information types and correspond to different user groups.

[0081] S430: The information in each of the sub-recall information sets is scored using the preset ranking model to obtain the predicted click pass rate of the information.

[0082] Specifically, the preset ranking model can be trained based on the actual information recommendation results pushed to users. That is, known information recommendation results can be used as samples, and a certain deep learning method can be used to train the model to obtain the preset ranking model. When using the preset ranking model to score the information in each of the sub-recall information sets, the information is input into the preset ranking model to output the predicted click-through rate of the information.

[0083] S440: Determine the distribution of the recall strategy's corresponding tiers under different categories based on the distribution of the predicted click-through rate of the information in each of the sub-recall information sets.

[0084] Specifically, the distribution of PCTR values ​​of all information in the sub-recall information set can be statistically analyzed. The distribution range of PCTR values ​​can be divided into N (N is a positive integer) sub-intervals from high to low frequency. Each sub-interval can be used as a level, and the minimum PCTR value in each sub-interval can be used as the level threshold of the level. This yields the level distribution under different categories, where at least one of the information types and user groups in the different categories is different.

[0085] In one example, refer to the attached reference manual. Figure 5 Users can be segmented based on gender and age. For each recall algorithm, the corresponding tier distribution can be determined according to different user groups and different information types, resulting in the tiers of recall algorithm X information type X gender X age and their corresponding tier thresholds. For example... Figure 5 As shown, taking a 100-level classification as an example, we can obtain the level thresholds (levels 1-100) for each level in each category of recall strategy 1, recall strategy 2, and recall strategy 3. Assuming that recall strategy 1 has 10,000 pieces of information in the sub-recall information set corresponding to the entertainment category among female users aged 19-23, we can arrange the PCTR values ​​of these 10,000 pieces of information from largest to smallest. The interval containing the top 100 PCTR values ​​is divided into level 1, with the level threshold being the 100th PCTR value; the interval containing PCTR values ​​from 101 to 200 is divided into level 2, with the level threshold being the 200th PCTR value; and so on. This allows us to obtain the levels of recall strategy 1 in the entertainment category among female users aged 19-23, from level 1 to 100, and their corresponding level thresholds.

[0086] It should be noted that different recall strategies may have the same total number of tiers in the same category, but the same recall strategy may have the same or different total number of tiers in different categories. This embodiment of the invention does not impose any restrictions on this.

[0087] In one possible embodiment, the PCTR distribution of different recall strategies under different information types can be determined based on the recall information records of the target user for different recall strategies, thereby determining the tier distribution of different recall strategies for the target user. However, it is necessary to ensure that the target user has sufficient behavioral records for the calculated results to have sufficient credibility.

[0088] S240: For each of the recall strategies, determine the candidate recommendation information set from the corresponding recall information set based on the preset ranking model and the actual recall tier.

[0089] In this embodiment of the invention, the preset ranking model can be used to score the information in each recall information set, and the information to be retained can be determined based on the score and information type of each information. The retained information is then added to the corresponding candidate recommendation information set.

[0090] In one possible embodiment, determining the candidate recommendation information set from the corresponding recall information set based on a preset ranking model and the actual recall tier for each of the recall strategies may include:

[0091] For each piece of information in the recall information set, determine the target information type of that information;

[0092] Obtain the actual recall tier of the recall strategy within the target information type;

[0093] The information is scored using a preset ranking model to obtain the predicted click-through rate of the information;

[0094] When the predicted click-through rate and the actual recall tier satisfy a preset relationship, the information is added to the candidate recommendation information set corresponding to the recall strategy.

[0095] Specifically, when the predicted click-through rate and the actual recall tier satisfy a preset relationship, adding the information to the candidate recommendation information set corresponding to the recall strategy may include:

[0096] Obtain the target user group to which the target user belongs;

[0097] The target gear distribution is determined based on the target user group and the target information type. The target gear distribution includes multiple gears and their corresponding gear thresholds.

[0098] The gear threshold corresponding to the actual recall gear is determined based on the target gear distribution.

[0099] Compare the predicted click pass rate with the tier threshold;

[0100] When the predicted click pass rate is greater than or equal to the threshold, the information is added to the candidate recommendation information set.

[0101] In one example, suppose an entertainment article is recalled using recall strategy A. The PCTR of the article can be obtained by scoring it using the preset ranking model. Also suppose that the actual recall level of recall strategy A in the entertainment category of the target user group is level 7. Then, the article will only be retained if its PCTR is greater than or equal to the level 7 threshold, that is, the article can be added to the candidate recommendation information set corresponding to recall strategy A; if the article's PCTR is less than the level 7 threshold, the article will be truncated.

[0102] In one possible embodiment, for the information in the recall information set of a certain recall strategy, the target tier of the information can be determined based on the distribution of the target tier and the predicted click pass rate of the information. The target tier is compared with the actual recall tier. When the target tier is not lower than the actual recall tier, the information can be added to the candidate recommendation information set corresponding to the recall strategy. When the target tier is lower than the actual recall tier, the information will be truncated.

[0103] S250: Determine a preset number of information from the candidate recommendation information set corresponding to each of the recall strategies and recommend it to the target user.

[0104] In this embodiment of the invention, several pieces of information that meet the requirements can be selected as target recommendation information from the candidate recommendation information sets corresponding to each of the recall strategies using different methods. For example, the information with the highest PCTR value can be selected as the target recommendation information from the candidate recommendation information sets corresponding to each of the recall strategies. After determining the target recommendation information, a recommendation information list can be generated based on preset rules, and the recommendation information list can be sent to the terminal corresponding to the information recommendation request. After receiving the recommendation information list, the terminal can display each piece of information in the recommendation information list to the target user.

[0105] The following describes the information recommendation method provided in this embodiment of the invention using a specific application scenario as an example. The information recommendation method provided in this embodiment of the invention can be applied to various information flow recommendation scenarios, such as the image and text recommendation scenario of a certain application. In this scenario, the method can be further described in conjunction with the appendix to the specification. Figure 6After receiving an information recommendation request from a target user sent by a terminal, the application server can use the information recommendation method provided in this embodiment of the invention to determine the recall queues recalled by different recall algorithms. By penetrating the different recall queues in the recommendation process, the truncated queues are returned to the image and text gateway. The image and text gateway performs diversity selection (i.e., each recall queue selects some articles) to determine the articles finally recommended to the user. The application server can then send the articles finally recommended to the user to the terminal. After receiving the articles, the terminal can display them to the target user in a designated user interface. For example, as shown... Figure 7 As shown, the terminal can display the received articles on the main interface of the application. These articles can include articles from different dimensions, such as current articles, articles of long-term user interest, and articles from accounts followed by the user, etc.

[0106] The information recommendation method provided in this invention can improve the user's reading experience in image and text recommendation scenarios, especially for new users. Because new users have limited activity, their profiles are less reliable, and ranking models trained on new user behavior data are prone to bias, leading to convergent high-scoring content (e.g., recommending articles related to variety shows and TV dramas to female users). By penetrating various recall strategies during the recommendation process, the inherent bias of the ranking model can be avoided, resulting in more diverse articles recommended to users. The method significantly improves the diversity of image and text information recommended to users, and also increases the number of daily active users.

[0107] The following provides an example illustrating the recommendation effect of the information recommendation method provided in this embodiment of the invention. This embodiment of the invention can be tested online. The control group uses a recommendation method that aggregates information retrieved by multiple recall strategies and sorts it using a ranking model. The experimental group uses the recommendation method provided in this embodiment of the invention that truncates information retrieved by each recall strategy based on the corrected actual recall tier. Statistical analysis of the recommended text and image information data and user click data reveals that the experimental group's data on the diversity of clicks in the secondary categories of main information stream text and image content, the diversity of clicks in the primary categories of main information stream text and image content, and the diversity of exposures in the secondary categories of main information stream text and image content are significantly higher than those in the control group. The experimental group also shows significantly higher daily active users, number of clicks in the information stream, and cumulative usage time compared to the control group, indicating a significant improvement in recommendation effectiveness.

[0108] In summary, the information recommendation method of this invention utilizes multiple recall strategies for information retrieval, determines the correction parameters for each recall strategy based on historical data from the entire network, and adjusts the preset recall tiers according to the correction parameters to obtain the actual recall tiers for each recall strategy. Finally, the recalled information is truncated according to its actual recall tier. This approach ensures both the effectiveness of information recommendation and enhances the diversity of recommended information, avoiding the underperformance of certain recall strategies due to the bias of the ranking model itself. Furthermore, the amount of information retained by each recall strategy is determined by its performance in historical data; the better the recall strategy, the more information is retained, thus avoiding sacrificing the final recommendation effect for the sake of diversity.

[0109] Reference manual attached Figure 8 This illustrates the structure of an information recommendation device provided in one embodiment of the present invention. For example... Figure 8 As shown, the device may include:

[0110] The recall information set determination module 810 is used to respond to an information recommendation request and determine the corresponding recall information set from a preset information database using multiple different recall strategies. The recall information set includes at least one piece of information.

[0111] The correction parameter determination module 820 is used to acquire historical data of information recommendation and determine the correction parameters of each of the recall strategies based on the historical data.

[0112] The actual recall tier determination module 830 is used to determine the actual recall tier corresponding to each of the recall strategies based on the correction parameters of each of the recall strategies and the preset recall tiers.

[0113] The candidate recommendation information set determination module 840 is used to determine the candidate recommendation information set from the corresponding recall information set for each recall strategy based on a preset ranking model and the actual recall tier.

[0114] The recommendation module 850 is used to determine a preset number of information from the candidate recommendation information set corresponding to each of the recall strategies and recommend it to the target user.

[0115] In one possible embodiment, the device may further include a gear distribution determination module, used to determine the gear distribution corresponding to each of the recall strategies; the gear distribution determination module may include:

[0116] The recall information record acquisition unit is used to acquire the recall information record of each recall strategy, wherein the recall information record includes multiple pieces of information and the corresponding user group;

[0117] A classification unit is used to classify the information in the recall information record according to the user group corresponding to the information and the information type of the information, so as to obtain multiple sub-recall information sets; wherein, at least one of the information type and user group corresponding to different sub-recall information sets is different.

[0118] The scoring unit is used to score the information in each of the sub-recall information sets using the preset sorting model, and to obtain the predicted click pass rate of the information.

[0119] The tier distribution determination unit is used to determine the tier distribution of the recall strategy under different categories based on the distribution of the predicted click pass rate of the information in each of the sub-recall information sets.

[0120] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus provided in the above embodiments and the corresponding method embodiments belong to the same concept, and the specific implementation process can be found in the corresponding method embodiments, which will not be repeated here.

[0121] One embodiment of the present invention also provides an electronic device, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the information recommendation method provided in the above method embodiments.

[0122] Memory can be used to store software programs and modules. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory. Memory can primarily include a program storage area and a data storage area. The program storage area can store the operating system, application programs required for the functions, etc.; the data storage area can store data created based on the use of the device, etc. Furthermore, memory can include high-speed random access memory, and can also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, memory can also include a memory controller to provide the processor with access to the memory.

[0123] The method embodiments provided in this invention can be executed in a terminal, server, or similar computing device; that is, the aforementioned electronic device may include a terminal, server, or similar computing device. Taking running on a server as an example, such as... Figure 9The diagram illustrates the structural schematic of a server for the runtime information recommendation method provided in this embodiment of the invention. The server 900 can vary significantly due to different configurations or performance characteristics, and may include one or more Central Processing Units (CPUs) 910 (e.g., one or more processors) and a memory 930, and one or more storage media 920 (e.g., one or more mass storage devices) for storing application programs 923 or data 922. The memory 930 and storage media 920 can be temporary or persistent storage. The program stored in the storage media 920 may include one or more modules, each module including a series of instruction operations on the server. Furthermore, the CPU 910 may be configured to communicate with the storage media 920 and execute the series of instruction operations in the storage media 920 on the server 900. Server 900 may also include one or more power supplies 960, one or more wired or wireless network interfaces 950, one or more input / output interfaces 940, and / or one or more operating systems 921, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0124] The input / output interface 940 can be used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of server 900. In one example, the input / output interface 940 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In one example, the input / output interface 940 can be a radio frequency (RF) module for wireless communication with the Internet. This wireless communication can use any communication standard or protocol, including but not limited to Global System for Mobile Communications (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, and Short Messaging Service (SMS).

[0125] Those skilled in the art will understand that Figure 9 The structure shown is for illustrative purposes only; the server 900 may also include more advanced components. Figure 9 The more or fewer components shown, or having the same Figure 9 The different configurations shown.

[0126] One embodiment of the present invention also provides a computer-readable storage medium, which can be disposed in an electronic device to store at least one instruction or at least one program related to implementing an information recommendation method, wherein the at least one instruction or the at least one program is loaded and executed by the processor to implement the information recommendation method provided in the above-described method embodiment.

[0127] Optionally, in embodiments of the present invention, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0128] One embodiment of the present invention also provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the information recommendation method provided in the various optional implementations described above.

[0129] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0130] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0131] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0132] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An information recommendation method characterized by comprising: The method comprises the following steps: in response to an information recommendation request, determining a respective recall information set corresponding to each of a plurality of different recall strategies from a preset information base, the recall information set comprising at least one information; obtaining historical data of information recommendation, and determining a correction parameter of each of the recall strategies based on the historical data; determining an actual recall gear corresponding to each of the recall strategies according to the correction parameter of each of the recall strategies and a preset recall gear; the preset recall gear is determined based on an expected amount of information remaining after being truncated by a preset sorting model; for each information in the recall information set, determining a target information type of the information; obtaining an actual recall gear of the recall strategy in the target information type; scoring the information by using a preset sorting model to obtain a predicted click-through rate of the information; when the predicted click-through rate and the actual recall gear satisfy a preset relationship, adding the information to a candidate recommended information set corresponding to the recall strategy; the preset relationship refers to that the predicted click-through rate is greater than or equal to a gear threshold corresponding to the actual recall gear; determining a preset number of information from the candidate recommended information set corresponding to each of the recall strategies and recommending the information to a target user.

2. The method of claim 1, wherein, The method comprises the following steps: analyzing the information recommendation request to determine a target user; obtaining feature information of the target user, and determining a target user group in which the target user is located according to the feature information; obtaining a historical recall information set of each of the recall strategies for the target user group from the historical data; for each of the recall strategies, determining click data of the target user group on information in the historical recall information set as a correction parameter of the recall strategy.

3. The method of claim 2, wherein, The method comprises the following steps: obtaining an information type of information in the historical recall information set; classifying the information in the historical recall information set according to the information type to obtain a plurality of sub historical recall information sets; determining click data of the target user group on information in each of the sub historical recall information sets as a correction parameter of the recall strategy in each information type.

4. The method according to claim 2 or 3, characterized in that, The correction parameter of the recall strategy comprises correction parameters of the recall strategy in a plurality of information types, and the actual recall gear corresponding to the recall strategy comprises actual recall gears of the recall strategy in a plurality of information types; The method comprises the following steps: for each information type, calculating an average value of the correction parameter of each of the recall strategies in the information type; determining an actual recall gear of each of the recall strategies in the information type according to the correction parameter of each of the recall strategies in the information type, the average value, and the preset recall gear.

5. The method of claim 1, wherein, The information is added to the candidate recommended information set corresponding to the recall strategy when the predicted click-through rate and the actual recall level satisfy a preset relationship. Obtain a target user group in which the target user is located; Determine a target level distribution according to the target user group and the target information type, the target level distribution including multiple levels and corresponding level thresholds; Determine a level threshold corresponding to the actual recall level according to the target level distribution; Compare the predicted click-through rate with the level threshold; When the predicted click-through rate is greater than or equal to the level threshold, the information is added to the candidate recommended information set.

6. The method according to claim 1 or 5, characterized in that, The method further includes determining a level distribution corresponding to each recall strategy; the determination of the level distribution corresponding to each recall strategy includes: For each recall strategy, obtain a recall information record of the recall strategy, the recall information record including multiple information and corresponding user groups; Classify the information in the recall information record according to the user groups corresponding to the information and the information types of the information, to obtain multiple sub-recall information sets; different sub-recall information sets correspond to at least one different information type and user group; Score the information in each sub-recall information set using the preset ranking model to obtain a predicted click-through rate of the information; Determine a level distribution corresponding to the recall strategy under different classifications according to the distribution of the predicted click-through rates of the information in each sub-recall information set.

7. An information recommendation device characterized by comprising: It includes: A recall information set determination module for determining, in response to an information recommendation request, a recall information set corresponding to each of multiple different recall strategies from a preset information database, the recall information set including at least one information; A correction parameter determination module for obtaining historical data of information recommendation and determining correction parameters of each recall strategy based on the historical data; An actual recall level determination module for determining an actual recall level corresponding to each recall strategy according to correction parameters of each recall strategy and a preset recall level; The preset recall level is determined based on an expected amount of information remaining after being truncated by a preset ranking model; A candidate recommended information set determination module for determining, for each information in the recall information set, a target information type of the information; Obtain an actual recall level of the recall strategy in the target information type; Score the information using a preset ranking model to obtain a predicted click-through rate of the information; when the predicted click-through rate and the actual recall level satisfy a preset relationship, the information is added to a candidate recommended information set corresponding to the recall strategy; The preset relationship means that the predicted click-through rate is greater than or equal to a level threshold corresponding to the actual recall level; A recommendation module for determining a preset number of information from each candidate recommended information set corresponding to the recall strategy and recommending the information to a target user.

8. The apparatus of claim 7, wherein, The device further comprises a gear distribution condition determining module for determining gear distribution conditions corresponding to each of the recall strategies; the gear distribution condition determining module comprises: a recall information record obtaining unit for obtaining, for each of the recall strategies, a recall information record of the recall strategy, the recall information record comprising a plurality of information and a user group corresponding thereto; a classifying unit for classifying the information in the recall information record according to the user group corresponding to the information and the information type of the information, to obtain a plurality of sub-recall information sets; wherein different sub-recall information sets differ in at least one of the information type and the user group corresponding thereto; a scoring unit for scoring the information in each of the sub-recall information sets by using the preset ranking model, to obtain a predicted click-through rate of the information; a gear distribution condition determining unit for determining the gear distribution conditions corresponding to the recall strategy under different classifications according to the distribution of the predicted click-through rate of the information in each of the sub-recall information sets.

9. An electronic device, comprising: The electronic device comprises a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the information recommendation method according to any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the information recommendation method according to any one of claims 1-6.

Images