Media recommendation method and device, electronic equipment, medium and product

Abstract

The application provides a media recommendation method and device, electronic equipment, medium and product, and belongs to the technical field of Internet. The method comprises the following steps: dividing media data elements of each to-be-recommended media into a media quality category, a media volume category and a media benefit category; for each to-be-recommended media, inputting the media data elements of the media quality category, the media volume category and the media benefit category into a first scoring model, a second scoring model and a third scoring model based on a factor decomposition machine model to obtain media quality scores, media volume scores and media benefit scores output by the corresponding scoring models; performing a superior-inferior solution distance analysis on the media quality scores, the media volume scores and the media benefit scores corresponding to each to-be-recommended media to obtain media scores of each to-be-recommended media; and performing media recommendation on each to-be-recommended media based on the media scores. The application can realize media recommendation in a scientific, reasonable and notarized manner, and effectively improve the accuracy of media recommendation.

Description

Technical Field

[0001] This application relates to the field of Internet technology, and in particular to a media asset recommendation method, apparatus, electronic device, medium, and product. Background Technology

[0002] Currently, when recommending media assets, television stations often rely on vague, quantitative, and subjective criteria such as the provider's historical reputation within the industry and the station's close relationship with the provider. These criteria lack scientific, reasonable, and impartial quantitative ranking standards. Furthermore, there is limited user-side data, making it impossible to provide analytical suggestions from the user's perspective.

[0003] Therefore, there is a lack of accuracy in current media asset recommendations. Summary of the Invention

[0004] This application provides a media asset recommendation method, apparatus, electronic device, medium, and product to address the current problem of insufficient accuracy in media asset recommendation and improve the accuracy of media asset recommendation.

[0005] This application provides a media asset recommendation method, the method comprising: dividing the media asset data elements of each media asset to be recommended into media asset quality category, media asset volume category, and media asset benefit category; for each media asset to be recommended, inputting the media asset data elements of the media asset quality category, media asset volume category, and media asset benefit category into a first scoring model, a second scoring model, and a third scoring model respectively, to obtain the media asset quality score, media asset volume score, and media asset benefit score output by the first scoring model, the second scoring model, and the third scoring model respectively; the first scoring model, the second scoring model, and the third scoring model are obtained by training a factorization machine model based on the sample media asset data elements and their scoring labels of each of the media asset quality category, media asset volume category, and media benefit category of the sample media assets; performing a distance analysis of the superior and inferior solutions based on the media asset quality score, media asset volume score, and media benefit score corresponding to each media asset to be recommended, to obtain the media asset score of each media asset to be recommended; and recommending the media assets based on the media asset scores.

[0006] According to the media asset recommendation method provided in this application, the step of performing a distance analysis of superior and inferior solutions based on the media asset quality score, media asset volume score, and media asset benefit score corresponding to each media asset to be recommended, to obtain the media asset score of each media asset to be recommended, includes: determining a benchmark media asset from among the media assets to be recommended based on the media asset indicators of each media asset to be recommended; performing positive index processing on the media asset quality score, media asset volume score, and media asset benefit score corresponding to each media asset to be recommended based on the media asset quality score, media asset volume score, and media asset benefit score of the benchmark media asset, to obtain a positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each media asset to be recommended; and performing a distance analysis of superior and inferior solutions based on the positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each media asset to be recommended, to obtain the media asset score of each media asset to be recommended.

[0007] According to the media asset recommendation method provided in this application, the step of performing a distance analysis of superior and inferior solutions based on the positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each media asset to be recommended, to obtain the media asset score of each media asset to be recommended, includes: constructing a positive matrix by taking the positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each media asset to be recommended as a row of a matrix; standardizing the positive matrix to obtain a standardized matrix; and performing a distance analysis of superior and inferior solutions based on the standardized matrix to obtain the media asset score of each media asset to be recommended.

[0008] According to a media asset recommendation method provided in this application, the step of performing a distance analysis of superior and inferior solutions based on the standardized matrix to obtain the media asset score of each media asset to be recommended includes: determining the maximum and minimum values ​​based on each element of the standardized matrix; wherein the maximum value includes the element with the largest value in each column of the standardized matrix; the minimum value includes the element with the smallest value in each column of the standardized matrix; determining a first distance between the corresponding media asset to be recommended and the maximum value based on the elements corresponding to each media asset to be recommended in the standardized matrix; determining a second distance between the corresponding media asset to be recommended and the minimum value based on the elements corresponding to each media asset to be recommended in the standardized matrix; and determining the media asset score of each media asset to be recommended based on the first and second distances.

[0009] According to the media asset recommendation method provided in this application, the step of determining the media asset score of each media asset to be recommended based on the first distance and the second distance corresponding to each media asset to be recommended includes: determining the unnormalized score of each media asset to be recommended based on the first distance and the second distance corresponding to each media asset to be recommended; determining the normalized score of each media asset to be recommended based on each unnormalized score; and determining the normalized score of each media asset to be recommended as the media asset score of the corresponding media asset to be recommended.

[0010] According to the media asset recommendation method provided in this application, when determining the unnormalized score of the corresponding media asset to be recommended based on the first distance and the second distance corresponding to each media asset to be recommended, the following steps are performed for each media asset to be recommended: determining the sum of the first distance and the second distance of the current media asset to be recommended; and determining the ratio of the second distance of the current media asset to the sum of the sums as the unnormalized score of the current media asset to be recommended.

[0011] According to the media asset recommendation method provided in this application, when determining the normalized score of each media asset to be recommended based on each unnormalized score, the following steps are performed for each media asset to be recommended: determining the sum of each unnormalized score; and determining the ratio of the unnormalized score of the current media asset to be recommended to the sum of the unnormalized scores as the normalized score of the current media asset to be recommended.

[0012] This application also provides a media asset recommendation device, comprising: a segmentation module for segmenting the media asset data elements of each media asset to be recommended into media asset quality, media asset volume, and media asset benefit categories; an input module for inputting the media asset data elements of the media asset quality, media asset volume, and media asset benefit categories into a first scoring model, a second scoring model, and a third scoring model for each media asset to be recommended, thereby obtaining media asset quality scores, media asset volume scores, and media asset benefit scores output by the first scoring model, the second scoring model, and the third scoring model, respectively; the first scoring model, the second scoring model, and the third scoring model are obtained by training a factorization machine model based on the sample media asset data elements and their scoring labels in each of the media asset quality, media asset volume, and media benefit categories of the sample media assets; an analysis module for performing a distance analysis of superior and inferior solutions based on the media asset quality scores, media asset volume scores, and media benefit scores corresponding to each media asset to be recommended, thereby obtaining a media asset score for each media asset to be recommended; and a determination module for recommending each media asset based on its score.

[0013] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the media asset recommendation method as described above.

[0014] This application also provides a medium, which is a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the media asset recommendation method as described above.

[0015] This application also provides a product, which is a computer program product, including a computer program that, when executed by a processor, implements any of the media asset recommendation methods described above.

[0016] The media asset recommendation method, apparatus, electronic device, medium, and product provided in this application divide the media asset data elements of each media asset to be recommended into media asset quality, media asset volume, and media asset benefit categories. Since a first scoring model, a second scoring model, and a third scoring model are obtained by pre-training a factorization machine model based on the sample media asset data elements and their scoring labels for each of the three categories (media asset quality, media volume, and media benefit), the media asset data elements of each category can be input into the first, second, and third scoring models respectively for each media asset to be recommended, resulting in accurate media asset quality, media asset volume, and media benefit scores output by the first, second, and third scoring models, respectively. Furthermore, a distance analysis of the superior and inferior solutions can be performed based on the media asset quality, media volume, and media benefit scores corresponding to each media asset to be recommended, obtaining an accurate media asset score for each media asset to be recommended, enabling media asset recommendation based on each media asset score. By employing user-side media asset data elements and categorizing them into media asset quality, media asset volume, and media asset effectiveness, a scoring model trained on a factorization machine model can be used to score each type of media asset data element. Furthermore, a distance analysis of the superior and inferior solutions is performed based on each score, and media asset recommendations are made for each media asset to be recommended based on the scores obtained from the analysis. This enables accurate media asset recommendations through a scientific, reasonable, and impartial approach, effectively improving the accuracy of media asset recommendations. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the media asset recommendation method provided in this application.

[0019] Figure 2 This is a schematic diagram illustrating the application scenario of the media asset recommendation method provided in this application.

[0020] Figure 3 This is a schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. The terms "upper," "lower," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are used only for the convenience of describing this application and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Unless otherwise expressly specified and limited, the terms "installed," "connected," and "linked" should be interpreted broadly, for example, as a fixed connection, a detachable connection, or an integral connection; a mechanical connection or an electrical connection; a direct connection or an indirect connection through an intermediate medium; or a connection within two elements. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0023] The terms "first," "second," etc., used in this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class, without limiting the number of objects; for example, a first object can be one or more. Furthermore, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects have an "or" relationship.

[0024] The following is combined Figures 1-3This application describes the media asset recommendation method, apparatus, electronic device, medium, and product.

[0025] Figure 1 This is a flowchart illustrating the media asset recommendation method provided in this application, such as... Figure 1 As shown, the method includes the following steps 100 to 400.

[0026] Step 100: Divide the media asset data elements of each media asset to be recommended into media asset quality, media asset volume, and media asset benefit categories.

[0027] Step 200: For each media asset to be recommended, input the media asset data elements of the media asset quality category, media asset volume category, and media asset benefit category into the first scoring model, the second scoring model, and the third scoring model, respectively, to obtain the media asset quality score, media asset volume score, and media asset benefit score output by the first scoring model, the second scoring model, and the third scoring model, respectively. The first scoring model, the second scoring model, and the third scoring model are obtained by training the factorization machine model based on the sample media asset data elements and their scoring labels of each category of the sample media asset quality category, the media asset volume category, and the media benefit category.

[0028] Step 300: Perform a distance analysis of the superior and inferior solutions based on the media asset quality score, media asset volume score, and media asset benefit score corresponding to each media asset to be recommended, and obtain the media asset score of each media asset to be recommended.

[0029] Step 400: Recommend each media asset to be recommended based on its score.

[0030] It should be noted that the execution subject of the media asset recommendation method provided in this application embodiment can be a server, computer equipment, such as a mobile phone, tablet computer, laptop computer, handheld computer, vehicle electronic equipment, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc.

[0031] The server or computer equipment of this application may be equipped with or connected to a media asset recommendation device, thereby controlling the media asset recommendation device to execute the media asset recommendation method of this application.

[0032] This application defines each media asset that needs to be recommended as a media asset to be recommended, such as the content of a channel's navigation bar homepage. Furthermore, it can obtain the key data elements of each media asset to be recommended as media asset data elements. The media assets in this application can include, but are not limited to, programs, movies, TV series, documentaries, variety shows, news programs, etc.

[0033] Furthermore, for each media asset to be recommended, this application can classify the media asset data elements into media asset quality, media asset volume, and media asset effectiveness categories according to attribute type.

[0034] In one embodiment, for a TV series, this type of media asset may include, but is not limited to, key data elements such as video resolution, release date, copyright holding period, whether it is three-dimensional (3D), whether it is exclusively broadcast, episode length, episode length, number of episodes, completion rate, user rating, awards, and subscription product categories and prices.

[0035] Furthermore, for media assets of a certain type of TV series, key data elements can be divided into media asset quality categories according to attributes: video clarity, release time, copyright holding time, whether it is 3D, and whether it is exclusively broadcast; media asset volume categories: series length, sub-episode length, number of sub-episodes, and completion rate; and media asset benefit categories: user ratings, awards, and subscription product categories and prices.

[0036] It should be noted that this application can define a certain number of media assets as sample media assets, and obtain key data elements of media asset quality, media asset volume, and media asset effectiveness as sample media asset data elements. Simultaneously, it obtains media asset scores set for each type of sample media asset data element (media asset quality, media asset volume, and media asset effectiveness) as scoring labels for the corresponding type of sample media asset. That is, each type of each sample media asset has a corresponding scoring label; specifically, the media asset quality type has one scoring label, the media asset volume type has one scoring label, and the media asset effectiveness type also has one scoring label.

[0037] Furthermore, the sample media asset data elements of the media asset quality class and their rating labels can be used to train the Factorization Machine (FM) model. After training, a first rating model can be obtained that can score based on the input media asset quality data elements.

[0038] Furthermore, the factorization machine model can be trained using sample media asset data elements of the media asset volume class and their rating labels. After training, a second rating model can be obtained that can score based on the input media asset data elements of the media asset volume class.

[0039] Furthermore, the factorization machine model can be trained using sample media asset data elements of the media asset benefit category and their rating labels. After training, a third rating model can be obtained that can score based on the input media asset benefit metadata.

[0040] It can be understood that for different types of media assets, such as movies, TV dramas, etc., the factorization machine model may be trained respectively according to various sample media asset metadata and their rating labels of the corresponding sample media assets, to obtain a rating model for the corresponding type of media assets, and the rating model for each media asset may include a first rating model, a second rating model, and a third rating model. For example, media assets of the movie type may have one corresponding rating model, and media assets of the TV drama type may have another corresponding rating model. A plurality of to-be-recommended media assets of the same type may be rated using the same rating model.

[0041] It should be noted that the FM model includes an FM algorithm, which is a matrix factorization-based machine learning algorithm suitable for recommendation systems in highly sparse data scenarios. FM considers the interaction between features and performs cross-combination on the features.

[0042] To express the interrelationship between features, a polynomial is added on the basis of the Logistic Regression (LR) linear model, for example, the i-th feature X in the media asset data element i and the j-th feature X j are combined and represented by X i X j , and the model function is rewritten as: .

[0043] This model is the POLY2 model (Degree-2 Polynomial Margin), where n represents the number of features of one sample, and pairwise interactions can obtain cross terms; ω ij is the weight corresponding to the feature combination, used for characterizing the importance of this pair of feature combinations, and the parameters to be learned by the polynomial are ω coefficients.

[0044] Due to the widespread use of One-hot Encoding in machine learning, such brute-force combination increases the complexity from n to n 2 . At the same time, when there is a data sparsity problem, the parameters will be updated only when X i and X j are both non-zero, which leads to difficulty in learning ω ij .

[0045] To solve this problem, the FM algorithm integrates the idea of matrix factorization on the basis of the POLY2 model, and for each feature component X i , introduces a k-dimensional (k<<n) auxiliary vector (a total of n auxiliary vectors for n features): .

[0046] FM learns a one-dimensional vector of size k for each feature, where X is a pair of features. i and X j The weight values ​​of the feature combination can be obtained through the feature vector V. i and V j Represented by the inner product: .

[0047] function The formula for calculating FM can be rewritten as follows: .

[0048] The time complexity of the current formula is O(kn). 2 Further simplification of the second-order crossover feature reduces the time complexity to O(kn). The optimized FM calculation formula is rewritten as follows: .

[0049] Taking the partial derivative of the current formula, we can calculate the gradient of the FM model with respect to the parameters: When the parameter is ω0, we have: .

[0050] When the parameter is ω i At that time, there were: .

[0051] When the parameter is v if At that time, there were: .

[0052] Therefore, the gradient of the FM model with respect to the parameters is: .

[0053] The loss function is the squared difference loss function, which is solved using gradient descent. .

[0054] The gradient of the squared loss function is: .

[0055] This application can use "Channel Navigation Bar - TV Series" as an example to illustrate the application of the actual algorithm: Set up a scheduled task: Write a Kettle script to periodically output the latest ranking results based on the following criteria, ensuring continuous updates to the recommendation results. Kettle scripts are data processing tools that can perform complex data integration tasks in a simple and intuitive way.

[0056] The time interval for scheduled tasks is set at 168 hours to ensure that the latest media assets are included in the sorting process.

[0057] The target data pool is the media assets entered into the database in the last three months, ensuring the timeliness of the output results.

[0058] Furthermore, based on a unit of time, such as 168 hours, the media asset with the highest click-through rate is used as the benchmark, and the output value of that media asset calculated with the initial weight is used as the true value of the benchmark.

[0059] Because some features lack sufficient correlation, training all features uniformly would result in predictive values ​​with limited reference value due to the weak correlation between pairs of features. Therefore, all features are first divided into three groups based on their attributes: quality, size, and efficiency. Then, a set of weight coefficients is determined for each group separately.

[0060] Since some features are discrete features, they need to be encoded using one-hot encoding. For example, video clarity in the quality category has values ​​of standard definition, high definition, and ultra-high definition, and their corresponding one-hot encodings are (1,0,0), (0,2,0), and (0,0,3).

[0061] Furthermore, based on historical media asset rating experience and the attributes held by the media assets used as benchmarks, initial weights are set for the FM formula. Taking the quality category group as an example, the initial weights are set with experience and benchmarks as references. The initial weight settings are shown in Table 1 below: .

[0062] Furthermore, within a single group, by substituting each feature value, calculations are performed and the results are obtained. .

[0063] Furthermore, gradient descent iterations are required, specifically: For each sample, we have: .

[0064] Features ,have: .

[0065] right ,have: .

[0066] Import media asset data from the database for training, observe the loss function curve to determine convergence, and define the convergence threshold of the loss function as 0.005. If the convergence threshold cannot be reached, then training all samples in the database is considered the convergence condition.

[0067] Therefore, we can obtain the scoring models corresponding to the various key data elements of media assets.

[0068] Therefore, for each media asset to be recommended, the media asset data elements of media asset quality, media asset volume, and media asset benefit are input into the first scoring model, the second scoring model, and the third scoring model, respectively, to obtain the media asset quality score, media asset volume score, and media asset benefit score output by the first scoring model, the second scoring model, and the third scoring model, respectively.

[0069] For example, the media asset quality data elements of the first media asset to be recommended are input into the first scoring model to obtain the media asset quality score output by the first scoring model; the media asset volume data elements of the first media asset to be recommended are input into the second scoring model to obtain the media asset volume score output by the second scoring model; the media asset effectiveness data elements of the first media asset to be recommended are input into the third scoring model to obtain the media asset effectiveness score output by the third scoring model. Based on the same operation, the media asset quality score, media asset volume score, and media asset effectiveness score of the remaining media assets to be recommended are obtained.

[0070] For example, the quality-related data elements, benefit-related data elements, and volume-related data elements of media assets A, B, C, and D are input into the corresponding scoring models (where media asset A is determined as the best media asset based on its click-through rate, and the best media asset can also be referred to as the benchmark media asset in the following discussion), thereby obtaining the corresponding scores of each type of media asset data element in each media asset as shown in Table 2 below: .

[0071] Furthermore, based on the media asset quality score, media asset volume score, and media asset benefit score corresponding to each media asset to be recommended, the Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) can be used to conduct a distance analysis of the distance between the superior and inferior solutions, thereby obtaining the media asset score of each media asset to be recommended.

[0072] Furthermore, each media asset can be ranked according to its score, and the higher-ranked media assets will be displayed on the channel navigation page as needed, thus completing the media asset recommendation.

[0073] The media asset recommendation method provided in this application divides the media asset data elements of each media asset to be recommended into media asset quality, media asset volume, and media asset benefit categories. Since a first scoring model, a second scoring model, and a third scoring model are obtained by pre-training a factorization machine model based on the sample media asset data elements and their rating labels for each of the three categories, a first scoring model, a second scoring model, and a third scoring model can be obtained. Then, for each media asset to be recommended, the media asset data elements of the three categories can be input into the first, second, and third scoring models respectively, resulting in accurate media asset quality scores, media asset volume scores, and media asset benefit scores output by the first, second, and third scoring models. Furthermore, a distance analysis of the superior and inferior solutions can be performed based on the media asset quality scores, media asset volume scores, and media benefit scores corresponding to each media asset to be recommended, obtaining accurate media asset scores for each media asset to be recommended. This allows for media asset recommendations based on the scores of each media asset. By employing user-side media asset data elements and categorizing them into media asset quality, media asset volume, and media asset effectiveness, a scoring model trained on a factorization machine model can be used to score each type of media asset data element. Furthermore, a distance analysis of the superior and inferior solutions is performed based on each score, and media asset recommendations are made for each media asset to be recommended based on the scores obtained from the analysis. This enables accurate media asset recommendations through a scientific, reasonable, and impartial approach, effectively improving the accuracy of media asset recommendations.

[0074] Furthermore, the latest ranking results are dynamically and periodically output. The calibration benchmark is reselected every unit of time to ensure dynamic updates; the ranking results are re-output every unit of time to ensure coverage of all newly entered media assets within that unit of time; and data from the past three months is used as the target data pool to ensure the timeliness of the recommendation results.

[0075] In one embodiment, a distance analysis of superior and inferior solutions is performed based on the media asset quality score, media asset volume score, and media asset benefit score corresponding to each media asset to be recommended, to obtain the media asset score of each media asset to be recommended. This includes: determining a benchmark media asset from among the media assets to be recommended based on the media asset indicators of each media asset to be recommended; performing positive index processing on the media asset quality score, media asset volume score, and media asset benefit score corresponding to each media asset to be recommended based on the media asset quality score, media asset volume score, and media asset benefit score of the benchmark media asset, to obtain a positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each media asset to be recommended; and performing a distance analysis of superior and inferior solutions based on the positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each media asset to be recommended, to obtain the media asset score of each media asset to be recommended.

[0076] Furthermore, based on the positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each media asset to be recommended, a distance analysis of superior and inferior solutions is performed to obtain the media asset score of each media asset to be recommended. This includes: constructing a positive matrix by taking the positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each media asset to be recommended as a row of a matrix; standardizing the positive matrix to obtain a standardized matrix; and performing a distance analysis of superior and inferior solutions based on the standardized matrix to obtain the media asset score of each media asset to be recommended.

[0077] Furthermore, a distance analysis of superior and inferior solutions is performed based on the standardized matrix to obtain the media asset score for each media asset to be recommended. This includes: determining the maximum and minimum values ​​for each element of the standardized matrix; where the maximum value includes the element with the largest value in each column of the standardized matrix; the minimum value includes the element with the smallest value in each column of the standardized matrix; determining the first distance between the corresponding media asset and the maximum value based on the elements and the maximum value corresponding to each media asset to be recommended in the standardized matrix; determining the second distance between the corresponding media asset and the minimum value based on the elements and the minimum value corresponding to each media asset to be recommended in the standardized matrix; and determining the media asset score for each media asset to be recommended based on the first and second distances.

[0078] Furthermore, based on the first distance and the second distance corresponding to each media asset to be recommended, the media asset score of the corresponding media asset to be recommended is determined, including: based on the first distance and the second distance corresponding to each media asset to be recommended, the unnormalized score of the corresponding media asset to be recommended is determined; based on each unnormalized score, the normalized score of each media asset to be recommended is determined; and the normalized score of each media asset to be recommended is determined as the media asset score of the corresponding media asset to be recommended.

[0079] Furthermore, when determining the unnormalized score of each media asset to be recommended based on the first distance and the second distance, the following steps are performed for each media asset to be recommended: determine the sum of the first distance and the second distance of the current media asset to be recommended; and determine the ratio of the second distance of the current media asset to the sum as the unnormalized score of the current media asset to be recommended.

[0080] Furthermore, when determining the normalized score of each media asset to be recommended based on each unnormalized score, the following steps are performed for each media asset to be recommended: determine the sum of each unnormalized score; and determine the ratio of the unnormalized score of the current media asset to the sum as the normalized score of the current media asset to be recommended.

[0081] Specifically, in this application, the media asset quality score, media asset volume score, and media asset effectiveness score corresponding to each recommended media asset can be standardized by converting all indicators into extremely large values, i.e., positiveizing the indicators. The output scores of the three groups in this application can all be defined as intermediate indicators; the closer the result is to the value of the optimal media asset, the better.

[0082] More specifically, a benchmark media asset can be determined from among the media assets to be recommended based on their media asset metrics. For example, the media asset with the highest click-through rate can be determined as the benchmark media asset.

[0083] Furthermore, the indicator can be positively processed according to the following formula: ; in, This represents the score of the j-th type of media asset data element of the i-th media asset to be recommended. This represents the score of the j-th type of media asset data element representing the best media asset. This represents the positive rating of the j-th type of media asset data element for the i-th media asset to be recommended.

[0084] Therefore, we can obtain the positive media asset quality score, positive media asset volume score, and positive media asset benefit score for each media asset to be recommended.

[0085] Furthermore, the positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each recommended media asset can be used as rows of a matrix to construct a positive matrix. For example, the following positive matrix can be constructed: ; The matrix above is formed by using the positive media asset quality score, positive media asset volume score, and positive media asset benefit score of n media assets to be recommended as matrix elements.

[0086] To eliminate the influence of different indicator dimensions, the already positiveized matrix needs to be standardized.

[0087] Specifically, each element in the normalized matrix can be standardized using the following formula: ; Thus, we obtain the standardized matrix composed of the normalized elements: ; Furthermore, the maximum value can be determined based on the elements of the normalized matrix using the following formula: Z + =(Z1 + Z2 + Z3+ )=(max{z 11 , z 21 , ..., z n1}, max{z 12 , z 22 , ..., z n2}, max{z 13 , z 23 , ..., z n3}).

[0088] Furthermore, the minimum value is determined by combining the elements of the standardized matrix using the following formula: Z - =(Z1 - Z2 - Z3 - )=(min{z 11 , z 21 , ..., z n1}, min{z 12 , z 22 , ..., z n 2},min{z 13 , z 23 , ..., z n3}).

[0089] Furthermore, the first distance between each media asset to be recommended and the maximum value can be determined based on the following formula: .

[0090] Furthermore, the second distance between each media asset to be recommended and the minimum value is determined based on the following formula: .

[0091] Furthermore, based on the first and second distances corresponding to each media asset to be recommended, the unnormalized score of the corresponding media asset to be recommended can be determined using the following formula: .

[0092] Furthermore, the normalized score for each media asset to be recommended can be determined based on its unnormalized score, using the following formula: .

[0093] Furthermore, the normalized score of each media asset to be recommended can be determined as the media asset score of the corresponding media asset to be recommended.

[0094] Since the results closer to the optimal value are evaluated higher in this application, the TOPSIS algorithm is used to sort the multi-group results of the FM algorithm in an intermediate calibration manner, generating a unique ranking after comprehensive evaluation of each group, so as to realize the comprehensive ranking results of media assets under each media asset category, support the bureau's subsequent display decisions, and thus achieve accurate media asset recommendation through a scientific, reasonable and fair method, effectively improving the accuracy of media asset recommendation.

[0095] Figure 2 This is a schematic diagram illustrating the application scenario of the media asset recommendation method provided in this application, such as... Figure 2 As shown, in one embodiment, media asset data (specifically, key data elements) are divided into three groups based on attributes: media asset quality, media asset volume, and media asset effectiveness. Further, for a specific channel navigation bar category, the FM algorithm is used to calculate a corresponding output value for each group. Then, TOPSIS is used to sort the output values ​​of each media asset, thereby obtaining the ranking result of the media asset recommendations for the homepage content of a specific channel navigation bar.

[0096] In the current field of media asset management platforms, where there is a lack of quantifiable evaluation mechanisms for content recommendation in the large-screen channel navigation bar homepage, this application innovatively introduces a quantitative mechanism using the FM+TOPSIS algorithm combination to scientifically, reasonably, and fairly score and rank all media assets under each media asset category, effectively solving the problem of how to prioritize the display of the best media assets on the homepage.

[0097] In the current field of content recommendation for large-screen channel navigation bars, the lack of sufficient user-side data prevents the training of models from the user's perspective. This application fully leverages the advantages of channel-side media assets, which are abundant and have diverse attributes, and takes a different approach by using the attributes of the media assets themselves as feature values ​​required for the FM algorithm, effectively solving the problem of insufficient user-side data.

[0098] This paper optimizes the FM algorithm for use cases where media asset attributes are used as features. Traditional methods combine all features in pairs within a group. However, since the features in this application have different attribute categories, pairwise combinations of features from different attribute categories do not perform well. Therefore, this application innovatively divides media assets into three feature groups based on their meaning attributes: quality, efficiency, and volume. The FM algorithm is trained separately for each group to obtain the optimal model for each group.

[0099] Because the optimized FM algorithm is used, multiple grouping results are generated. Within each group, results closer to the optimal value receive higher evaluations, necessitating a comprehensive ranking method that outputs a unique conclusion. Therefore, this application innovatively combines FM and TOPSIS. Since results closer to the optimal value receive higher evaluations, the TOPSIS algorithm is used to sort the multiple grouping results from the FM algorithm using an intermediate calibration method, generating a unique ranking after comprehensive evaluation of each group. This achieves a comprehensive ranking result for media assets under each category, supporting the authority's subsequent display decisions.

[0100] The media asset recommendation device provided in this application is described below. The media asset recommendation device described below can be referred to in correspondence with the media asset recommendation method described above.

[0101] Furthermore, this application also provides a media asset recommendation device.

[0102] The media asset recommendation device may also include: a segmentation module, used to segment the media asset data elements of each media asset to be recommended into media asset quality category, media asset volume category, and media asset benefit category; an input module, used to input the media asset data elements of the media asset quality category, media asset volume category, and media asset benefit category into a first scoring model, a second scoring model, and a third scoring model for each media asset to be recommended, to obtain the media asset quality score, media asset volume score, and media asset benefit score output by the first scoring model, the second scoring model, and the third scoring model, respectively; the first scoring model, the second scoring model, and the third scoring model are obtained by training the factorization machine model with the sample media asset data elements and their scoring labels for each of the media asset quality category, media asset volume category, and media benefit category of the sample media assets; an analysis module, used to perform a distance analysis of the superior and inferior solutions based on the media asset quality score, media asset volume score, and media benefit score corresponding to each media asset to be recommended, to obtain the media asset score of each media asset to be recommended; and a determination module, used to recommend each media asset to be recommended based on the media asset score.

[0103] In one embodiment, the analysis module is specifically used to: determine benchmark media assets from among the media assets to be recommended based on the media asset indicators of each media asset to be recommended; perform positive index processing on the media asset quality score, media asset volume score, and media asset benefit score corresponding to each media asset to be recommended based on the media asset quality score, media asset volume score, and media asset benefit score of the benchmark media asset, to obtain the positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each media asset to be recommended; and perform a distance analysis of superior and inferior solutions based on the positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each media asset to be recommended, to obtain the media asset score of each media asset to be recommended.

[0104] In one embodiment, the analysis module includes a first analysis unit, which is configured to: construct a positive matrix by taking the positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each media asset to be recommended as a row of a matrix; perform standardization processing on the positive matrix to obtain a standardized matrix; and perform superior-inferiority distance analysis based on the standardized matrix to obtain the media asset score of each media asset to be recommended.

[0105] In one embodiment, the analysis module includes a second analysis unit, which is configured to: determine a maximum value and a minimum value based on each element of the normalization matrix; wherein the maximum value includes the element with the largest value in each column of the normalization matrix; the minimum value includes the element with the smallest value in each column of the normalization matrix; determine a first distance between the corresponding media asset to be recommended and the maximum value based on the elements corresponding to each media asset to be recommended in the normalization matrix; determine a second distance between the corresponding media asset to be recommended and the minimum value based on the elements corresponding to each media asset to be recommended in the normalization matrix; and determine a media asset score for each media asset to be recommended based on the first distance and the second distance.

[0106] In one embodiment, the analysis module includes a first determining unit, which is configured to: determine the unnormalized score of the corresponding media asset to be recommended based on the first distance and the second distance corresponding to each media asset to be recommended; determine the normalized score of each media asset to be recommended based on each unnormalized score; and determine the normalized score of each media asset to be recommended as the media asset score of the corresponding media asset to be recommended.

[0107] In one embodiment, the analysis module includes a second determining unit, which is used to perform the following steps for each media asset to be recommended when determining the unnormalized score of the corresponding media asset based on the first distance and the second distance corresponding to each media asset to be recommended: determining the sum of the first distance and the second distance of the current media asset to be recommended; and determining the ratio of the second distance of the current media asset to the sum as the unnormalized score of the current media asset to be recommended.

[0108] In one embodiment, the analysis module includes a third determining unit, which is used to perform the following steps for each media asset to be recommended when determining the normalized score of each media asset to be recommended based on each unnormalized score: determining the sum of each unnormalized score; and determining the ratio of the unnormalized score of the current media asset to be recommended to the sum of the unnormalized scores as the normalized score of the current media asset to be recommended.

[0109] The media asset recommendation device provided in this application divides the media asset data elements of each media asset to be recommended into media asset quality, media asset volume, and media asset benefit categories. Since a first scoring model, a second scoring model, and a third scoring model are obtained by pre-training a factorization machine model based on the sample media asset data elements and their rating labels for each of the three categories, the media asset data elements of each category can be input into the first, second, and third scoring models for each media asset to be recommended. This yields accurate media asset quality scores, media asset volume scores, and media asset benefit scores output by the first, second, and third scoring models, respectively. Furthermore, a distance analysis of the superior and inferior solutions can be performed based on the media asset quality scores, media asset volume scores, and media benefit scores corresponding to each media asset to be recommended, resulting in accurate media asset scores for each media asset to be recommended. This allows for media asset recommendations based on the scores of each media asset. By employing user-side media asset data elements and categorizing them into media asset quality, media asset volume, and media asset effectiveness, a scoring model trained on a factorization machine model can be used to score each type of media asset data element. Furthermore, a distance analysis of the superior and inferior solutions is performed based on each score, and media asset recommendations are made for each media asset to be recommended based on the scores obtained from the analysis. This enables accurate media asset recommendations through a scientific, reasonable, and impartial approach, effectively improving the accuracy of media asset recommendations.

[0110] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330, and a communication bus 340, wherein the processor 310, the communication interface 320, and the memory 330 communicate with each other through the communication bus 340. The processor 310 can call logic instructions in the memory 330 to execute a media asset recommendation method. This method includes: dividing the media asset data elements of each media asset to be recommended into media asset quality, media asset volume, and media asset benefit categories; for each media asset to be recommended, inputting the media asset data elements of the media asset quality, media asset volume, and media asset benefit categories into a first scoring model, a second scoring model, and a third scoring model, respectively, to obtain the media asset quality score, media asset volume score, and media asset benefit score output by the first scoring model, the second scoring model, and the third scoring model, respectively; the first scoring model, the second scoring model, and the third scoring model are obtained by training a factorization machine model based on the sample media asset data elements and their scoring labels for each of the media asset quality, media asset volume, and media benefit categories of the sample media assets; performing a distance analysis of the superior and inferior solutions based on the media asset quality score, media asset volume score, and media benefit score corresponding to each media asset to be recommended, to obtain the media asset score for each media asset to be recommended; and recommending media assets based on each media asset score.

[0111] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a 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 a 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 a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0112] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the media asset recommendation method provided by the above methods. The method includes: dividing the media asset data elements of each media asset to be recommended into media asset quality category, media asset volume category, and media asset benefit category; for each media asset to be recommended, inputting the media asset data elements of the media asset quality category, media asset volume category, and media asset benefit category into a first scoring model, a second scoring model, and a third scoring model, respectively, to obtain a first... The scoring model, the second scoring model, and the third scoring model output media asset quality scores, media asset volume scores, and media asset benefit scores, respectively. The first, second, and third scoring models are obtained by training the factor decomposition machine model with sample media asset data elements and their scoring labels for each of the media asset quality, media asset volume, and media asset benefit categories of the sample media assets. Based on the media asset quality scores, media asset volume scores, and media benefit scores corresponding to each media asset to be recommended, a distance analysis of the superior and inferior solutions is performed to obtain the media asset score of each media asset to be recommended. Media asset recommendations are made for each media asset to be recommended based on the media asset scores.

[0113] Furthermore, this application also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, this computer program is implemented to perform the media asset recommendation method provided by the methods described above. This method includes: classifying the media asset data elements of each media asset to be recommended into media asset quality, media asset volume, and media asset benefit categories; for each media asset to be recommended, inputting the media asset data elements of the media asset quality, media asset volume, and media asset benefit categories into a first scoring model, a second scoring model, and a third scoring model, respectively, to obtain the first scoring model, the second scoring model, and the third scoring model. The scoring models output media asset quality scores, media asset volume scores, and media asset benefit scores, respectively. The first, second, and third scoring models are obtained by training the factor decomposition machine model with sample media asset data elements and their scoring labels for each of the three categories of media asset quality, media asset volume, and media asset benefit. Based on the media asset quality scores, media asset volume scores, and media benefit scores corresponding to each media asset to be recommended, a distance analysis of the superior and inferior solutions is performed to obtain the media asset score of each media asset to be recommended. Media asset recommendations are made for each media asset to be recommended based on the media asset scores.

[0114] The device embodiments described above are merely illustrative. 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0115] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A media asset recommendation method, characterized in that, The method includes: The media asset data elements of each media asset to be recommended are divided into media asset quality, media asset volume, and media asset effectiveness categories; For each media asset to be recommended, media asset data elements of media asset quality, media asset volume, and media asset effectiveness are input into the first scoring model, the second scoring model, and the third scoring model, respectively, to obtain the media asset quality score, media asset volume score, and media asset effectiveness score output by the first scoring model, the second scoring model, and the third scoring model, respectively. The first scoring model, the second scoring model, and the third scoring model are obtained by training the factorization machine model based on the sample media asset data elements and their scoring labels for each of the media asset quality, media asset volume, and media effectiveness categories of the sample media assets. Based on the media asset indicators of each media asset to be recommended, a benchmark media asset is determined from all media assets to be recommended; Based on the media asset quality score, media asset volume score, and media asset benefit score of the benchmark media asset, the media asset quality score, media asset volume score, and media asset benefit score corresponding to each media asset to be recommended are respectively positively processed to obtain the positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each media asset to be recommended. Construct a positive matrix by taking the positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each media asset to be recommended as a row of a matrix; The normalized matrix is ​​standardized to obtain the standardized matrix; The maximum and minimum values ​​are determined based on each element of the normalized matrix; wherein the maximum value includes the element with the largest value in each column of the normalized matrix; and the minimum value includes the element with the smallest value in each column of the normalized matrix. Based on the elements corresponding to each media asset to be recommended in the standardized matrix and the maximum value, the first distance between the corresponding media asset to be recommended and the maximum value is determined respectively; Based on the elements corresponding to each media asset to be recommended in the standardized matrix and the minimum value, the second distance between the corresponding media asset to be recommended and the minimum value is determined respectively. Based on the first distance and the second distance corresponding to each media asset to be recommended, the media asset score of the corresponding media asset to be recommended is determined; Media assets are recommended based on their scores.

2. The media asset recommendation method according to claim 1, characterized in that, The process of determining the media asset score for each media asset to be recommended based on the first distance and the second distance includes: Based on the first distance and the second distance corresponding to each media asset to be recommended, the unnormalized score of the corresponding media asset to be recommended is determined; Based on the unnormalized scores, the normalized score of each media asset to be recommended is determined separately. The normalized score of each media asset to be recommended is determined as the media asset score of the corresponding media asset to be recommended.

3. The media asset recommendation method according to claim 2, characterized in that, When determining the unnormalized score of each media asset to be recommended based on the first distance and the second distance, the following steps are performed for each media asset to be recommended: Determine the sum of the first and second distances of the media asset to be recommended; The ratio of the second distance of the current media asset to be recommended to the sum of the sums is determined as the unnormalized score of the current media asset to be recommended.

4. The media asset recommendation method according to claim 2, characterized in that, When determining the normalized score for each media asset to be recommended based on its unnormalized scores, the following steps are performed for each media asset to be recommended: Determine the sum of the unnormalized scores; The ratio of the unnormalized score of the current media asset to be recommended to the summed value is determined as the normalized score of the current media asset to be recommended.

5. A media asset recommendation device, characterized in that, include: The segmentation module is used to classify the media asset data elements of each media asset to be recommended into media asset quality, media asset volume, and media asset effectiveness categories. The input module is used to input media asset data elements of media asset quality, media asset volume, and media asset benefit categories into the first scoring model, the second scoring model, and the third scoring model respectively for each media asset to be recommended, so as to obtain the media asset quality score, media asset volume score, and media asset benefit score output by the first scoring model, the second scoring model, and the third scoring model respectively; the first scoring model, the second scoring model, and the third scoring model are obtained by training the factorization machine model based on the sample media asset data elements and their scoring labels in each of the media asset quality, media asset volume, and media benefit categories of the sample media assets respectively. The analysis module is used to determine benchmark media assets from among the media assets to be recommended based on their media asset indicators; based on the media asset quality score, media asset volume score, and media asset benefit score of the benchmark media assets, the module performs positive index processing on the media asset quality score, media asset volume score, and media asset benefit score corresponding to each media asset to be recommended, obtaining positive media asset quality score, positive media asset volume score, and positive media asset benefit score for each media asset to be recommended; constructing a positive matrix by taking each of the positive media asset quality score, positive media asset volume score, and positive media asset benefit score corresponding to each media asset to be recommended as a row; and standardizing the positive matrix to obtain a standard... The standardization matrix is ​​used to determine the maximum and minimum values ​​for each element. The maximum value includes the element with the largest value in each column of the standardization matrix, and the minimum value includes the element with the smallest value in each column of the standardization matrix. Based on the elements corresponding to each media asset to be recommended in the standardization matrix and the maximum value, a first distance between the media asset to be recommended and the maximum value is determined. Based on the elements corresponding to each media asset to be recommended in the standardization matrix and the minimum value, a second distance between the media asset to be recommended and the minimum value is determined. Based on the first and second distances corresponding to each media asset to be recommended, a media asset score is determined. The determination module is used to recommend media assets based on their scores.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the media asset recommendation method as described in any one of claims 1 to 4.

7. A medium, said medium being a non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the media asset recommendation method as described in any one of claims 1 to 4.

8. A product, said product being a computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the media asset recommendation method as described in any one of claims 1 to 4.

Images