Video marketing recommendation method and system, medium and program product

Abstract

The invention provides a video marketing recommendation method and system, a medium and a program product, and relates to the technical field of intelligent marketing, and the method comprises the steps: taking a parameter vector set corresponding to a historical consumption feature set of a to-be-promoted user as a population, carrying out the individual selection of the population, and obtaining the features of the to-be-promoted user from the historical consumption feature set; according to a to-be-predicted marketing feature set and a consumption intention prediction model, a predicted value of the influence degree of the consumption intention of each to-be-predicted marketing feature in the to-be-predicted marketing feature set is obtained, and the to-be-predicted marketing feature set comprises historical marketing success features and to-be-promoted user features; the consumption intention prediction model is used for predicting the influence degree of the marketing features on the consumption intention of the user; according to the predicted value, obtaining a target marketing feature for the to-be-promoted user from a to-be-predicted marketing feature set; and generating a marketing video for the to-be-promoted user according to the target marketing feature, thereby realizing accurate marketing recommendation for the user, and improving the recommendation success rate.

Description

Technical Field

[0001] This application relates to the field of intelligent marketing technology, specifically to a video marketing recommendation method, system, medium, and program product. Background Technology

[0002] Currently, product and service marketing typically involves push notifications via SMS or pop-ups to promote products and services, or short videos or advertisements broadcast on television or other media platforms. However, these marketing methods cannot provide targeted recommendations based on user needs, and the production cycle for short videos or advertisements is long and costly, making timely adjustments difficult. Consequently, existing marketing promotion technologies suffer from low recommendation success rates. Summary of the Invention

[0003] At least one embodiment of this application provides a video marketing recommendation method, system, medium, and program product to solve the problem of low recommendation success rate in existing marketing promotion technologies.

[0004] To solve the above-mentioned technical problems, this application is implemented as follows:

[0005] In a first aspect, embodiments of this application provide a video marketing recommendation method, including:

[0006] By using the set of parameter vectors corresponding to the set of historical consumption characteristics of users to be promoted as a population, individual selection is performed on the population to obtain the characteristics of users to be promoted from the set of historical consumption characteristics.

[0007] Based on the set of marketing features to be predicted and the consumption intention prediction model, the predicted value of the influence of each marketing feature to be predicted in the set of marketing features to be predicted is obtained. The set of marketing features to be predicted includes historical marketing success features and the user features to be promoted. The consumption intention prediction model is used to predict the influence of marketing features on user consumption intention.

[0008] Based on the predicted value, the target marketing features for the users to be promoted are obtained from the set of marketing features to be predicted;

[0009] Based on the target marketing characteristics, generate marketing videos for the users to be promoted.

[0010] Optionally, the video marketing recommendation method, wherein the parameter vector set corresponding to the historical consumption feature set of the users to be promoted is used as a population, and individual selection is performed on the population to obtain the features of the users to be promoted from the historical consumption feature set, includes:

[0011] For each individual in the population, obtain the accuracy parameter and the area under the curve parameter corresponding to that individual;

[0012] The fitness value of an individual is obtained based on its accuracy parameter and area under the curve parameter.

[0013] Individuals are selected from the population based on their fitness values, and user characteristics to be promoted are obtained from the set of historical consumption characteristics.

[0014] Optionally, the video marketing recommendation method, wherein selecting individuals from the population based on their fitness values ​​and obtaining user characteristics to be promoted from the historical consumption characteristic set includes:

[0015] For every three different individuals in the population, based on the fitness value of the individuals, the first individual is used as the mutation baseline vector, and the difference between the second and third individuals is used as the difference vector to obtain the mutation vector, with the fitness values ​​of the first individual, the second individual, and the third individual increasing sequentially.

[0016] The mutation vector is recombined with the individual to obtain the experimental vector;

[0017] Individual selection is performed based on the fitness values ​​of the experimental vector and the individual to obtain the optimal individual;

[0018] The historical consumption characteristics corresponding to the optimal individual are used as the characteristics of users to be promoted.

[0019] Optionally, the video marketing recommendation method further includes:

[0020] Obtain the first distribution divergence between the first user group to be promoted and the second user group to be promoted;

[0021] For each of the marketing features to be predicted in the set of marketing features to be predicted, the set of marketing features to be predicted is divided into two subsets according to the marketing features to be predicted, and the second distribution divergence after splitting with the marketing features to be predicted as nodes is obtained.

[0022] Obtain the first marketing feature to be predicted from the set of marketing features to be predicted that has the largest difference between the second distribution divergence and the first distribution divergence, and divide the set of marketing features to be predicted into two subsets based on the first marketing feature to be predicted;

[0023] The step of obtaining the first marketing feature to be predicted is repeated for each subset until the set can no longer be divided, thereby generating the consumer intention prediction model.

[0024] Optionally, in the video marketing recommendation method, obtaining the first distribution divergence between the first user group to be promoted and the second user group to be promoted includes:

[0025] A marketing video including the predicted marketing features is sent to the users in the first user group to be promoted, and a first result is obtained as to whether the users in the first user group to be promoted make a purchase.

[0026] Marketing videos including the predicted marketing features are not sent to the users in the second user group to be promoted, and a second result is obtained as to whether the users in the second user group to be promoted have made a purchase.

[0027] Based on the first result and the second result, obtain the types of users to be promoted in the first user group to be promoted and the second user group to be promoted.

[0028] Based on the type of users to be promoted, obtain the first distribution divergence between the first user group to be promoted and the second user group to be promoted.

[0029] Optionally, the video marketing recommendation method further includes:

[0030] Obtain historical consumption data from users on the sales platform;

[0031] Based on the historical consumption data, the users of the sales platform are divided into users to be promoted or users who have successfully been marketed.

[0032] For the users who have successfully engaged in marketing, obtain the characteristics of their historical marketing success.

[0033] Optionally, the video marketing recommendation method, wherein obtaining target marketing features for the user to be promoted from the set of marketing features to be predicted based on the predicted value includes:

[0034] The marketing features to be predicted in the set of marketing features to be predicted are sorted in descending order of the predicted values ​​to obtain the sorted set of marketing features to be predicted.

[0035] A preset number of target marketing features for the users to be promoted are sequentially obtained from the sorted set of marketing features to be predicted.

[0036] Optionally, the video marketing recommendation method, wherein generating a marketing video for the user to be promoted based on the target marketing characteristics includes:

[0037] Based on the target marketing characteristics, generate marketing text for the users to be promoted;

[0038] Based on the marketing text and the text-based video model, generate the marketing video for the user to be promoted.

[0039] Secondly, embodiments of this application also provide a video marketing recommendation system, including:

[0040] The selection module is used to select individuals from the set of parameter vectors corresponding to the set of historical consumption characteristics of users to be promoted, and obtain the characteristics of users to be promoted from the set of historical consumption characteristics.

[0041] The prediction module is used to obtain the predicted value of the influence of each marketing feature in the marketing feature set to be predicted on the consumer intention based on the marketing feature set to be predicted and the consumer intention prediction model. The marketing feature set to be predicted includes historical marketing success features and the user features to be promoted. The consumer intention prediction model is used to predict the influence of marketing features on user consumption intention.

[0042] The acquisition module is used to acquire target marketing features for the user to be promoted from the set of marketing features to be predicted based on the predicted value.

[0043] The generation module is used to generate marketing videos for the users to be promoted based on the target marketing characteristics.

[0044] Thirdly, embodiments of this application also provide a video marketing recommendation system, including: a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the processor executes the program or instructions to implement the video marketing recommendation method as described in the first aspect.

[0045] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the video marketing recommendation method as described in the first aspect.

[0046] Fifthly, embodiments of this application also provide a computer program product, including computer instructions, which, when executed by a processor, implement the video marketing recommendation method as described in the first aspect.

[0047] Compared with existing technologies, this application provides a video marketing recommendation method, system, medium, and program product. The method includes: selecting individuals from a population of parameter vectors corresponding to a set of historical consumption characteristics of users to be promoted; obtaining user characteristics from the historical consumption characteristic set; obtaining a predicted value of the influence of each marketing characteristic in the marketing characteristic set to be predicted on user consumption intention based on a set of marketing characteristics to be predicted and a consumption intention prediction model, wherein the marketing characteristic set to be predicted includes historical marketing success characteristics and the user characteristics to be promoted, and the consumption intention prediction model is used to predict the influence of marketing characteristics on user consumption intention; obtaining target marketing characteristics for the user to be promoted from the marketing characteristic set to be predicted based on the predicted value; and generating a marketing video for the user to be promoted based on the target marketing characteristics. Because the marketing characteristics most influential on user consumption behavior are selected for generating the marketing video, the recommendation success rate of the marketing video is greatly improved, achieving precise marketing to different user groups. Attached Figure Description

[0048] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0049] Figure 1 This is a flowchart illustrating the video marketing recommendation method described in an embodiment of this application;

[0050] Figure 2 This is a flowchart illustrating one embodiment of the video marketing recommendation method described in this application.

[0051] Figure 3 This is a schematic diagram of the structure of the video marketing recommendation system described in the embodiments of this application;

[0052] Figure 4 This is a schematic diagram of one embodiment of the video marketing recommendation system described in this application.

[0053] Figure 5 This is a hardware block diagram of the video marketing recommendation system described in an embodiment of this application. Detailed Implementation

[0054] 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 terms can be used interchangeably 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" and "second" are generally of the same class, without limiting the number of objects; for example, the first object can be one or more. Furthermore, "or" in this application indicates at least one of the connected objects. For example, "A or B" covers three scenarios: Scenario 1: including A but not B; Scenario 2: including B but not A; Scenario 3: including both A and B. The character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0055] Please refer to Figure 1 This application provides a video marketing recommendation method, including:

[0056] Step 101: By using the set of parameter vectors corresponding to the set of historical consumption characteristics of the users to be promoted as a population, individual selection is performed on the population to obtain the characteristics of the users to be promoted from the set of historical consumption characteristics.

[0057] It is understood that each individual in the population corresponds to a parameter vector, and each parameter vector corresponds to a historical consumption feature of the user to be promoted.

[0058] In one embodiment, optionally, the method further includes:

[0059] Obtain historical consumption data from users on the sales platform;

[0060] Based on the historical consumption data, the users of the sales platform are divided into users to be promoted or users who have successfully been marketed.

[0061] For the users who have successfully engaged in marketing, obtain the characteristics of their historical marketing success.

[0062] In this embodiment of the application, the historical consumption data of the sales platform users are collected. Based on the historical consumption data and the types of products to be promoted, the sales platform users are divided into users to be promoted or users who have successfully been marketed. That is, the sales platform users are grouped, and groups of users to be promoted and groups of users who have successfully been marketed are selected.

[0063] Optionally, the historical consumption data includes, but is not limited to, at least one of the following:

[0064] Account registration date; gender; product name purchased; type; amount spent; purchase date; customer review.

[0065] Specifically, based on the historical consumption data, it is determined whether the sales platform users have consumption records of the same type as the product to be promoted, and then the sales platform users are classified as follows:

[0066] Type a, the users to be promoted:

[0067] Users on the sales platform who do not have a purchase record of the same type as the product to be recommended are classified as users to be promoted.

[0068] In addition, for the sales platform users whose consumption records are of the same type as the products to be recommended, it is necessary to determine the purchase date of the consumption record, further determine the last purchase interval, and classify the sales platform users whose purchase interval is greater than or equal to a preset time interval as the users to be promoted.

[0069] Type b, the successful marketing users:

[0070] Users of the sales platform whose consumption records are of the same type as the product to be recommended are classified as successful marketing users.

[0071] Alternatively, for the sales platform users whose consumption records are of the same type as the product to be recommended, it is necessary to determine the purchase date of the consumption record, further determine the last purchase interval, and classify the sales platform users whose purchase interval is less than a preset time interval as the successful marketing users.

[0072] In one implementation, optionally, before classifying the sales platform users into users to be promoted or successfully marketed users based on the historical consumption data, the method further includes:

[0073] The historical consumption data is preprocessed.

[0074] It should be noted that the processing of the collected historical consumption data includes, but is not limited to, at least one of the following:

[0075] Data cleaning; data labeling; data compression.

[0076] Data cleaning may include, but is not limited to, at least one of the following:

[0077] Deduplication (e.g., deleting duplicate consumption records); handling missing values ​​(e.g., deleting missing values, replacing missing values, or using interpolation methods); handling outliers (e.g., deleting or replacing); handling error values.

[0078] Furthermore, since the historical consumption data is extremely large and the data type is complex, in order to improve the accuracy of subsequent predictions, data annotation can also be performed manually (or automatically by machines) in this embodiment of the application.

[0079] Specifically, in this embodiment of the application, natural language processing methods are used to segment and annotate the data types of the historical consumption data. For example, the historical consumption data can be marked and annotated according to keywords, entities, and grammatical structures; for example, it can be marked and annotated according to the order date, order mobile number, order number, and order payment method in the historical consumption data.

[0080] Finally, in order to reduce the data redundancy of the historical consumption data, in this embodiment of the application, the historical consumption data can also be compressed using a compression algorithm, such as the dynamic Huffman tree algorithm.

[0081] Optionally, for the successful marketing users, the video marketing plans of the successful marketing users are obtained, and the historical marketing success features are extracted from the video marketing plans. The historical marketing success features are used to represent the user's interest points in the marketing recommended videos.

[0082] The video marketing plan may include pre-generated recommended videos. In this embodiment, the video marketing plan can be determined based on the purchase channels of the successful marketing users.

[0083] Specifically, when a successfully marketed user completes an order, the sales system records the purchase channel of that user. For example, if the user made the purchase by clicking a purchase link in video A played on the Douyin app, the sales system will record "Douyin - Video A" as the purchase channel for that user. Furthermore, the video marketing plan can be determined based on this purchase channel.

[0084] It should be noted that the historical marketing success characteristics include the product characteristics highlighted in the video marketing plan, including but not limited to: the target audience characteristics of the product, such as office workers or students; the functional characteristics of the product, such as low energy consumption, fast response speed, and high accuracy; and the user feedback characteristics of the product, such as user reviews.

[0085] Optionally, for the successful marketing users, the video marketing plan of the successful marketing users is obtained, key frames are extracted and voice is extracted from the video marketing plan to obtain the content data of the video marketing plan, and the historical marketing success features are obtained by performing target recognition on the extracted key frames and voice recognition on the extracted voice.

[0086] In one implementation, optionally, the feature of the user to be promoted is obtained by using a set of parameter vectors corresponding to the set of historical consumption features of the users to be promoted as a population, and by performing individual selection on the population, including:

[0087] For each individual in the population, obtain the accuracy parameter and the area under the curve parameter corresponding to that individual;

[0088] The fitness value of an individual is obtained based on its precision parameter and area under the curve (AUC) parameter.

[0089] Individuals are selected from the population based on their fitness values, and user characteristics to be promoted are obtained from the set of historical consumption characteristics.

[0090] In this embodiment of the application, feature extraction is performed based on the historical consumption data of the user to be promoted to obtain the historical consumption feature set, which includes, but is not limited to: order number; user; consumption amount.

[0091] It should be noted that the population is a set of multiple binary parameter vectors. If the set of historical consumption features obtained through feature extraction includes N features, it can be initialized into a parameter vector of 1×N features through an initialization operation. This parameter vector corresponds to the historical consumption features and is an individual of the first generation population, where the value of each parameter vector is 0 or 1.

[0092] Based on the dimension of the feature, set an appropriate population size Ps, repeat the above initialization operation Ps times, and obtain the first generation population with Ps individuals, which is represented by the following formula (1):

[0093]

[0094] Among them, X i (0) represents the first generation population, X i,1 (0),X i,2 (0),…,X i,P (0) represents the parameter vector corresponding to each historical consumption feature in the set of historical consumption features that constitute the first generation population.

[0095] Optionally, for each individual in the population, the accuracy parameter and area under the curve parameter corresponding to that individual are obtained through a decision tree classifier.

[0096] Optionally, the accuracy parameter and area under the curve parameter corresponding to the individual are obtained by performing five-fold cross-validation on the individual.

[0097] Optionally, the fitness value of the individual is obtained based on the precision parameter and the area under the curve (AUC) parameter corresponding to the individual, including:

[0098] The fitness value of an individual is obtained based on the accuracy parameter and the mean of the area under the curve corresponding to that individual.

[0099] Specifically, the fitness value of the individual is determined according to the following formula (2):

[0100]

[0101] Where X refers to the set of historical consumption characteristics identified by binary parameter vectors, i.e., the parameter vector set; precision 5-fold(X) and AUC 5-fold(X) and represent the accuracy parameter and area under the curve parameter obtained by the decision tree classifier through five-fold cross-validation on the individual, respectively.

[0102] In one implementation, optionally, individual selection is performed on the population based on the fitness value of the individuals, and the characteristics of users to be promoted are obtained from the set of historical consumption characteristics, including:

[0103] For every three different individuals in the population, based on the fitness value of the individuals, the first individual is used as the mutation baseline vector, and the difference between the second and third individuals is used as the difference vector to obtain the mutation vector, with the fitness values ​​of the first individual, the second individual, and the third individual increasing sequentially.

[0104] The mutation vector is recombined with the individual to obtain the experimental vector;

[0105] Individual selection is performed based on the fitness values ​​of the experimental vector and the individual to obtain the optimal individual;

[0106] The historical consumption characteristics corresponding to the optimal individual are used as the characteristics of users to be promoted.

[0107] In this embodiment of the application, performing evolutionary operations on the population to obtain the optimal individual includes the following steps:

[0108] Mutation operation: Randomly select three different individuals from the population, denoted as X. a1 (g), X a2 (g) and X a3 (g) Sort the three individuals in order of fitness value from low to high, assuming the sorting result is still X. a1 (g), X a2 (g) and X a3(g)

[0109] Select the first body X a1 (g) serves as the mutation baseline vector, and the second volume X a2 (g) and the third body X a3 The difference X between (g) a2 (g)-X a3 (g) is the difference vector. Then, the mutation operation is performed according to the following formula (3) to obtain the mutation vector (mutated individual):

[0110] v i(g) =X a1 (g)+U·(X a2 (g)-X a3 (g)) (3)

[0111] U is the mutation control factor, which is used to adjust the magnitude of the mutation operation. U can be set as needed, and is generally set to 0.1.

[0112] Correction: The above mutation vector v is modified by the following formula (4). i(g) The algorithm is modified to ensure that each individual is between 0 and 1, thus guaranteeing that each element of the mutated individual is a binary bit, ensuring the correct execution of the algorithm.

[0113]

[0114] Cross operation: By crossing the target vector X ij (g) Recombines with the mutation vector to generate the experimental vector E. i (g). The target vector is the parameter vector of those unmutated individuals selected from the population that need to be crossed with the mutated individuals (mutation vectors).

[0115] Selection operation: By evaluating the fitness values ​​of the experimental vectors obtained from the crossover operation and the original individuals (i.e., the unmutated individuals in the population), the strategy for selecting the next generation is determined, thereby selecting the best individual in the population, so that the population gradually converges until it reaches the global optimum.

[0116] Specifically, the selection is made according to the following formula (5):

[0117]

[0118] Where, the adaptive value is the fitness function, in the experimental vector E i (g) and target vector X ij (g) A greedy selection strategy is used to select the better individual as the new individual, and the selected individual is used as the user characteristics to be promoted. Subsequent analysis and processing are then performed based on these characteristics.i (g) represents the target vector X ij (g) is the vector obtained by cross-recombining with the mutated vector.

[0119] Understandably, to better uncover potential user behavior information and hidden patterns in the data, this application employs the aforementioned Adaptive Differential Evolution Algorithm (ADE) to extract user characteristics for subsequent marketing efforts to predict consumption intentions. This ADE is an improved differential evolution algorithm, belonging to the category of evolutionary algorithms. It introduces an adaptive mutation operator, automatically adjusting the mutation operator value based on the relationship between each individual's fitness value and the optimal individual, ensuring that individuals rapidly and stably approach the optimal value. The basic operations of this ADE algorithm include mutation, crossover, and selection, using iterative methods to search for the optimal individual by leveraging the differences between individuals and mutation strategies.

[0120] Step 102: Based on the set of marketing features to be predicted and the consumption intention prediction model, obtain the predicted value of the influence of each marketing feature to be predicted on the consumption intention in the set of marketing features to be predicted. The set of marketing features to be predicted includes historical marketing success features and the user features to be promoted. The consumption intention prediction model is used to predict the influence of marketing features on user consumption intention.

[0121] It should be noted that the impact of marketing features on consumer intention is used to represent the degree of influence of different marketing features on users' consumer intentions. The higher the impact, the higher the attention users pay to that marketing feature when consuming the product.

[0122] Optionally, an incremental prediction model, such as the Uplift model, can be used to calculate the impact of the marketing feature to be predicted on consumer intention. The Uplift model is mainly used to calculate the impact of marketing interventions (e.g., marketing videos) on users' consumption behavior, i.e. whether adding a certain marketing feature to a marketing video will increase users' intention to purchase the product.

[0123] The formula (6) for calculating the impact of the marketing features to be predicted on consumer intentions is as follows:

[0124] ITE = τ i =Y i (1)-Y i (0) (6)

[0125] Where i = 0, 1, 2, ..., N, N represents the number of users to be promoted, that is, the number of users to be promoted in the user group; Y i(1) represents the probability that the i-th user in the target user group will make a purchase after being influenced by a marketing video containing a certain marketing feature to be predicted; while Y i (0) represents the probability that the i-th user to be promoted will make a purchase after being unaffected by a marketing video containing a certain marketing feature to be predicted.

[0126] Suppose that the probability of user a making a purchase without being influenced by a marketing video containing a certain unpredictable marketing feature is 0.2, while the probability of user a making a purchase after being influenced by a marketing video containing a certain unpredictable marketing feature is 0.9. Then, according to the above formula, the predicted value of the influence of the unpredictable marketing feature on the consumption intention of user a is 0.7. That is, adding a certain unpredictable marketing feature to the marketing video will increase user a's willingness to consume the product.

[0127] However, in practical use, the consumption behavior of users to be promoted under different conditions—with and without marketing intervention—cannot be observed simultaneously. That is, for a given user, one can only choose to either implement marketing intervention or not, and observe the user's consumption behavior under that single condition. Therefore, in this embodiment, a consumption intention prediction model can be constructed using the following implementation method, and based on this model, the influence of marketing features on user consumption intentions can be predicted.

[0128] In one embodiment, optionally, the method further includes:

[0129] Obtain the first distribution divergence between the first user group to be promoted and the second user group to be promoted;

[0130] For each of the marketing features to be predicted in the set of marketing features to be predicted, the set of marketing features to be predicted is divided into two subsets according to the marketing features to be predicted, and the second distribution divergence after splitting with the marketing features to be predicted as nodes is obtained.

[0131] Obtain the first marketing feature to be predicted from the set of marketing features to be predicted that has the largest difference between the second distribution divergence and the first distribution divergence, and divide the set of marketing features to be predicted into two subsets based on the first marketing feature to be predicted;

[0132] The step of obtaining the first marketing feature to be predicted is repeated for each subset until the set can no longer be divided, thereby generating the consumer intention prediction model.

[0133] In this embodiment, the user group to be promoted is divided into a first user group and a second user group, both containing the same number of users. For ease of understanding, the first user group is referred to as the experimental group, and the second user group as the control group. Marketing intervention is performed on the users in the experimental group; no marketing intervention is performed on the users in the control group. Specifically, marketing videos containing specified marketing features are sent to the users in the experimental group to determine whether they make a purchase based on the marketing videos; while marketing videos without specified marketing features are sent to the users in the control group to determine whether they make a purchase.

[0134] The set of marketing features to be predicted is divided into two subsets based on the first marketing feature to be predicted L, and the second distribution divergence after splitting with the first marketing feature to be predicted L as the node is calculated.

[0135] The set of marketing features to be predicted includes the historical marketing success features and the selected user features to be promoted.

[0136] Specifically, the second distribution divergence after splitting can be calculated using the following formula (7):

[0137]

[0138] Where N represents the number of marketing features to be predicted in the set of marketing features to be predicted mentioned before partitioning. k This indicates the number of marketing features to be predicted contained in the subset after partitioning.

[0139] Calculate the difference D between the second distribution divergence after splitting and the first distribution divergence before splitting. Q ;

[0140] Specifically, the difference D can be calculated using the following formula (8). Q :

[0141] D Q =D(P) T ,P C |L)-D(P T ,P C (8)

[0142] It should be noted that distribution divergence is used to represent the difference between two probability distributions (i.e., the first user group to be promoted and the second user group to be promoted). The higher the similarity between the two probability distributions, the smaller their distribution divergence (i.e., the second distribution divergence versus the first distribution divergence); conversely, the lower the similarity between the two probability distributions, the larger their distribution divergence (i.e., the second distribution divergence versus the first distribution divergence). If the probability distributions of the first user group to be promoted and the second user group to be promoted are the same, then their distribution divergence is 0. The purpose of this application embodiment is to increase the distribution divergence of the two distributions by splitting the set of marketing features to be predicted, thereby maximizing the difference between the users to be promoted in the first user group to be promoted and the second user group to be promoted. The difference D... Q The larger the value of D, the greater the divergence of the two distributions after the split compared to the divergence before the split. Q When the value is at its maximum, it indicates that the difference between the first user group to be promoted and the second user group to be promoted after the split is the greatest.

[0143] Repeat the above process until all marketing features in the set of marketing features to be predicted have been traversed, and take the difference D. Q The marketing feature L2 corresponding to the maximum value is the first marketing feature to be predicted, and the first marketing feature to be predicted is used as a splitting node to divide the set of marketing features to be predicted into two subsets.

[0144] Repeat the process of dividing into two subsets as described above to generate a consumer intention prediction model.

[0145] Assume that the set of marketing features to be predicted before partitioning is C, which contains 10 marketing features to be predicted (let's say marketing feature 1 to marketing feature 10). Using marketing feature 5 as the first marketing feature to be predicted, the set of marketing features to be predicted C is partitioned to obtain subsets C1 and C2.

[0146] In subset C1, there are 5 marketing features to be predicted (marketing features 1 to 5), and correspondingly, there are 5 marketing features to be predicted (marketing features 6 to 10). According to the above formula (7), the distribution divergence corresponding to subset C1 and the distribution divergence corresponding to subset C2 are calculated respectively, and the difference D corresponding to subset C1 is calculated according to the above formula (8). Q1 and the difference D corresponding to subset C2 Q2 Iterate through all the marketing features to be predicted in the set C of marketing features to be predicted, divide the set C of marketing features to be predicted into subsets, and calculate the difference D for each subset. QSelect the difference D Q The marketing feature to be predicted at its maximum is the first marketing feature to be predicted.

[0147] Assuming that the marketing feature to be predicted 4 is the first marketing feature to be predicted, and based on the marketing feature to be predicted 4, the set of marketing features to be predicted C is divided into subsets Ci. a (including marketing features to be predicted 1 to marketing features to be predicted 4) and subset C b (Including marketing features to be predicted 5 to marketing features to be predicted 10), for subset C a and subset C b Repeat the above process to generate the consumption intention prediction model.

[0148] It should be noted that the consumer intention prediction model is an incremental prediction tree model. Each marketing feature in the set of marketing features to be predicted is input into the consumer intention prediction model. Following the tree model structure, the predicted value of the influence of each marketing feature on consumer intention is determined. The influence of a marketing feature further from the root node is higher, and so on.

[0149] In one implementation, optionally, obtaining the first distribution divergence between the first user group to be promoted and the second user group to be promoted includes:

[0150] A marketing video including the predicted marketing characteristics is sent to users in the first user group to be promoted, and a first result is obtained as to whether the users in the first user group to be promoted have made a purchase.

[0151] Marketing videos including the predicted marketing features are not sent to users in the second user group to be promoted, thus obtaining a second result as to whether users in the second user group to be promoted make a purchase.

[0152] Based on the first result and the second result, obtain the types of users to be promoted in the first user group to be promoted and the second user group to be promoted.

[0153] Based on the type of users to be promoted, obtain the first distribution divergence between the first user group to be promoted and the second user group to be promoted.

[0154] In the embodiments of this application, the distribution divergence is used to represent the difference between two probability distributions. The higher the similarity between the two probability distributions, the smaller the distribution divergence; the lower the similarity between the two probability distributions, the larger the distribution divergence.

[0155] Optionally, the first distribution divergence between the first user group to be promoted and the second user group to be promoted can be calculated using the KL divergence algorithm, as shown in the following formula (9):

[0156]

[0157] Among them, P T and P C Let represent the probability distributions of the first user group to be promoted and the second user group to be promoted, respectively. and These represent the probabilities that the users in the first and second user groups to be promoted belong to the i-th type of the four user categories, respectively.

[0158] It should be noted that users can be divided into four categories based on whether marketing interventions (such as advertising) result in consumer behavior. The user types are shown in Table 1 below:

[0159] Table 1

[0160]

[0161] Step 103: Based on the predicted value, obtain the target marketing features for the user to be promoted from the set of marketing features to be predicted;

[0162] In one implementation, optionally, based on the predicted value, target marketing features for the user to be promoted are obtained from the set of marketing features to be predicted, including:

[0163] The marketing features to be predicted in the set of marketing features to be predicted are sorted in descending order of the predicted values ​​to obtain the sorted set of marketing features to be predicted.

[0164] A preset number of target marketing features for the users to be promoted are sequentially obtained from the sorted set of marketing features to be predicted.

[0165] In this embodiment of the application, a preset number of target marketing features specific to the users to be promoted are selected in descending order of the predicted values ​​of the influence of consumption intention. It is understood that these preset number of target marketing features specific to the users to be promoted can be referred to as a target marketing feature combination.

[0166] Step 104: Generate a marketing video for the target users based on the target marketing characteristics.

[0167] In one implementation, optionally, generating a marketing video for the target user based on the target marketing characteristics includes:

[0168] Based on the target marketing characteristics, generate marketing text for the users to be promoted;

[0169] Based on the marketing text and the text-based video model, generate the marketing video for the user to be promoted.

[0170] In this embodiment of the application, the target marketing features are combined to obtain the target marketing feature combination, and marketing text for the user to be promoted is generated based on the target marketing feature combination.

[0171] For example, the target marketing feature combination includes: abundant data allowance, high speed, zero lag, low cost, suitability for students, and gaming scenarios. If the product to be recommended is a "carrier data plan," the generated marketing text can include: "Based on xx game, generate a data plan recommendation video targeting students, highlighting the data plan's features: abundant data allowance, high speed, zero lag, and low cost."

[0172] Optionally, the text-based video model includes the Sora text-based video model.

[0173] The marketing text is input into the Sora text-based video model, which generates a marketing video for the user to be promoted based on the marketing text and plays the marketing video to the user, thereby achieving precise marketing recommendations to the user.

[0174] It should be noted that the Sora text-based video model uses a diffusion-type transformer model to construct a deep neural network model to simulate the connection patterns of neurons in the human brain, thereby enabling the processing and learning of complex data. The diffusion-type transformer in the Sora text-based video model extends the single encoder-single decoder model into a dual encoder with dual text semantic information extraction and a single decoder that can fuse guidance signal features. Then, a pointer generation network model is used to select whether to copy words from the source text or generate new summary information using a vocabulary, thus solving the OOV (Out of Vocabulary) problem commonly encountered in summarizing tasks. Furthermore, to efficiently encode positional information, the Sora text-based video model uses relative position representation in the attention layer to introduce sequential information from the text.

[0175] Figure 2 This is a flowchart illustrating one embodiment of the video marketing recommendation method described in this application. Figure 2 As shown, the method includes:

[0176] Step 201: Collect historical consumption data of users on the sales platform;

[0177] Step 202: Preprocess historical consumption data;

[0178] Step 203: Obtain the video marketing plans of successful users and extract historical marketing success features from the video marketing plans;

[0179] Step 204: Obtain the characteristics of the users to be promoted based on their historical consumption data;

[0180] Step 205: Input the historical marketing success features and the user features to be promoted into the pre-trained consumer intention prediction model to obtain the predicted value of the influence of the historical marketing success features and the user features to be promoted on the consumer intention.

[0181] Step 206: Based on the predicted values, obtain the target marketing characteristics for the users to be promoted;

[0182] Step 207: Generate marketing text for the target users based on the target marketing characteristics;

[0183] Step 208: Input the marketing text into the Wensheng video model to generate marketing videos for the users to be promoted.

[0184] In summary, the video marketing recommendation method described in this application, through the analysis of historical consumption data, selects user characteristics with a high correlation to the consumption intention of users to be promoted, as the basis for subsequent analysis. By inputting these selected user characteristics and historical marketing success characteristics into a consumption intention prediction model, the predicted value of the influence of each marketing characteristic on the user's purchase intention is determined. The predicted value of the influence of the marketing characteristic on the user's purchase intention represents the influence of different marketing characteristics on the user's purchase intention. The higher the influence, the higher the user's attention to that characteristic when purchasing a product. Therefore, based on this, the recommendation method can be further developed. The system predicts the impact of the marketing features on consumer intention, identifies the target marketing features for the target users, and generates marketing text for these users based on the target marketing features using a text generation module. This generated marketing text is then input into the Sora text-based video model, which generates a marketing video based on the marketing text and plays it to the target users. This achieves precise marketing recommendations to users. By selecting the target marketing features that most significantly influence user consumption behavior to construct the marketing text for marketing video generation, the success rate of marketing video recommendations is greatly improved, enabling precise marketing to different user groups.

[0185] Please refer to Figure 3 This application also provides a video marketing recommendation system, including:

[0186] The selection module 301 is used to take the parameter vector set corresponding to the historical consumption feature set of the users to be promoted as a population, perform individual selection on the population, and obtain the features of the users to be promoted from the historical consumption feature set.

[0187] Prediction module 302 is used to obtain the predicted value of the influence of each marketing feature to be predicted on the consumption intention in the marketing feature set to be predicted and the consumption intention prediction model. The marketing feature set to be predicted includes historical marketing success features and the user features to be promoted. The consumption intention prediction model is used to predict the influence of marketing features on user consumption intention.

[0188] The acquisition module 303 is used to acquire target marketing features for the user to be promoted from the set of marketing features to be predicted based on the predicted value;

[0189] The generation module 304 is used to generate marketing videos for the users to be promoted based on the target marketing characteristics.

[0190] Optionally, in the video marketing recommendation system, the selection module 301 includes:

[0191] The acquisition unit is used to acquire the accuracy parameter and the area under the curve parameter for each individual in the population.

[0192] The first obtaining unit is used to obtain the fitness value of the individual based on the accuracy parameter and the area under the curve parameter corresponding to the individual;

[0193] The second acquisition unit is used to select individuals from the population based on the fitness value of the individuals, and to obtain the characteristics of users to be promoted from the set of historical consumption characteristics.

[0194] Optionally, in the video marketing recommendation system, the second obtaining unit is specifically used for:

[0195] For every three different individuals in the population, based on the fitness value of the individuals, the first individual is used as the mutation baseline vector, and the difference between the second and third individuals is used as the difference vector to obtain the mutation vector, with the fitness values ​​of the first individual, the second individual, and the third individual increasing sequentially.

[0196] The mutation vector is recombined with the individual to obtain the experimental vector;

[0197] Individual selection is performed based on the fitness values ​​of the experimental vector and the individual to obtain the optimal individual;

[0198] The historical consumption characteristics corresponding to the optimal individual are used as the characteristics of users to be promoted.

[0199] Optionally, the video marketing recommendation system further includes:

[0200] The distribution divergence acquisition module is used to acquire the first distribution divergence between the first user group to be promoted and the second user group to be promoted.

[0201] The first partitioning module is used to divide the set of marketing features to be predicted into two subsets based on each of the marketing features to be predicted in the set of marketing features to be predicted, and to obtain the second distribution divergence after splitting with the marketing features to be predicted as nodes.

[0202] The second partitioning module is used to obtain the first marketing feature to be predicted in the set of marketing features to be predicted with the largest difference between the second distribution divergence and the first distribution divergence, and to partition the set of marketing features to be predicted into two subsets based on the first marketing feature to be predicted.

[0203] The model generation module is used to repeatedly execute the step of obtaining the first marketing feature to be predicted for each subset until the set can no longer be divided, thereby generating the consumer intention prediction model.

[0204] Optionally, in the aforementioned video marketing recommendation system, the distribution divergence acquisition module is specifically used for:

[0205] A marketing video including the predicted marketing features is sent to the users in the first user group to be promoted, and a first result is obtained as to whether the users in the first user group to be promoted make a purchase.

[0206] Marketing videos including the predicted marketing features are not sent to the users in the second user group to be promoted, thus obtaining a second result as to whether the users in the second user group to be promoted have made a purchase.

[0207] Based on the first result and the second result, obtain the types of users to be promoted in the first user group to be promoted and the second user group to be promoted.

[0208] Based on the type of users to be promoted, obtain the first distribution divergence between the first user group to be promoted and the second user group to be promoted.

[0209] Optionally, the video marketing recommendation system further includes:

[0210] The data acquisition module is used to acquire historical consumption data of users on the sales platform;

[0211] The third segmentation module is used to segment the sales platform users into users to be promoted or users who have successfully been marketed, based on the historical consumption data.

[0212] The feature acquisition module is used to acquire the historical marketing success features for the successful marketing users.

[0213] Optionally, in the aforementioned video marketing recommendation system, the acquisition module 303 is specifically used for:

[0214] The marketing features to be predicted in the set of marketing features to be predicted are sorted in descending order of the predicted values ​​to obtain the sorted set of marketing features to be predicted.

[0215] A preset number of target marketing features for the users to be promoted are sequentially obtained from the sorted set of marketing features to be predicted.

[0216] Optionally, in the aforementioned video marketing recommendation system, the generation module 304 is specifically used for:

[0217] Based on the target marketing characteristics, generate marketing text for the users to be promoted;

[0218] Based on the marketing text and the text-based video model, generate the marketing video for the user to be promoted.

[0219] It should be noted that the apparatus provided in this application embodiment can implement all the method steps implemented in the above method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0220] Figure 4 This is a schematic diagram of one embodiment of the video marketing recommendation system described in this application. Figure 4 As shown, the system includes:

[0221] The data acquisition module 401 is used to collect historical consumption data of users on the sales platform;

[0222] The data processing module 402 is used to classify sales platform users into users to be promoted or users who have successfully marketed, based on historical consumption data and the types of products to be promoted.

[0223] The feature selection module 403 is used to obtain the video marketing plans of successful users, extract historical marketing success features from the video marketing plans, and select features based on the historical consumption data of the users to be promoted to obtain the features of the users to be promoted.

[0224] The consumer intention prediction module 404 is used to input historical marketing success features and user features to be promoted into the consumer intention prediction model to obtain the predicted value of the influence of historical marketing success features and user features to be promoted on consumer intention.

[0225] The text generation module 405 is used to obtain target marketing features based on the predicted values, and generate marketing text for the users to be promoted based on the target marketing features;

[0226] The Wensheng video module 406 is used to generate marketing videos for users to be promoted based on marketing text.

[0227] This application also provides a video marketing recommendation system, such as... Figure 5 As shown, it includes:

[0228] The processor 501, memory 502, transceiver 503, and programs or instructions stored in the memory 502 and executable on the processor 501; when the processor 501 executes the programs or instructions, it implements the various processes of the above-described aggregated payment method embodiments and achieves the same technical effect. To avoid repetition, these will not be described again here.

[0229] The transceiver 503 is used to receive and send data under the control of the processor 501.

[0230] Among them, Figure 5 In this context, the bus architecture can include any number of interconnected buses and bridges, specifically connecting various circuits of one or more processors represented by processor 501 and memory represented by memory 502. The bus architecture can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. Transceiver 503 can be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. For different user equipment, the user interface 504 can also be an interface capable of connecting external or internal devices, including but not limited to keypads, displays, speakers, microphones, joysticks, etc.

[0231] The processor 501 is responsible for managing the bus architecture and general processing, while the memory 502 can store the data used by the processor 501 when performing operations.

[0232] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the above-described video marketing recommendation method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0233] This application also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, they implement the various processes of the above-described video marketing recommendation method embodiments and achieve the same technical effects. To avoid repetition, they will not be described again here.

[0234] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover 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 that element.

[0235] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0236] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A video marketing recommendation method, characterized in that, include: By using the set of parameter vectors corresponding to the set of historical consumption characteristics of users to be promoted as a population, individual selection is performed on the population to obtain the characteristics of users to be promoted from the set of historical consumption characteristics. Based on the set of marketing features to be predicted and the consumption intention prediction model, the predicted value of the influence of each marketing feature to be predicted in the set of marketing features to be predicted is obtained. The set of marketing features to be predicted includes historical marketing success features and the user features to be promoted. The consumption intention prediction model is used to predict the influence of marketing features on user consumption intention. Based on the predicted value, the target marketing features for the users to be promoted are obtained from the set of marketing features to be predicted; Based on the target marketing characteristics, generate marketing videos for the users to be promoted.

2. The video marketing recommendation method according to claim 1, characterized in that, By using the set of parameter vectors corresponding to the historical consumption feature set of users to be promoted as a population, and performing individual selection on the population, the characteristics of users to be promoted are obtained from the historical consumption feature set, including: For each individual in the population, obtain the accuracy parameter and the area under the curve parameter corresponding to that individual; The fitness value of an individual is obtained based on its accuracy parameter and area under the curve parameter. Individuals are selected from the population based on their fitness values, and user characteristics to be promoted are obtained from the set of historical consumption characteristics.

3. The video marketing recommendation method according to claim 2, characterized in that, Individual selection is performed on the population based on the fitness values ​​of the individuals, and the characteristics of users to be promoted are obtained from the historical consumption characteristic set, including: For every three different individuals in the population, based on the fitness value of the individuals, the first individual is used as the mutation baseline vector, and the difference between the second and third individuals is used as the difference vector to obtain the mutation vector, with the fitness values ​​of the first individual, the second individual, and the third individual increasing sequentially. The mutation vector is recombined with the individual to obtain the experimental vector; Individual selection is performed based on the fitness values ​​of the experimental vector and the individual to obtain the optimal individual; The historical consumption characteristics corresponding to the optimal individual are used as the characteristics of users to be promoted.

4. The video marketing recommendation method according to claim 1, characterized in that, The method further includes: Obtain the first distribution divergence between the first user group to be promoted and the second user group to be promoted; For each of the marketing features to be predicted in the set of marketing features to be predicted, the set of marketing features to be predicted is divided into two subsets according to the marketing features to be predicted, and the second distribution divergence after splitting with the marketing features to be predicted as nodes is obtained. Obtain the first marketing feature to be predicted from the set of marketing features to be predicted that has the largest difference between the second distribution divergence and the first distribution divergence, and divide the set of marketing features to be predicted into two subsets based on the first marketing feature to be predicted; The step of obtaining the first marketing feature to be predicted is repeated for each subset until the set can no longer be divided, thereby generating the consumer intention prediction model.

5. The video marketing recommendation method according to claim 4, characterized in that, Obtain the first distribution divergence between the first user group to be promoted and the second user group to be promoted, including: A marketing video including the predicted marketing features is sent to the users in the first user group to be promoted, and a first result is obtained as to whether the users in the first user group to be promoted make a purchase. Marketing videos including the predicted marketing features are not sent to the users in the second user group to be promoted, and a second result is obtained as to whether the users in the second user group to be promoted have made a purchase. Based on the first result and the second result, obtain the types of users to be promoted in the first user group to be promoted and the second user group to be promoted. Based on the type of users to be promoted, obtain the first distribution divergence between the first user group to be promoted and the second user group to be promoted.

6. The video marketing recommendation method according to claim 1, characterized in that, The method further includes: Obtain historical consumption data from users on the sales platform; Based on the historical consumption data, the users of the sales platform are divided into users to be promoted or users who have successfully been marketed. For the users who have successfully engaged in marketing, obtain the characteristics of their historical marketing success.

7. The video marketing recommendation method according to claim 1, characterized in that, Based on the predicted value, target marketing features for the user to be promoted are obtained from the set of marketing features to be predicted, including: The marketing features to be predicted in the set of marketing features to be predicted are sorted in descending order of the predicted values ​​to obtain the sorted set of marketing features to be predicted. A preset number of target marketing features for the users to be promoted are sequentially obtained from the sorted set of marketing features to be predicted.

8. The video marketing recommendation method according to claim 1, characterized in that, Generate marketing videos for the target users based on the target marketing characteristics, including: Based on the target marketing characteristics, generate marketing text for the users to be promoted; Based on the marketing text and the text-based video model, generate the marketing video for the user to be promoted.

9. A video marketing recommendation system, characterized in that, include: The selection module is used to select individuals from the set of parameter vectors corresponding to the set of historical consumption characteristics of users to be promoted, and obtain the characteristics of users to be promoted from the set of historical consumption characteristics. The prediction module is used to obtain the predicted value of the influence of each marketing feature in the marketing feature set to be predicted on the consumer intention based on the marketing feature set to be predicted and the consumer intention prediction model. The marketing feature set to be predicted includes historical marketing success features and the user features to be promoted. The consumer intention prediction model is used to predict the influence of marketing features on user consumption intention. The acquisition module is used to acquire target marketing features for the user to be promoted from the set of marketing features to be predicted based on the predicted value. The generation module is used to generate marketing videos for the users to be promoted based on the target marketing characteristics.

10. A video marketing recommendation system, characterized in that, include: A processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the processor, when executing the program or instructions, implements the video marketing recommendation method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the video marketing recommendation method as described in any one of claims 1 to 8.

12. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the video marketing recommendation method as described in any one of claims 1 to 8.

Images