Methods, devices, storage media, and electronic devices for recalling media information

By combining feature similarity and publication results to adjust the recall ratio during the media information recall process, the problem of low recall accuracy was solved, effective exposure of high-quality advertisements was achieved, and recall accuracy was improved.

CN116485469BActive Publication Date: 2026-11-17TENCENT TECHNOLOGY (SHENZHEN) CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202210044874.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-14
Publication Date
2026-11-17
Estimated Expiration
2042-01-14

AI Technical Summary

Technical Problem

In existing technologies, the accuracy of media information recall is low, making it difficult for high-quality advertisements to be exposed, thus creating a vicious cycle.

Method used

By obtaining recall requests associated with the target object, the first media information is recalled from N media information using information features, and supplementary recall is carried out according to the number of remaining media information. The recall ratio is adjusted by combining feature similarity and release results to achieve accurate recall.

Benefits of technology

It improves the accuracy of media information recall, reduces the probability of high-quality media information being missed, and ensures the effectiveness of advertising exposure.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116485469B_ABST
    Figure CN116485469B_ABST
Patent Text Reader

Abstract

This application discloses a method, apparatus, storage medium, and electronic device for recalling media information. The method includes: obtaining a first target recall request associated with a target object, wherein the first target recall request is used to request the recall of target media information from N media information, where N is a natural number greater than or equal to 2; responding to the first target recall request, obtaining information features corresponding to each of the N media information, wherein the information features are content description features of the media information; recalling first media information from the N media information based on the information features; and recalling second media information from M media information based on the number of remaining media information among the N media information after the first media information has been recalled. This application solves the technical problem of low accuracy in media information recall.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of computers, and more specifically, to a method, apparatus, storage medium, and electronic device for recalling media information. Background Technology

[0002] In recent years, businesses have placed increasing emphasis on the quality of media information (such as advertising). To ensure the quality of advertising, businesses need to find the right audience for their ads, that is, to recall ads that users might be interested in when a user makes a request.

[0003] For ad recall, related technologies often utilize feature similarity for ad recall. However, the recall accuracy of this method is difficult to guarantee, and it is easy to fall into the trap of local optima, preventing some high-quality ads from getting recalled and exposed, creating a vicious cycle. Therefore, there is a problem of low recall accuracy for media information.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides a method, apparatus, storage medium, and electronic device for recalling media information, in order to at least solve the technical problem of low accuracy in recalling media information.

[0006] According to one aspect of the embodiments of this application, a method for recalling media information is provided, comprising: obtaining a first target recall request associated with a target object, wherein the first target recall request is used to request the recall of target media information from N media information, where N is a natural number greater than or equal to 2; responding to the first target recall request, obtaining information features corresponding to each of the N media information, wherein the information features are content description features of the media information; recalling the first media information from the N media information based on the information features, wherein the similarity between the information features of the first media information and the target features corresponding to the target object reaches a target threshold, and the target media information includes the first media information; recalling the second media information from M media information based on the number of remaining media information in the N media information after the first media information is recalled, wherein the M media information are the remaining media information in the N media information after the first media information is recalled, where M is a natural number less than N, and the target media information includes the second media information.

[0007] According to another aspect of the embodiments of this application, a media information recall device is also provided, comprising: a first acquisition unit, configured to acquire a first target recall request associated with a target object, wherein the first target recall request is used to request the recall of target media information from N media information, where N is a natural number greater than or equal to 2; a second acquisition unit, configured to respond to the first target recall request and acquire information features corresponding to each of the N media information, wherein the information features are content description features of the media information; and a first recall unit, configured to recall the first media information from the N media information based on the information features. Wherein, the similarity between the information features of the first media information and the target features corresponding to the target object reaches a target threshold, and the target media information includes the first media information; the second recall unit is used to recall the second media information from M media information based on the number of media information remaining in the N media information after the first media information is recalled, wherein the M media information are the media information remaining in the N media information after the first media information is recalled, M is a natural number less than N, and the target media information includes the second media information.

[0008] As an optional solution, the second recall unit includes: a recall module, used to recall a first target quantity of the second media information from M media information, wherein the ratio between the first target quantity and M is the first target ratio.

[0009] As an optional solution, it includes: a first acquisition module, used to acquire, after recalling the second media information from M media information based on the remaining number of media information from N media information after the first media information is recalled, a first release result after the first media information is recalled, and a second release result after the second media information is recalled; and an adjustment module, used to adjust the first target ratio based on the first release result and / or the second release result after recalling the second media information from M media information based on the remaining number of media information from N media information after the first media information is recalled, to obtain a second target ratio.

[0010] As an optional solution, it includes: a second acquisition module, used to acquire a second target recall request associated with the target object after adjusting the first target ratio according to the first release result and the second release result to obtain a second target ratio, wherein the target recall request is used to request the recall of the target media information from P media information, where P is a natural number greater than or equal to 2; and a third acquisition module, used to respond to the second target recall request after adjusting the first target ratio according to the first release result and the second release result to obtain a second target ratio, and acquire the information features corresponding to each of the P media information. The first recall module is used to recall the first media information from the P media information after adjusting the first target ratio according to the first release result and the second release result to obtain the second target ratio; the second recall module is used to recall the second target number of the second media information from the K media information after adjusting the first target ratio according to the first release result and the second release result to obtain the second target ratio, wherein the ratio between the second target number and K is the second target ratio, the K media information is the media information other than the first target media information from the P media information, and K is a natural number less than P.

[0011] As an optional solution, the aforementioned adjustment module includes: a first acquisition submodule, configured to acquire a first release indicator based on the first release result and the second release result, wherein the first release indicator is a release effect indicator of the second media information relative to the first media information; and adjust the first target ratio using the first release indicator to obtain the second target ratio; or, a second acquisition submodule, configured to acquire a second release indicator of the second media information based on the second release result, wherein the second release indicator is a release effect indicator of the second media information; and adjust the first target ratio using the second release indicator to obtain the second target ratio.

[0012] As an optional solution, the first acquisition submodule includes: an adjustment subunit, used to adjust the first target ratio using the target attenuation coefficient and the first release indicator to obtain the second target ratio, wherein the target attenuation coefficient is an attenuation coefficient calculated based on time data.

[0013] As an optional solution, it includes: a publishing unit, used to publish the first media information and the second media information at the target publishing location associated with the target object after recalling the second media information from the remaining media information among the N media information after the first media information has been recalled.

[0014] As an optional solution, it includes: a first input unit, configured to, after obtaining the first target recall request associated with the target object, respond to the first target recall request by inputting the N media information into the first tower structure of the target dual-tower model, wherein the target dual-tower model includes the first tower structure, the second tower structure, and a target processing structure; the first tower structure is used to extract information features corresponding to each media information; the second tower structure is used to extract features corresponding to each object; the features corresponding to each object include the target object; the information features corresponding to each media information include target information features; and the target processing structure is used to process the features corresponding to each object and the media information. The processing unit is used to compare the feature similarity between the corresponding information features after obtaining the first target recall request associated with the target object, and after obtaining the information features corresponding to each of the N media information output by the first tower structure, to compare the feature similarity between the information features corresponding to each of the N media information and the target feature using the target processing structure, so as to obtain the target information feature whose similarity with the target feature reaches the target threshold; the third recall unit is used to recall the first media information corresponding to the target information feature after obtaining the first target recall request associated with the target object.

[0015] As an optional solution, it includes: a third acquisition unit, configured to acquire object reference information and context information associated with the target object before acquiring the first target recall request associated with the target object; and a second input unit, configured to input the object reference information and context information into the second tower structure before acquiring the first target recall request associated with the target object, to obtain the target feature output by the second tower structure.

[0016] According to another aspect of the embodiments of this application, a computer program product or computer program is provided, the 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 the processor executes the computer instructions, causing the computer device to perform the media information recall method as described above.

[0017] According to another aspect of the embodiments of this application, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the media information recall method described above through the computer program.

[0018] In this embodiment, a first target recall request associated with a target object is obtained, wherein the first target recall request is used to request the recall of target media information from N media information, where N is a natural number greater than or equal to 2; in response to the first target recall request, information features corresponding to each of the N media information are obtained, wherein the information features are content description features of the media information; the first media information is recalled from the N media information based on the information features, wherein the similarity between the information features of the first media information and the target features corresponding to the target object reaches a target threshold, and the target media information includes the first media information; the number of remaining media information in the N media information after the first media information is recalled is used to recall the first media information. The remaining media information from the last N media information is used to recall the second media information from M media information. Here, the M media information refers to the remaining media information from the last N media information after the first media information is recalled, and M is a natural number less than N. The target media information includes the second media information. Based on the recall of media information using feature similarity, the remaining media information is also recalled. In this way, even if some high-quality media information is missed by the advertising recall method using feature similarity, the missed media information can still be recalled by this supplementary recall method, thereby reducing the probability of high-quality media information being missed. This achieves the technical effect of improving the recall accuracy of media information and solves the technical problem of low recall accuracy of media information. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0020] Figure 1 This is a schematic diagram illustrating the application environment of an optional media information recall method according to an embodiment of this application;

[0021] Figure 2 This is a schematic diagram of the flow of an optional media information recall method according to an embodiment of this application;

[0022] Figure 3 This is a schematic diagram of an optional media information recall method according to an embodiment of this application;

[0023] Figure 4 This is a schematic diagram of another optional media information recall method according to an embodiment of this application;

[0024] Figure 5This is a schematic diagram of another optional media information recall method according to an embodiment of this application;

[0025] Figure 6 This is a schematic diagram of another optional media information recall method according to an embodiment of this application;

[0026] Figure 7 This is a schematic diagram of another optional media information recall method according to an embodiment of this application;

[0027] Figure 8 This is a schematic diagram of another optional media information recall method according to an embodiment of this application;

[0028] Figure 9 This is a schematic diagram of another optional media information recall method according to an embodiment of this application;

[0029] Figure 10 This is a schematic diagram of an optional media information recall device according to an embodiment of this application;

[0030] Figure 11 This is a schematic diagram of the structure of an optional electronic device according to an embodiment of this application. Detailed Implementation

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

[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application 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 this application 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 non-exclusive inclusion; for example, a process, method, system, product, or apparatus 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 apparatus.

[0033] For ease of understanding, the following explanations are provided for some key terms and technical terms:

[0034] Pay-per-click advertising: Online advertising that users can independently place and manage, ranked by ad prices, and paid according to ad performance.

[0035] Cost to achieve: |Actual cost of ad - Ad bid| / Ad bid ≤ Specified threshold α (α < 1).

[0036] Achievement rate (based on cost): Ad spend for achieving the cost / Total ad spend * 100%.

[0037] Achievement rate (based on the number of ads): Number of ads that achieve the cost of advertising / Total number of ads * 100%.

[0038] Waste ads: Ads that consume ≥ N conversion bids and have < M1 conversions (N and M1 are positive integers).

[0039] Scaling up: The first day of exposure is within the calculation period, the conversion and return time is the calculation period plus a delay of t days, and the cumulative number of conversions reaches M2 (t is a non-negative integer, M2 is a positive integer).

[0040] Scaling rate (based on spending): Scaling rate of ad spending / total ad spending * 100%.

[0041] Scaling rate (based on the number of ads): Number of ads that initiated traffic / Total number of ads * 100%.

[0042] Prediction deviation: |Predicted value - Actual value| / Actual value * 100%.

[0043] Consumption: Actual charges deducted.

[0044] Conversion count: such as activation count, download count, and valid leads.

[0045] Actual cost of advertising (Cpa): Cost per conversion.

[0046] Target CPA: The target CPA user on the advertising platform.

[0047] Click-through rate (CTR): This can be, but is not limited to, the ratio of the number of times a piece of content on a website page is clicked to the number of times it is displayed. It can reflect, but is not limited to, the level of attention a piece of content on a webpage receives and is used to measure the attractiveness of an advertisement.

[0048] Conversion rate: This can be, but is not limited to, the percentage or fraction of a particular outcome. In advertising scenarios, optionally, merchants can set a conversion goal, such as visitors completing a target task (registering, placing an order, making a payment, etc.) on the website. The number of times a visitor completes the target task is the number of times the conversion goal is achieved. The conversion rate can also be, but is not limited to, the ratio of the number of times the conversion goal is achieved to the total number of clicks on the information within a statistical period. The calculation formula can be, but is not limited to: Conversion rate = (Number of conversions / Number of clicks) * 100%. Optionally, the conversion rate can be, but is not limited to, the core factor in whether a website can ultimately be profitable; improving the website's conversion rate is a result of the website's overall operational strength.

[0049] Original targeting: The combination of targeting conditions set by advertisers themselves, in addition to using smart targeting features (including automatic expansion and system optimization).

[0050] Broad targeting and precise targeting: Broad targeting and precise targeting are the opposites. How are they specifically defined? The straightforward idea is to judge based on targeting conditions. However, different targeting conditions have many sub-labels and even complex logical expressions, so it is also necessary to look at the actual number of users covered. However, different users have different activity levels, and different traffic sources will also vary greatly, so precise calculation is not feasible.

[0051] Unbreakable targeting: also known as non-ignorable targeting, refers to targeting conditions that advertisers set themselves and must meet under any circumstances.

[0052] Automatic scaling: The system intelligently scales up the advertising volume based on the advertiser's manual selection of precise narrow targeting.

[0053] System optimization: Based on the advertiser's manual selection of broad targeting, the system performs intelligent volume adjustment.

[0054] LookAlike: Based on a small number of high-quality customer groups (existing customers) provided by the advertiser, the system automatically expands to a larger similar customer group (potential new customers).

[0055] Ad concentration problem: also known as the top ad problem, mainly manifests in the fact that different users tend to retrieve the same small set of ads, resulting in other ads not having an exposure opportunity.

[0056] According to one aspect of the embodiments of this application, a method for recalling media information is provided. Optionally, as an optional implementation, the above-mentioned method for recalling media information may be applied to, but is not limited to, [examples of other methods]. Figure 1 The environment shown may include, but is not limited to, user equipment 102, network 110, and server 112. The user equipment 102 may include, but is not limited to, a display 108, a processor 106, and a memory 104.

[0057] The specific process can be summarized in the following steps:

[0058] Step S102, user equipment 102 obtains the first target recall request associated with target object 1022;

[0059] In steps S104-S106, user equipment 102 sends the first target recall request to server 112 through network 110;

[0060] In step S108, server 112 searches for the N media information requested to be recalled in the first target recall request through database 114, and processes the N media information through processing engine 116, recalling the first media information from the N media information according to information characteristics, and supplementing the recall of the second media information from the remaining media information.

[0061] In steps S110-S112, server 112 sends the first media information and the second media information to user equipment 102 via network 110. The processor 106 in user equipment 102 displays the first media information and the second media information on display 108 and stores the first media information and the second media information in memory 104.

[0062] remove Figure 1 Beyond the illustrated example, the above steps can be performed independently by user equipment 102, i.e., user equipment 102 can perform steps such as recalling the first media information and the second media information, thereby reducing the processing load on the server. User equipment 102 includes, but is not limited to, handheld devices (such as mobile phones), laptops, desktop computers, in-vehicle devices, etc., and this application does not limit the specific implementation of user equipment 102.

[0063] Alternatively, as an optional implementation, such as Figure 2 As shown, methods for recalling media information include:

[0064] S202, Obtain the first target recall request associated with the target object, wherein the first target recall request is used to request the recall of target media information from N media information, where N is a natural number greater than or equal to 2;

[0065] S204, in response to the first target recall request, obtain the information features corresponding to each of the N media information, wherein the information features are the content description features of the media information;

[0066] S206, Recall the first media information from N media information based on information features, wherein the similarity between the information features of the first media information and the target features corresponding to the target object reaches a target threshold, and the target media information includes the first media information;

[0067] S208, based on the number of media information remaining among N media information after the first media information is recalled, recall the second media information from M media information, where M media information are the media information remaining among N media information after the first media information is recalled, M is a natural number less than N, and the target media information includes the second media information.

[0068] Optionally, in this embodiment, the media information recall method described above can be applied to, but is not limited to, advertising scenarios. For example, when a request (the first target recall request associated with the target object) arrives, useful information (of the target object) is extracted from the request, such as user information and context information; this information is processed into target features to describe the request; the similarity between the target features and the information features of the advertisements to be recalled (N media information items) is further calculated; advertisements with a similarity greater than or equal to a threshold can be used as recalled advertisements, while advertisements that are not recalled are randomly added to the overall recall queue.

[0069] Optionally, in this embodiment, the target object may be, but is not limited to, the location object of the media information to be published, and / or the audience user account of the recalled media information; and the first target recall request associated with the target object may be, but is not limited to, a request to recall the media information to be published in the above-mentioned location object, and / or to be targeted at the audience user account.

[0070] Optionally, in this embodiment, recalling the second media information from M media information based on the number of remaining media information among N media information after the first media information is recalled can be, but is not limited to, recalling the second media information from M media information based on the recall result associated with the target object. The recall result can be, but is not limited to, the publication result after the media information is recalled, or can be understood as the feedback information collected after the media information is published, such as the click-through rate, exposure conversion rate, click conversion rate, number of conversions, cost deviation (cost / GMV, number of conversions * actual bid associated with the media information), cumulative viewing time, and resource revenue associated with the media information within a certain period after the media information is published.

[0071] Optionally, in this embodiment, recalling the second media information from M media information based on the number of remaining media information among N media information after the first media information is recalled can be, but is not limited to, recalling the second media information from M media information based on the number of remaining media information among N media information after the first media information is recalled, according to the target data threshold matched by the recall result associated with the target object. The target data threshold can be, but is not limited to, a range data threshold, and each range data threshold corresponds to a recall reference information. Then, the matched target range data threshold is obtained first based on the recall result associated with the target object, and the target recall reference information corresponding to the target range data threshold is obtained. The target recall reference information is then used to recall the second media information from M media information based on the number of remaining media information among N media information after the first media information is recalled, thereby achieving accurate supplementary recall.

[0072] Optionally, in this embodiment, obtaining the information features corresponding to each of the N media information can be, but is not limited to, using the ANN (Artificial Neural Network) triggering branch method. The ANN triggering branch method can be, but is not limited to, understanding as building an artificial neural network model and using this model to observe from historical data which objects may be high-conversion groups and have relatively high advertising efficiency.

[0073] Further examples, such as Figure 3 As shown, historical data 302 is obtained (such as data on users who have consumed specific accounts, media resource delivery data corresponding to converted user accounts, characteristic data of user accounts and media resources, and a small amount of targeted data such as gender, age, and geographic attributes); this historical information 302 is then input into model structure 304, where model structure 304 can be, but is not limited to, a targeted model automatically constructed by machine learning algorithms; further, the output of model structure 304 is obtained, and based on this output, the confidence score 306 of each user account is obtained, such as the execution score of user account 308-1 being 0.56, user account 308-2 being 0.81, and user account 308-3 being 0.79. In addition, user accounts with a confidence score 306 higher than the confidence score threshold (such as user account 308-2 and user account 308-3) can also be used as targets for media resource delivery.

[0074] Optionally, in this embodiment, the aforementioned artificial neural network model can be abstracted as a probability estimation problem like P(1|user, context, ad), which is the probability that a user account will see a media resource in a certain context under a predefined goal. Naturally, the higher the probability, the better—this is easy to understand. So how is this goal defined, and what is this predefined goal? This goal should have the function of measuring the degree of matching between the user account and the media resource. Typically, it can be the click-through rate (CTR), click-to-conversion rate (CPC), or impression-to-conversion rate (PCP). Alternatively, multiple different predefined goals can be modeled separately, constructing different ANN triggering branches, and then multiple branches can work together. It can be seen that the ANN triggering branch strategy essentially involves constructing classic models such as CTR, CPC, and PCP models, and then using these models to determine whether the user account matches the media resource. It's easy to understand that the higher the probability of a user clicking on the ad, the more likely the user account is to match the media resource; the higher the probability of a user converting from the media resource, the more likely the user account is to match the media resource.

[0075] To further illustrate the process of collecting various features (such as object features, context features, advertising features, etc.) and using these features to construct corresponding target prediction models, such as... Figure 4 As shown, the target prediction model typically has a dual-tower structure: one tower (target prediction model 402) is formed from object features and context features, and the other tower is formed from resource features. The dual-tower structure is the integrated target prediction model 404. In other words, contextual information only appears on the user side and not on the resource side. This means that the information on the user side is context-dependent and may be dynamic features that change with the context. The information on the media resource side, however, is context-independent, i.e., static features. There is no fixed approach to refining context features; many different approaches can be used to optimize performance. Offline truncation, through training, obtains the model structure and parameters of the user-side tower, as well as the embedding vector of each media resource on the resource side. Furthermore, during online prediction, a context-based user embedding vector is obtained for a specific user account and context. Using this vector, several media resources corresponding to similar vectors are searched from the resource embedding vectors as the recalled media resources.

[0076] However, the ANN triggering branch method also has a technical flaw: its accuracy is highly dependent on historical data. This characteristic is unfair to media information without historical data accumulation, or in other words, the ANN triggering branch method is not suitable for media information without historical data accumulation and is prone to missing high-quality media information that lacks historical data accumulation.

[0077] Optionally, in this embodiment, the above-mentioned media information recall method may also adopt, but is not limited to, the TAG trigger branch method. The TAG trigger branch method may include, but is not limited to, tagging user accounts (audience groups), then tagging media resources, finally matching the tags of the audience group and the tags of the media resources, and recalling the media resources based on the matching results.

[0078] To further illustrate, such as Figure 5 The system retrieves user tags 504 (such as behavioral targeting, interest targeting, intention targeting, installation targeting, and custom groups in the broad sense, and category, keyword, installed list, and group package list in the specific sense) and resource tags 506 (such as keywords, installed list, and other information) of media resource 508 from user account 502. Then, the matcher 510 is used to match user tags 504 and resource tags 506 to obtain the matching results.

[0079] Optionally, in this embodiment, the target audience for media information can be determined by combining the aforementioned TAG triggering branch and ANN triggering branch, for example, as shown below. Figure 6 As shown, the specific steps are as follows:

[0080] Step S602: Obtain user input, wherein the user input may include, but is not limited to, data from multiple dimensions, such as object feature data, resource feature data, context information, etc.

[0081] In steps S604-1 and S604-2, for various high-dimensional sparse features, the algorithm maps a single feature to a low-dimensional dense vector, transforming the "exact matching" of high-dimensional sparse features into a "fuzzy matching" on the embedding vector, thereby improving the algorithm performance and greatly enhancing the usability of high-dimensional sparse features. Specifically, embedding lookup can be achieved by using matrix multiplication, but is not limited to this.

[0082] The model structure may include, but is not limited to, an SE Block (Sequence and Excision) to perform data compression and information aggregation on the embedding vectors of each feature obtained in steps S604-1 and S604-2, and represent each feature with a bit. The model interacts with the MLP to obtain the results. For all the input features, the model dynamically determines which features are important and which are not by establishing correlations between them. In other words, the model constructs an operation mode to dynamically strengthen important features and weaken noisy or unimportant features.

[0083] Furthermore, the above model structure can be, but is not limited to, the C3MM model (Complete Conversion Chain Multi-task Model). The construction process of the C3MM model can be understood as, but is not limited to, a multi-task learning process among the training of the Click-Through-Rate (CTR) dual-tower model, the shallow Conversion Rate (CVR) dual-tower model, and the deep CVR dual-tower model. The goal of constructing the C3MM model is to minimize the weighted sum of the losses of the CTR dual-tower model, the shallow CVR dual-tower model, and the deep CVR dual-tower model, i.e., step S606.

[0084] Furthermore, in the above embodiments, the dual-tower structure of the CTR dual-tower model, the shallow CVR dual-tower model, and the deep CVR dual-tower model can be, but is not limited to, referencing... Figure 7 As shown, the dual-tower structure 702 is divided into two relatively independent sub-network structures, such as object tower 704-1 and resource tower 704-2. Object tower 704-1 may, but is not limited to, use object feature data 706-1 as input data (such as object identification data, object attribute data, object historical operation data, etc.), while resource tower 704-2 may, but is not limited to, use media resource resource feature data 706-2 (such as media resource publishing data, media resource feedback data, etc.) as input data.

[0085] Furthermore, after obtaining the processing results of object tower 704-1 and resource tower 704-2 respectively, the cosine distance between the two processing results can be calculated, but is not limited to, and the model parameters of the dual-tower structure 702 can be updated using the calculation results. In actual use, the processing results of object tower 704-1 and resource tower 704-2 can be integrated to obtain the final recall result 708 output by the dual-tower structure 702 (such as which media resource is recalled and the publishing object of the recalled media resource).

[0086] It should be noted that the process involves: obtaining a first target recall request associated with the target object, where the first target recall request is used to request the recall of target media information from N media information, where N is a natural number greater than or equal to 2; responding to the first target recall request, obtaining the information features corresponding to each of the N media information, where the information features are the content description features of the media information; recalling the first media information from the N media information based on the information features, where the similarity between the information features of the first media information and the target features corresponding to the target object reaches a target threshold, and the target media information includes the first media information; and recalling the second media information from M media information based on the number of media information remaining in the N media information after the first media information is recalled, where M media information are the media information remaining in the N media information after the first media information is recalled, where M is a natural number less than N, and the target media information includes the second media information.

[0087] To further illustrate, optional examples include... Figure 8 As shown, in response to a first target recall request associated with a target object, information features corresponding to each of the N media information 802 are obtained, wherein the information features are the content description features of the media information; first media information 804 (such as media information 1) is recalled from the N media information 802, wherein the similarity between the information features of the first media information 804 and the target features corresponding to the target object reaches a target threshold, and the target media information 810 includes the first media information 804 (such as media information 1); second media information 808 is recalled from M media information 806, wherein the M media information 806 are the media information in the N media information 802 other than the first media information 804 (such as media information 1), and the target media information 810 includes the second media information 808.

[0088] The embodiments provided in this application involve obtaining a first target recall request associated with a target object, wherein the first target recall request is used to request the recall of target media information from N media information, where N is a natural number greater than or equal to 2; responding to the first target recall request, obtaining the information features corresponding to each of the N media information, wherein the information features are the content description features of the media information; recalling the first media information from the N media information based on the information features, wherein the similarity between the information features of the first media information and the target features corresponding to the target object reaches a target threshold, and the target media information includes the first media information; and retrieving the remaining media information from the N media information after the first media information is recalled. The second media information is recalled from M media information, where M media information is the remaining media information among N media information after the first media information is recalled, and M is a natural number less than N. The target media information includes the second media information. In addition to recalling media information using feature similarity, the remaining media information is also recalled. In this way, even if some high-quality media information is missed by the advertising recall method using feature similarity, the missed media information can still be recalled using this supplementary recall method, thereby reducing the probability of high-quality media information being missed and thus achieving the technical effect of improving the recall accuracy of media information.

[0089] As an optional approach, the second media information is recalled from M media information based on the number of remaining media information among N media information after the first media information is recalled, including:

[0090] Recall a first target number of second media information from M media information, wherein the ratio between the first target number and M is the first target ratio.

[0091] Optionally, in this embodiment, recalling a first target number of second media information from M media information may be, but is not limited to, recalling media information that accounts for a first target proportion of the total number of media information from M media information as second media information. The method of recalling the second media information may be, but is not limited to, random. The advantage of randomness is that, in the absence of recall conditions with high reference value, randomness can provide a more stable recall result.

[0092] Optionally, in this embodiment, the method for recalling the second media information may be, but is not limited to, referring to the similarity between the information features corresponding to each of the M media information and the target features. To minimize the omission problem that exists in the feature similarity method, the similarity between the information features corresponding to each of the M media information and the target features may be sorted in descending order to recall a first target number of second media information from the M media information; or,

[0093] To compensate for the shortcomings of the feature similarity method, it is possible, but not limited to, to retrieve the first target number of second media information from the M media information by sorting the similarity between the information features corresponding to each media information and the target features in ascending order.

[0094] As an optional approach, after recalling the second media information from M media information based on the number of remaining media information among N media information after the first media information has been recalled, the process includes:

[0095] S1, obtain the first release result after the first media information is recalled, and the second release result after the second media information is recalled;

[0096] S2, adjust the first target ratio based on the first release result and / or the second release result to obtain the second target ratio.

[0097] Optionally, in this embodiment, the publishing result may be, but is not limited to, the publishing result after the media information is recalled, or it can be understood as the feedback information collected after the media information is published, such as the click-through rate, exposure conversion rate, click conversion rate, number of conversions, cost deviation (cost / GMV, number of conversions * actual bid associated with the media information), cumulative viewing time, and resource revenue associated with the media information within a certain period of time after the media information is published.

[0098] As an optional approach, after adjusting the first target ratio based on the first and second release results to obtain the second target ratio, the following steps are included:

[0099] S1, obtain the second target recall request associated with the target object, wherein the target recall request is used to request the recall of target media information from P media information, where P is a natural number greater than or equal to 2;

[0100] S2, respond to the second target recall request, and obtain the information features corresponding to each of the P media information;

[0101] S3, Retrieve the first media information from P media information;

[0102] S4, recall the second target quantity of second media information from K media information, where the ratio between the second target quantity and K is the second target ratio, K media information is the media information other than the first target media information from P media information, and K is a natural number less than P.

[0103] Optionally, in this embodiment, the target ratio can be adjusted in real time as the release results are released, in order to ensure the timeliness of media information recall.

[0104] As an optional approach, the first target ratio is adjusted based on the first and / or second release results to obtain the second target ratio, including:

[0105] S1, obtain a first release indicator based on the first release result and the second release result, wherein the first release indicator is a release effect indicator of the second media information relative to the first media information; adjust the first target ratio using the first release indicator to obtain the second target ratio; or,

[0106] S2, obtain the second release index of the second media information based on the second release result, wherein the second release index is the release effect index of the second media information; adjust the first target ratio using the second release index to obtain the second target ratio.

[0107] Optionally, in this embodiment, the first target ratio is adjusted specifically using the publication effect index of the second media information, which is suitable for application scenarios where the similarity judgment between the information features of the first media information and the target features corresponding to the target object is relatively vague; while the first target ratio is adjusted relatively using the publication effect indexes of both the first media information and the second media information, which is suitable for application scenarios where the similarity judgment between the information features of the first media information and the target features corresponding to the target object is relatively accurate.

[0108] Optionally, in this embodiment, obtaining the first release index based on the first release result and the second release result may include, but is not limited to, obtaining: the first achievement rate P1 of the first media information in the first release result and the second achievement rate P2 of the second media information in the second release result, wherein the aforementioned achievement rate refers to the proportion of media information consumption with cost achievement to the total media information consumption, and media information cost achievement refers to the media information cost being controlled within 20% of the target cost fluctuation range; further calculating the target calculation ratio Pr (first release index) between the second achievement rate P2 and the first achievement rate P1, as shown in the following formula (1):

[0109] P r =P2 / P1 (1)

[0110] As an optional approach, the first target ratio can be adjusted using the first release indicator to obtain the second target ratio, including:

[0111] Once the target attenuation coefficient is obtained, the first target ratio is adjusted using the target attenuation coefficient and the first release indicator to obtain the second target ratio. The target attenuation coefficient is an attenuation coefficient calculated based on time data.

[0112] Optionally, in this embodiment, a target decay coefficient alpha is set that decays over time. The target decay coefficient can be calculated and updated on an hourly basis, that is, the target decay coefficient for the next hour is calculated at the beginning of a certain hour. For details, please refer to the following formula (2):

[0113] alpha=alpha0*H*step (2)

[0114] Where alpha0 is the initial value of the target attenuation coefficient, for example, it can be set to 1.0; H is the current hour, for example, 0 o'clock or 1 o'clock; Step is the step size, which is a variable that can be set, for example, 0.02.

[0115] Further combining the target calculation ratio Pr (first release index) obtained by formula (1) above, and combining it with the following formula (3), the probability value p that can be used to adjust the first target ratio can be calculated:

[0116] p = p0 * (1 - alpha) * P r (3)

[0117] Where P0 is the initial value, for example, 0.1; alpha is the target attenuation coefficient; P r is the target calculation ratio (first published indicator); p is the probability value of adjusting the first target ratio.

[0118] As an optional approach, after recalling the second media information from M media information based on the number of remaining media information among N media information after the first media information has been recalled, the process includes:

[0119] Publish the first media information and the second media information in the target publication location associated with the target object.

[0120] Optionally, in this embodiment, the target object may be, but is not limited to, the audience user account of the recalled media information; the target publishing location associated with the target object may be, but is not limited to, the location of the media information to be published; and the first target recall request associated with the target object may be, but is not limited to, a request to obtain media information for the audience user account and the location of the media information to be published.

[0121] As an optional approach, after obtaining the first target recall request associated with the target object, the following steps are included:

[0122] S1, in response to the first target recall request, input N media information into the first tower structure of the target dual-tower model, wherein the target dual-tower model includes a first tower structure, a second tower structure and a target processing structure. The first tower structure is used to extract the information features corresponding to each media information, the second tower structure is used to extract the features corresponding to each object, the features corresponding to each object include the target object, the information features corresponding to each media information include the target information features, and the target processing structure is used to compare the feature similarity between the features corresponding to each object and the information features corresponding to each media information.

[0123] S2, after obtaining the information features corresponding to each of the N media information output by the first tower structure, the target processing structure is used to compare the feature similarity between the information features corresponding to each of the N media information and the target features to obtain the target information features whose similarity with the target features reaches the target threshold.

[0124] S3, recall the first media information corresponding to the target information feature.

[0125] Optionally, in this embodiment, the target dual-tower model can be, but is not limited to, an artificial neural network model. This artificial neural network model can be abstracted as a probability estimation problem like P(1|user, context, ad), which represents the probability of a user account seeing a media resource in a given context under a predefined target. Naturally, a higher probability is better, which is easy to understand. So how is this target defined, and what is this predefined target? This target should have the function of measuring the degree of matching between the user account and the media resource. Typically, it can be the click-through rate (CTR), click-to-conversion rate (CPC), or impression-to-conversion rate (PCP). Alternatively, multiple different predefined targets can be modeled separately, constructing different ANN triggering branches, and then multiple branches can work together. It can be seen that the ANN triggering branch strategy essentially involves constructing classic models such as CTR, CPC, and PCP models, and then using these models to determine whether the user account matches the media resource. It's easy to understand that the higher the probability of a user clicking on the ad, the more likely the user account is to match the media resource; the higher the probability of a user converting from the media resource, the more likely the user account is to match the media resource.

[0126] To further illustrate the process of collecting various features (such as object features, context features, advertising features, etc.) and using these features to construct corresponding target prediction models, such as... Figure 4As shown, the target prediction model typically has a dual-tower structure: one tower (target prediction model 402) is formed from object features and context features, and the other tower is formed from resource features. The dual-tower structure is the integrated target prediction model 404. In other words, contextual information only appears on the user side and not on the resource side. This means that the information on the user side is context-dependent and may be dynamic features that change with the context. The information on the media resource side, however, is context-independent, i.e., static features. There is no fixed approach to refining context features; many different approaches can be used to optimize performance. Offline truncation, through training, obtains the model structure and parameters of the user-side tower, as well as the embedding vector of each media resource on the resource side. Furthermore, during online prediction, a context-based user embedding vector is obtained for a specific user account and context. Using this vector, several media resources corresponding to similar vectors are searched from the resource embedding vectors as the recalled media resources.

[0127] As an optional approach, before obtaining the first target recall request associated with the target object, the following steps are included:

[0128] S1, obtain the object reference information and context information associated with the target object;

[0129] S2, input the object reference information and context information into the second tower structure to obtain the target features output by the second tower structure.

[0130] As an optional approach, the aforementioned media information recall method can be applied, but is not limited to, in advertising scenarios. For example, when a request (the first target recall request associated with the target object) arrives, useful information (of the target object) can be extracted from this request, such as user information and contextual information. This information can be processed into target features to describe the aforementioned request. Furthermore, the similarity between the target features and the information features of the advertisements to be recalled (N media information items) can be calculated. Ads with a similarity greater than or equal to a threshold can be used as recalled advertisements, while advertisements that are not recalled are randomly added to the overall recall queue.

[0131] Further examples, such as Figure 9 As shown, the specific steps are as follows:

[0132] Step S902: Collect user features, context features, and ad features to train a dual-tower model, obtaining two types of data: one is the model of the user-side tower; the other is the ad embedding vector of the ad-side tower. The user-side tower model is used to process new input request information; the ad embedding vector of the ad-side tower is used to represent each ad message.

[0133] Step S904: When a request arrives, extract useful information from the request, such as user information and context information. Assemble this information into the input required for the model of the user side, and then obtain the output of this side, which is a vector used to represent the request;

[0134] Step S906: Calculate the similarity between the vector of the expression request obtained in step S904 and the vector of the expression advertisement. Ads with close similarity can be used as recall ads; otherwise, they can be discarded. A threshold can be set to measure the similarity. Ads with similarity greater than this threshold can be used as recall ads.

[0135] Step S908: For ads that were not recalled, randomly add them to the overall recall queue with probability p. Assuming there are 100 ads that were not recalled and probability p is 0.1, then 10 ads will be randomly added to the recall queue.

[0136] In step S910, the advertisement obtained in step S906 is marked as "hard recall", and the advertisement obtained in step S908 is marked as "soft recall".

[0137] Step S912, probability p is a crucial algorithm parameter. It can be updated using a time-decay mechanism combined with ad performance analysis. The update strategy for probability p is as follows: A time-decay coefficient alpha is set. Time decay is calculated and updated hourly. That is, at the beginning of a certain hour, the decay coefficient for the next hour is calculated. Furthermore, the update of probability p is also related to real-time performance. Here, ad achievement rate is used to represent real-time performance. Assuming the achievement rate of hard-recall ads is P1 and the achievement rate of soft-recall ads is P2, the performance metric is defined as the ratio of the two. The aforementioned ad achievement rate refers to the proportion of cost-effective ad spending to total ad spending. Cost-effective ad spending means that the ad cost is controlled within 20% of the target cost.

[0138] Through the embodiments of this application, when a request (a first target recall request associated with the target object) arrives, useful information (of the target object), such as user information and contextual information, is extracted from this request; this information is processed into target features to describe the aforementioned request; further, the similarity between the target features and the information features of the advertisements to be recalled (N media information) is calculated; advertisements with a similarity greater than or equal to a threshold can be used as recalled advertisements, while advertisements that are not recalled are randomly added to the overall recall queue, which is more conducive to cost control of advertisements in operation, agency and other scenarios; it is also conducive to improving the reliability and usability of products such as OCPA. Controllable advertising costs help build reputation; by improving the performance of the advertising system, it is conducive to building and consolidating advertisers' confidence in placing orders.

[0139] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0140] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0141] According to another aspect of the embodiments of this application, a media information recall apparatus for implementing the above-described media information recall method is also provided. For example... Figure 10 As shown, the device includes:

[0142] The first acquisition unit 1002 is used to acquire the first target recall request associated with the target object, wherein the first target recall request is used to request the recall of target media information from N media information, where N is a natural number greater than or equal to 2.

[0143] The second acquisition unit 1004 is used to respond to the first target recall request and acquire the information features corresponding to each of the N media information, wherein the information features are the content description features of the media information;

[0144] The first recall unit 1006 is used to recall first media information from N media information based on information features, wherein the similarity between the information features of the first media information and the target features corresponding to the target object reaches a target threshold, and the target media information includes the first media information.

[0145] The second recall unit 1008 is used to recall the second media information from M media information based on the number of media information remaining in N media information after the first media information is recalled. Here, M media information are the media information remaining in N media information after the first media information is recalled, M is a natural number less than N, and the target media information includes the second media information.

[0146] Optionally, in this embodiment, the media information recall method described above can be applied to, but is not limited to, advertising scenarios. For example, when a request (the first target recall request associated with the target object) arrives, useful information (of the target object) is extracted from the request, such as user information and context information; this information is processed into target features to describe the request; the similarity between the target features and the information features of the advertisements to be recalled (N media information items) is further calculated; advertisements with a similarity greater than or equal to a threshold can be used as recalled advertisements, while advertisements that are not recalled are randomly added to the overall recall queue.

[0147] Optionally, in this embodiment, the target object may be, but is not limited to, the location object of the media information to be published, and / or the audience user account of the recalled media information; and the first target recall request associated with the target object may be, but is not limited to, a request to recall the media information to be published in the above-mentioned location object, and / or to be targeted at the audience user account.

[0148] Optionally, in this embodiment, recalling the second media information from M media information based on the number of remaining media information among N media information after the first media information is recalled can be, but is not limited to, recalling the second media information from M media information based on the recall result associated with the target object. The recall result can be, but is not limited to, the publication result after the media information is recalled, or can be understood as the feedback information collected after the media information is published, such as the click-through rate, exposure conversion rate, click conversion rate, number of conversions, cost deviation (cost / GMV, number of conversions * actual bid associated with the media information), cumulative viewing time, and resource revenue associated with the media information within a certain period after the media information is published.

[0149] Optionally, in this embodiment, recalling the second media information from M media information based on the number of remaining media information among N media information after the first media information is recalled can be, but is not limited to, recalling the second media information from M media information based on the number of remaining media information among N media information after the first media information is recalled, according to the target data threshold matched by the recall result associated with the target object. The target data threshold can be, but is not limited to, a range data threshold, and each range data threshold corresponds to a recall reference information. Then, the matched target range data threshold is obtained first based on the recall result associated with the target object, and the target recall reference information corresponding to the target range data threshold is obtained. The target recall reference information is then used to recall the second media information from M media information based on the number of remaining media information among N media information after the first media information is recalled, thereby achieving accurate supplementary recall.

[0150] Optionally, in this embodiment, obtaining the information features corresponding to each of the N media information can be, but is not limited to, using the ANN (Artificial Neural Network) triggering branch method. The ANN triggering branch method can be, but is not limited to, understanding as building an artificial neural network model and using this model to observe from historical data which objects may be high-conversion groups and have relatively high advertising efficiency.

[0151] It should be noted that the process involves: obtaining a first target recall request associated with the target object, where the first target recall request is used to request the recall of target media information from N media information, where N is a natural number greater than or equal to 2; responding to the first target recall request, obtaining the information features corresponding to each of the N media information, where the information features are the content description features of the media information; recalling the first media information from the N media information based on the information features, where the similarity between the information features of the first media information and the target features corresponding to the target object reaches a target threshold, and the target media information includes the first media information; and recalling the second media information from M media information based on the number of media information remaining in the N media information after the first media information is recalled, where M media information are the media information remaining in the N media information after the first media information is recalled, where M is a natural number less than N, and the target media information includes the second media information.

[0152] For specific implementation examples, please refer to the examples shown in the above media information recall method, which will not be repeated here.

[0153] The embodiments provided in this application involve obtaining a first target recall request associated with a target object, wherein the first target recall request is used to request the recall of target media information from N media information, where N is a natural number greater than or equal to 2; responding to the first target recall request, obtaining the information features corresponding to each of the N media information, wherein the information features are the content description features of the media information; recalling the first media information from the N media information based on the information features, wherein the similarity between the information features of the first media information and the target features corresponding to the target object reaches a target threshold, and the target media information includes the first media information; and retrieving the remaining media information from the N media information after the first media information is recalled. The second media information is recalled from M media information, where M media information is the remaining media information among N media information after the first media information is recalled, and M is a natural number less than N. The target media information includes the second media information. In addition to recalling media information using feature similarity, the remaining media information is also recalled. In this way, even if some high-quality media information is missed by the advertising recall method using feature similarity, the missed media information can still be recalled using this supplementary recall method, thereby reducing the probability of high-quality media information being missed and thus achieving the technical effect of improving the recall accuracy of media information.

[0154] As an optional solution, the second recall unit 1008 includes:

[0155] The recall module is used to recall a first target number of second media information from M media information, wherein the ratio between the first target number and M is the first target ratio.

[0156] For specific implementation examples, please refer to the examples shown in the above media information recall method, which will not be repeated here.

[0157] As an optional solution, it includes:

[0158] The first acquisition module is used to acquire the first release result after the first media information is recalled and the second release result after the second media information is recalled, after recalling the second media information from M media information based on the number of media information remaining in N media information after the first media information is recalled.

[0159] The adjustment module is used to, after recalling the second media information from M media information based on the number of media information remaining in N media information after the first media information has been recalled, adjust the first target ratio according to the first release result and / or the second release result to obtain the second target ratio.

[0160] For specific implementation examples, please refer to the examples shown in the above media information recall method, which will not be repeated here.

[0161] As an optional solution, it includes:

[0162] The second acquisition module is used to acquire the second target recall request associated with the target object after adjusting the first target ratio according to the first release result and the second release result to obtain the second target ratio. The target recall request is used to request the recall of target media information from P media information, where P is a natural number greater than or equal to 2.

[0163] The third acquisition module is used to adjust the first target ratio according to the first release result and the second release result to obtain the second target ratio, and then respond to the second target recall request to acquire the information features corresponding to each media information in P media information.

[0164] The first recall module is used to recall the first media information from P media information after adjusting the first target ratio according to the first release result and the second release result to obtain the second target ratio;

[0165] The second recall module is used to recall the second target number of second media information from K media information after adjusting the first target ratio according to the first release result and the second release result to obtain the second target ratio. The ratio between the second target number and K is the second target ratio, and the K media information are the media information other than the first target media information from P media information, where K is a natural number less than P.

[0166] For specific implementation examples, please refer to the examples shown in the above media information recall method, which will not be repeated here.

[0167] As an optional solution, the module is adjusted, including:

[0168] The first acquisition submodule is used to acquire a first release indicator based on the first release result and the second release result, wherein the first release indicator is a release effect indicator of the second media information relative to the first media information; adjust the first target ratio using the first release indicator to obtain the second target ratio; or,

[0169] The second acquisition submodule is used to acquire the second release index of the second media information based on the second release result, wherein the second release index is the release effect index of the second media information; and to adjust the first target ratio using the second release index to obtain the second target ratio.

[0170] For specific implementation examples, please refer to the examples shown in the above media information recall method, which will not be repeated here.

[0171] As an optional approach, the first acquisition submodule includes:

[0172] The adjustment subunit is used to adjust the first target ratio using the target attenuation coefficient and the first release index, based on the obtained target attenuation coefficient, to obtain the second target ratio. The target attenuation coefficient is an attenuation coefficient calculated based on time data.

[0173] For specific implementation examples, please refer to the examples shown in the above media information recall method, which will not be repeated here.

[0174] As an optional solution, it includes:

[0175] The publishing unit is used to recall the second media information from M media information based on the number of media information remaining in N media information after the first media information has been recalled, and then publish the first media information and the second media information at the target publishing location associated with the target object.

[0176] For specific implementation examples, please refer to the examples shown in the above media information recall method, which will not be repeated here.

[0177] As an optional solution, it includes:

[0178] The first input unit is used to respond to the first target recall request after obtaining the first target recall request associated with the target object, and input N media information into the first tower structure in the target dual-tower model. The target dual-tower model includes a first tower structure, a second tower structure, and a target processing structure. The first tower structure is used to extract the information features corresponding to each media information. The second tower structure is used to extract the features corresponding to each object. The features corresponding to each object include the target object. The information features corresponding to each media information include the target information features. The target processing structure is used to compare the feature similarity between the features corresponding to each object and the information features corresponding to each media information.

[0179] The processing unit is used to, after obtaining the first target recall request associated with the target object, and after obtaining the information features corresponding to each of the N media information output by the first tower structure, use the target processing structure to compare the feature similarity between the information features corresponding to each of the N media information and the target features to obtain the target information features whose similarity to the target features reaches the target threshold.

[0180] The third recall unit is used to recall the first media information corresponding to the target information features after obtaining the first target recall request associated with the target object.

[0181] For specific implementation examples, please refer to the examples shown in the above media information recall method, which will not be repeated here.

[0182] As an optional solution, it includes:

[0183] The third acquisition unit is used to acquire object reference information and context information associated with the target object before acquiring the first target recall request associated with the target object;

[0184] The second input unit is used to input object reference information and context information into the second tower structure before obtaining the first target recall request associated with the target object, so as to obtain the target features output by the second tower structure.

[0185] For specific implementation examples, please refer to the examples shown in the above media information recall method, which will not be repeated here.

[0186] According to another aspect of the embodiments of this application, an electronic device for implementing the above-described media information recall method is also provided, such as... Figure 11 As shown, the electronic device includes a memory 1102 and a processor 1104. The memory 1102 stores a computer program, and the processor 1104 is configured to execute the steps of any of the above method embodiments via the computer program.

[0187] Optionally, in this embodiment, the aforementioned electronic device may be located in at least one of a plurality of network devices in a computer network.

[0188] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0189] S1, obtain the first target recall request associated with the target object, wherein the first target recall request is used to request the recall of target media information from N media information, where N is a natural number greater than or equal to 2;

[0190] S2, responding to the first target recall request, obtain the information features corresponding to each of the N media information, where the information features are the content description features of the media information;

[0191] S3, based on information features, recall the first media information from N media information, wherein the similarity between the information features of the first media information and the target features corresponding to the target object reaches the target threshold, and the target media information includes the first media information;

[0192] S4. Based on the number of media information remaining among N media information after the first media information is recalled, recall the second media information from M media information. Here, M media information are the media information remaining among N media information after the first media information is recalled, and M is a natural number less than N. The target media information includes the second media information.

[0193] Alternatively, as those skilled in the art will understand, Figure 11 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones (such as Android phones, iOS phones, etc.), tablets, PDAs, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 11 This does not limit the structure of the aforementioned electronic devices. For example, the electronic device may also include components that are more... Figure 11 The more or fewer components shown (such as network interfaces, etc.), or having the same Figure 11 The different configurations shown.

[0194] The memory 1102 can be used to store software programs and modules, such as the program instructions / modules corresponding to the media information recall method and apparatus in this embodiment. The processor 1104 executes various functional applications and data processing by running the software programs and modules stored in the memory 1102, thereby realizing the aforementioned media information recall method. The memory 1102 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 1102 may further include memory remotely located relative to the processor 1104, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. Specifically, the memory 1102 may be used, but is not limited to, to store information such as the first target recall request, target features, and target media information. As an example, such as Figure 11 As shown, the memory 1102 may include, but is not limited to, the first acquisition unit 1002, the second acquisition unit 1004, the first recall unit 1006, and the second recall unit 1008 in the media information recall device. Furthermore, it may include, but is not limited to, other module units in the media information recall device, which will not be elaborated upon in this example.

[0195] Optionally, the transmission device 1106 described above is used to receive or send data via a network. Specific examples of the network described above may include wired networks and wireless networks. In one example, the transmission device 1106 includes a Network Interface Controller (NIC), which can be connected to other network devices and a router via a network cable to communicate with the Internet or a local area network. In another example, the transmission device 1106 is a radio frequency (RF) module, used for wireless communication with the Internet.

[0196] In addition, the aforementioned electronic device also includes: a display 1108 for displaying information such as the first target recall request, target characteristics, and target media information; and a connection bus 1110 for connecting various module components in the aforementioned electronic device.

[0197] In other embodiments, the aforementioned terminal device or server can be a node in a distributed system, wherein the distributed system can be a blockchain system, which is a distributed system formed by connecting multiple nodes through network communication. The nodes can form a peer-to-peer (P2P) network, and any form of computing device, such as a server, terminal, or other electronic device, can become a node in the blockchain system by joining this peer-to-peer network.

[0198] According to one aspect of this application, a computer program product is provided, comprising a computer program / instructions containing program code for performing the methods shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit, it performs various functions provided in embodiments of this application.

[0199] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0200] It should be noted that the computer system of the electronic device is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0201] A computer system includes a Central Processing Unit (CPU), which performs various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) or loaded from RAM. ROM also stores various programs and data required for system operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output interfaces (I / O interfaces) are also connected to the bus.

[0202] The following components are connected to the input / output interface: input sections including keyboards, mice, etc.; output sections including cathode ray tubes (CRTs), liquid crystal displays (LCDs), and speakers; storage sections including hard drives; and communication sections including network interface cards such as LAN cards and modems. The communication section performs communication processing via a network such as the Internet. Drives are also connected to the input / output interface as needed. Removable media, such as disks, optical discs, magneto-optical discs, semiconductor memories, etc., are installed on the drive as needed so that computer programs read from them can be installed into the storage section as required.

[0203] Specifically, according to embodiments of this application, the processes described in the various method flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit, it performs various functions defined in the system of this application.

[0204] According to one aspect of this application, a computer-readable storage medium is provided, wherein a processor of a computer device reads computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the computer device to perform the methods provided in the various alternative implementations described above.

[0205] Optionally, in this embodiment, the computer-readable storage medium described above may be configured to store a computer program for performing the following steps:

[0206] S1, obtain the first target recall request associated with the target object, wherein the first target recall request is used to request the recall of target media information from N media information, where N is a natural number greater than or equal to 2;

[0207] S2, in response to the first target recall request, obtain the information features corresponding to each of the N media information, wherein the information features are the content description features of the media information;

[0208] S3, recall the first media information from the N media information, wherein the similarity between the information features of the first media information and the target features corresponding to the target object reaches a target threshold, and the target media information includes the first media information;

[0209] S4, based on the number of remaining media information among N media information after the first media information is recalled, recall the second media information from M media information, wherein the M media information are the media information other than the first media information among the N media information, M is a natural number less than N, and the target media information includes the second media information.

[0210] Optionally, in this embodiment, those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0211] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0212] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0213] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0214] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between units or modules, and may be electrical or other forms.

[0215] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0216] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0217] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for recalling media information, characterized in that, include: Obtain the first target recall request associated with the target object, wherein the first target recall request is used to request the recall of target media information from N media information, where N is a natural number greater than or equal to 2; In response to the first target recall request, obtain the information features corresponding to each of the N media information, wherein the information features are the content description features of the media information; Based on the information features, first media information is recalled from the N media information, wherein the similarity between the information features of the first media information and the target features corresponding to the target object reaches a target threshold, and the target media information includes the first media information; Recalling second media information from M media information based on the number of remaining media information among N media information after the first media information is recalled includes: recalling the second media information from the M media information by sorting the similarity between the information features corresponding to each media information and the target feature in descending or ascending order, wherein the M media information are the remaining media information among the N media information after the first media information is recalled, M is a natural number less than N, and the target media information includes the second media information.

2. The method according to claim 1, characterized in that, The step of recalling the second media information from M media information based on the number of remaining media information among N media information after the first media information has been recalled includes: Recall a first target number of second media information from M media information, wherein the ratio between the first target number and M is the first target ratio.

3. The method according to claim 2, characterized in that, After recalling the second media information from M media information based on the remaining number of media information among N media information after the first media information has been recalled, the process includes: Obtain the first release result after the first media information is recalled, and the second release result after the second media information is recalled; The first target ratio is adjusted based on the first release result and / or the second release result to obtain the second target ratio.

4. The method according to claim 3, characterized in that, After adjusting the first target ratio based on the first release result and the second release result to obtain the second target ratio, the process includes: Obtain the second target recall request associated with the target object, wherein the target recall request is used to request the recall of the target media information from P media information, where P is a natural number greater than or equal to 2; In response to the second target recall request, obtain the information features corresponding to each of the P media information items; Recall the first media information from the P media information; Recall a second target number of the second media information from K media information, wherein the ratio between the second target number and K is the second target ratio, and the K media information are the media information other than the first target media information from the P media information, where K is a natural number less than P.

5. The method according to claim 3, characterized in that, The step of adjusting the first target ratio based on the first release result and / or the second release result to obtain the second target ratio includes: A first release indicator is obtained based on the first release result and the second release result, wherein the first release indicator is a release effect indicator of the second media information relative to the first media information; the first target ratio is adjusted using the first release indicator to obtain the second target ratio; or, The second release index of the second media information is obtained based on the second release result, wherein the second release index is the release effect index of the second media information; the first target ratio is adjusted using the second release index to obtain the second target ratio.

6. The method according to claim 5, characterized in that, The step of adjusting the first target ratio using the first release indicator to obtain the second target ratio includes: Once the target attenuation coefficient is obtained, the first target ratio is adjusted using the target attenuation coefficient and the first release indicator to obtain the second target ratio, wherein the target attenuation coefficient is an attenuation coefficient calculated based on time data.

7. The method according to any one of claims 1 to 6, characterized in that, After recalling the second media information from M media information based on the remaining number of media information among N media information after the first media information has been recalled, the process includes: The first media information and the second media information are published at the target publication location associated with the target object.

8. The method according to any one of claims 1 to 6, characterized in that, After obtaining the first target recall request associated with the target object, the process includes: In response to the first target recall request, the N media information items are input into the first tower structure of the target dual-tower model. The target dual-tower model includes the first tower structure, the second tower structure, and the target processing structure. The first tower structure is used to extract information features corresponding to each media information item. The second tower structure is used to extract features corresponding to each object. The features corresponding to each object include the target object. The information features corresponding to each media information item include target information features. The target processing structure is used to compare the feature similarity between the features corresponding to each object and the information features corresponding to each media information item. When the information features corresponding to each of the N media information output by the first tower structure are obtained, the target processing structure is used to compare the feature similarity between the information features corresponding to each of the N media information and the target feature to obtain the target information feature whose similarity with the target feature reaches the target threshold. Recall the first media information corresponding to the target information feature.

9. The method according to claim 8, characterized in that, Before obtaining the first target recall request associated with the target object, the process includes: Obtain the object reference information and context information associated with the target object; The object reference information and the context information are input into the second tower structure to obtain the target feature output by the second tower structure.

10. A media information recall device, characterized in that, include: The first acquisition unit is used to acquire the first target recall request associated with the target object, wherein the first target recall request is used to request the recall of target media information from N media information, where N is a natural number greater than or equal to 2; The second acquisition unit is used to respond to the first target recall request and acquire the information features corresponding to each of the N media information, wherein the information features are the content description features of the media information; The first recall unit is configured to recall first media information from the N media information based on the information features, wherein the similarity between the information features of the first media information and the target features corresponding to the target object reaches a target threshold, and the target media information includes the first media information. The second recall unit is configured to recall second media information from M media information based on the information features and the number of remaining media information among N media information after the first media information has been recalled. This includes: recalling the second media information from the M media information by ranking the similarity between the information features corresponding to each of the M media information and the target features in descending or ascending order. The M media information refers to the remaining media information among the N media information after the first media information has been recalled, M is a natural number less than N, and the target media information includes the second media information.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when executed, performs the method according to any one of claims 1 to 9.

12. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1 to 9.

13. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the method described in any one of claims 1 to 9 through the computer program.

Citation Information

Patent Citations

  • Content recommendation method and device, server and storage medium

    CN113010790A