Method, device, equipment and storage medium for controlling advertisement delivery of video files

Abstract

The present application relates to a method, device, equipment and storage medium for controlling the advertising delivery of video files. The method comprises: obtaining a video file to be scored; inputting the video file to be scored into a pre-trained scoring model to obtain a score value of the video file to be scored, so as to deliver advertisements to the video file to be scored according to the score value of the video file to be scored, wherein the scoring model is trained based on the advertising delivery data of each historical advertising video file within a first preset time period, and the advertising delivery data includes the click-through rate of each historical advertising video file and the hot word information contained therein. The present application solves the problem of the inefficiency caused by the need for manual screening of advertising materials in traditional technologies, and the problem of inaccurate screening of advertising materials due to the single dimension of screening advertising materials.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device, equipment and storage medium for controlling advertisement delivery of video files. Background Art

[0002] At present, when placing advertisements, the video materials used are all placed based on expert experience. There is no quantitative indicator for the quality of video content. Advertising materials are screened and placed based only on expert experience or feeling or through a single dimension.

[0003] The traditional technical solution has at least the following defects:

[0004] 1. It takes a lot of manpower to screen the materials produced by suppliers before creating advertising plans. This will lead to subjective preferences playing a big role, which may cause good materials or materials that may run out of volume to be screened out.

[0005] 2. Screening advertising materials through a single dimension can easily lead to inaccurate screening of advertising materials, resulting in poor quality of customers brought by the advertising materials. Summary of the invention

[0006] Based on this, it is necessary to provide a method, device, equipment and storage medium for controlling advertisement delivery of video files in response to the above technical problems.

[0007] A method for controlling advertisement delivery of a video file, the method comprising:

[0008] Get the video file to be rated;

[0009] Determine the scoring result of the video file to be scored according to the video file to be scored and a pre-trained scoring model, wherein the scoring model is trained based on the advertising delivery data of each historical advertising video file, and the advertising delivery data includes the click rate of each historical advertising video file and the hot word information contained therein;

[0010] The advertisement delivery of the video file to be rated is controlled according to the rating result.

[0011] In one embodiment, the method further comprises:

[0012] Acquire advertisement delivery data of each historical advertisement video file, the advertisement delivery data also including video category information of the corresponding historical advertisement video file;

[0013] The click rate, hot word information and video category information of each historical advertising video file are used as model training data for model training to obtain a scoring model.

[0014] In one embodiment, the above-mentioned model training is performed using the click rate, hot word information and video category information of each historical advertisement video file as model training data to obtain a scoring model, including:

[0015] Input the click rate, hot word information and video category information of each historical advertising video file into the original scoring model as model training data in turn, and obtain the predicted score value corresponding to the model training data. The predicted score value is determined according to the score value predicted by the original scoring model for the input hot word information and video category information;

[0016] Determine the target loss function according to the prediction score value corresponding to the model training data and the click-through rate of the historical advertising video file to which the model training data belongs;

[0017] The original scoring model is trained using the target loss function;

[0018] When the preset training stop condition is reached, the classification model is obtained.

[0019] In one embodiment, the above-mentioned determining the scoring result of the video file to be scored according to the video file to be scored and the pre-trained scoring model includes:

[0020] Extracting text information from each key frame in the video file to be rated, and determining video category information of the video file to be rated based on the text information;

[0021] Inputting the video category information of the video file to be rated into the rating model, and obtaining a first rating value predicted by the rating model for the input video category information;

[0022] Extracting hot word information from the text information, inputting the extracted hot word information into a scoring model, and obtaining a second scoring value predicted by the scoring model for the input hot word information;

[0023] A scoring result is determined by the scoring model according to the first scoring value and the second scoring value.

[0024] In one embodiment, the advertisement delivery data of each of the above historical advertisement video files also includes the credit data of each of the above historical advertisement video files, and the click rate, hot word information and video category information of each of the above historical advertisement video files are used as model training data to perform model training to obtain a scoring model, including:

[0025] The click-through rate, hot word information, video category information and credit data of each historical advertising video file are used as model training data to train the model and obtain a scoring model;

[0026] The target loss function is determined based on the predicted score value corresponding to the model training data and the click-through rate of the historical advertising video file to which the model training data belongs, including:

[0027] The target loss function is determined based on the predicted score value corresponding to the model training data, the click-through rate of the historical advertising video file to which the model training data belongs, and the credit data.

[0028] In one embodiment, the target loss function is determined based on the prediction score value corresponding to the model training data and the click rate of the historical advertisement video file to which the model training data belongs, including:

[0029] Convert the click-through rate of the historical advertising video file to which the model training data belongs into a target score value;

[0030] Construct the target loss function based on the predicted score value and target score value corresponding to the model training data.

[0031] In one embodiment, the above-mentioned controlling the advertisement delivery of the to-be-rated video file according to the rating result includes:

[0032] According to the scoring results, determine the bid price of the video file to be scored and the advertisement placement position;

[0033] Advertisements are delivered to the video file to be rated according to the bids of the video file to be rated and the advertisement delivery positions.

[0034] A device for controlling advertisement delivery of a video file, characterized in that the device comprises:

[0035] An acquisition module is used to obtain the video file to be rated;

[0036] A scoring module is used to determine the scoring result of the video file to be scored based on the video file to be scored and a pre-trained scoring model, wherein the scoring model is trained based on the advertising delivery data of each historical advertising video file, and the advertising delivery data includes the click-through rate of each historical advertising video file and the hot word information contained therein;

[0037] The control module is used to control the advertisement delivery of the video file to be rated according to the rating result.

[0038] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0039] Get the video file to be rated;

[0040] Determine the scoring result of the video file to be scored according to the video file to be scored and a pre-trained scoring model, wherein the scoring model is trained based on the advertising delivery data of each historical advertising video file, and the advertising delivery data includes the click rate of each historical advertising video file and the hot word information contained therein;

[0041] The advertisement delivery of the video file to be rated is controlled according to the rating result.

[0042] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0043] Get the video file to be rated;

[0044] Determine the scoring result of the video file to be scored according to the video file to be scored and a pre-trained scoring model, wherein the scoring model is trained based on the advertising delivery data of each historical advertising video file, and the advertising delivery data includes the click rate of each historical advertising video file and the hot word information contained therein;

[0045] The advertisement delivery of the video file to be rated is controlled according to the rating result.

[0046] The above-mentioned video file advertising delivery control method, device, equipment and storage medium obtain a video file to be scored; input the video file to be scored into a pre-trained scoring model to obtain a score value of the video file to be scored, so as to deliver advertisements to the video file to be scored according to the score value of the video file to be scored, and the scoring model is trained based on the advertising delivery data of each historical advertising video file within a first preset time period, and the advertising delivery data includes the click-through rate of each historical advertising video file and the hot word information contained therein. The present application can score the video file to be scored through a pre-trained scoring model, and the scoring model is trained based on the click-through rate of each historical advertising video file within a preset time period and the hot word information contained therein. Scoring the video file to be scored based on such a model solves the inefficiency caused by the need for manual screening of advertising materials in traditional technologies, and the problem of inaccurate screening of advertising materials due to the single dimension of screening advertising materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A schematic diagram of a flow chart of a method for controlling advertisement delivery of a video file in one embodiment;

[0048] Figure 2 A flowchart of the steps of inputting the click rate, hot word information and video category information of each historical advertisement video file as model training data into the original scoring model in sequence in one embodiment to obtain a predicted score value corresponding to the model training data;

[0049] Figure 3 In another embodiment, a scoring result of the video file to be scored is determined according to the video file to be scored and a pre-trained scoring model;

[0050] Figure 4 In another embodiment, a target loss function is determined according to a prediction score value corresponding to the model training data and a click-through rate of a historical advertisement video file to which the model training data belongs;

[0051] Figure 5 It is a structural block diagram of a device for controlling advertisement delivery of video files in one embodiment;

[0052] Figure 6 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0054] In one embodiment, Figure 1 As shown, a method for controlling the advertisement delivery of a video file is provided, and the method is applied to a server as an example for explanation, and includes the following steps:

[0055] S11. Obtain the video file to be rated.

[0056] In the present application, the above-mentioned video file to be evaluated may be an advertising material to be evaluated.

[0057] S12. Determine a scoring result of the video file to be scored based on the video file to be scored and a pre-trained scoring model. The scoring model is trained based on the advertising delivery data of each historical advertising video file. The advertising delivery data includes the click rate of each historical advertising video file and the hot word information contained therein.

[0058] In the present application, the above-mentioned scoring model is a model for scoring video files to be evaluated. The scoring model is obtained by training using the advertising delivery data of each historical advertising video file within a preset time period. Among them, the above-mentioned historical advertising video files can be advertising video files that have been delivered in history. The above-mentioned advertising delivery data can be data on the effects brought about after the advertising delivery and attribute information of the advertisement itself. Among them, the data on the effects brought about after the advertising delivery can be data such as the exposure rate, click-through rate, registration rate, first-time rate, review rate and credit data of the advertisement. The attribute information of the advertisement itself can be data such as the category of the advertisement and the size of the advertisement. The above-mentioned hot word information can be information on the hot words contained in each historical advertising video file. Among them, the hot word can be a hot word in the current preset time period. The above-mentioned hot word information can be the ranking of the hot word and the number of hot words. Among them, the method for determining the hot word can include:

[0059] Get each key frame of each historical advertisement video file;

[0060] Extracting text information of each key frame of each historical advertisement video file;

[0061] Segment the text information of each key frame of each historical advertisement video file to obtain each word;

[0062] Sort the words and obtain the first preset number of words in the sorting results as hot words;

[0063] Get the ranking of each hot word in the sorting results and obtain the hot word information of each hot word.

[0064] Among them, when determining hot words, the advertising video files of the most recent period of time should be selected for determination. For example, when determining the hot words of the current time, the advertising video files of the most recent month or two months can be selected for determination, so that the hot words obtained have a certain timeliness. The hot words corresponding to each historical time period are different, but a preset number of hot words with the highest ranking can be selected each time.

[0065] S13. Controlling advertisement delivery of the video file to be rated according to the rating result.

[0066] In the present application, the above-mentioned scoring result may be an evaluation score value. The above-mentioned control of advertisement placement of the video file to be rated according to the scoring result may give a higher advertisement bid for the video file to be rated with a high scoring result, while for the video file to be rated with a low scoring result, it may be chosen not to place advertisements or give a lower advertisement bid, because the customer base brought by such material is of poor quality.

[0067] In one embodiment, the above method may further include:

[0068] Acquire advertisement delivery data of each historical advertisement video file, the advertisement delivery data also including video category information of the corresponding historical advertisement video file;

[0069] The click rate, hot word information and video category information of each historical advertising video file are used as model training data for model training to obtain a scoring model.

[0070] In the present application, the video category information may be the advertisement type of each historical advertisement video file. For example, the advertisement type may be oral, creative, subtitled, unmanned, music, books, sports, etc.

[0071] The above-mentioned model training using the click rate, hot word information and video category information of each historical advertising video file as model training data to obtain a scoring model may include:

[0072] According to the hot word information and video category information of each historical advertising video file, a prediction score is given to each historical advertising video file;

[0073] Evaluate the predicted score of each historical advertising video file according to the click rate of each historical advertising video file;

[0074] Determine whether the scoring model is trained based on the evaluation results.

[0075] In the present application, the video type information, hot word information and click-through rate of each historical advertising video file can be trained through this implementation method, so that the obtained scoring model can score the video files to be scored based on the dimensions of video type information and hot word information, and the click-through rate is used to evaluate the accuracy of the scoring results of the model. Therefore, the present application can score the video files to be evaluated from three dimensions: video type information, hot word information and click-through rate, which solves the problem of inefficiency caused by the need for manual screening of advertising materials in traditional technologies, and the problem of insufficient screening accuracy caused by screening advertising materials from a single dimension.

[0076] Please refer to Figure 2 In one embodiment, the above-mentioned model training using the click rate, hot word information and video category information of each historical advertisement video file as model training data to obtain a scoring model may include:

[0077] S21, inputting the click rate, hot word information and video category information of each historical advertising video file into the original scoring model as model training data in turn, obtaining a predicted score value corresponding to the model training data, and the predicted score value is determined according to the score value predicted by the original scoring model for the input hot word information and video category information;

[0078] S22, determining a target loss function according to the prediction score value corresponding to the model training data and the click-through rate of the historical advertising video file to which the model training data belongs;

[0079] S23, training the original scoring model through the target loss function;

[0080] S24. When the preset training stop condition is reached, a classification model is obtained.

[0081] In the present application, the above-mentioned original scoring model may include multiple model parameters, and the model parameters may include preset score values ​​corresponding to the two dimensions of hot word information and video category information. The original scoring model processes the click-through rate, hot word information and video category information of each input historical advertising video file according to the model parameters to obtain the predicted score value corresponding to the model training data.

[0082] In one embodiment, the above-mentioned original scoring model may include initial score values ​​corresponding to the two dimensions of hot word information and video category information. For example, the initial score value for the dimension of hot word information may be set as follows:

[0083] The included hot words are ranked first and have an initial score of X1;

[0084] The ranking of the included hot words is second, and the initial score is X2;

[0085] The ranking of the included hot words is third, and the initial score is X3;

[0086] Alternatively, the initial score value for the hot word information dimension can be set as follows:

[0087] The number of hot words included is one, and the initial score is X4;

[0088] The number of hot words included is two, and the initial score is X5;

[0089] The number of hot words included is three, and the initial score is X6;

[0090] Alternatively, the initial score value for the hot word information dimension can be set as follows:

[0091] The ranking of the included hot words is the first, and the number is one, and the initial score is X7;

[0092] The ranking of the included hot words is the first, and the number is two, and the initial score is X8;

[0093] The ranking of the included hot words is second, the number is one, and the initial score is X9.

[0094] The specific setting method can be set according to actual needs and is not specifically limited here.

[0095] Furthermore, the initial score value for the dimension of video category information may be set as follows:

[0096] Multiplayer interactive plot category, the initial score is A1;

[0097] Oral broadcast (announcement of products in a hosting manner), the initial score is A2;

[0098] Creative category, initial score is A3;

[0099] Subtitle broadcast category, initial score is A4;

[0100] The initial score for unmanned plot-based games (no human interaction, using a small scene such as mobile phone clicks and voice to explain the content) is A5.

[0101] When training the model, the above-mentioned scores are continuously adjusted to adjust the value of the corresponding target loss function. When the value of the target loss function reaches a preset threshold or the number of iterations reaches a preset number, it is determined that the above-mentioned preset training stop condition is met.

[0102] The target loss function is used to measure the mismatch between the evaluation result of the scoring model and the click-through rate.

[0103] Taking the current training as an example, the model training results are explained as follows:

[0104] Obtain each historical advertisement video file and the click rate, hot word information and video category information of each historical advertisement video file and input them into the current scoring model as model training data in turn;

[0105] Extracting text information from each key frame of each historical advertisement video file, and determining video category information of each historical advertisement video file according to the text information from each key frame of each historical advertisement video file;

[0106] Inputting the video category information of each historical advertisement video file into the current scoring model to obtain the score predicted by the current scoring model for the video list information of each historical advertisement video file;

[0107] Determine the hot word information of each historical advertising video file according to the text information in each key frame of each historical advertising video file;

[0108] Input the hot word information of each historical advertisement video file into the current scoring model to obtain the prediction score of the hot word information of each historical advertisement video file by the current scoring model;

[0109] According to the scores predicted by the current scoring model for the video list information of each historical advertising video file, and the scores predicted by the current scoring model for the hot word information of each historical advertising video file, the evaluation results of the current scoring model for each historical advertising video file are obtained;

[0110] Obtain the click-through rate of each historical advertising video file, and obtain the value of the current loss function according to the click-through rate of each historical advertising video file and the evaluation result of each historical advertising video file;

[0111] Determine whether the value of the current loss function meets the above preset training stop condition. If so, stop training, otherwise proceed to the next training.

[0112] Through this implementation, the present application can enable the trained scoring model to evaluate the video file to be evaluated from multiple dimensions, making the evaluation result more accurate.

[0113] Please refer to Figure 3 In one embodiment, the above-mentioned step of determining the scoring result of the video file to be scored based on the video file to be scored and the pre-trained scoring model may include:

[0114] S31, extracting text information in each key frame in the video file to be rated, and determining video category information of the video file to be rated according to the text information;

[0115] S32, inputting the video category information of the video file to be rated into the rating model, and obtaining a first rating value predicted by the rating model for the input video category information;

[0116] S33, extracting hot word information from the text information, inputting the extracted hot word information into a scoring model, and obtaining a second scoring value predicted by the scoring model for the input hot word information;

[0117] S34: Determine a scoring result through the scoring model and according to the first scoring value and the second scoring value.

[0118] In the present application, the above-mentioned key frames can be frames with key information in the video file to be scored. For example, for a video file of food recommendation, the pictures containing food and the pictures with text descriptions of food can be key frames, and the remaining blank pictures or pictures without important information can be non-key frames. The present application extracts key frames for processing, without processing each frame, thereby improving the efficiency of data processing.

[0119] The above-mentioned text information can be text information in each key frame, and text information converted from voice information in each key frame. The above-mentioned determination of the video category information of the video file to be rated based on the text information can be to match the extracted text information with a pre-set mapping relationship between text information and video category to obtain the corresponding video category information.

[0120] The step of inputting the video category information of the to-be-rated video file into the rating model to obtain a first rating value predicted by the rating model for the input video category information may include:

[0121] Matching the video category information of the video file to be rated with the score values ​​of each video category in the rating model to obtain the score value corresponding to the video category information of the video file to be rated;

[0122] The product of the score value corresponding to the video category information of the video file to be rated and a preset first weight value is determined to obtain the above-mentioned first rating value.

[0123] The above-mentioned step of extracting hot word information from text information, inputting the extracted hot word information into a scoring model, and obtaining a second scoring value predicted by the scoring model for the input hot word information may include:

[0124] Obtain a target video file within a preset time period closest to the current time, the target video file being an already released advertisement video file;

[0125] Extracting text information from each target video file, segmenting the extracted text information to obtain multiple words;

[0126] Sort the obtained multiple words, and select a preset number of words with the highest ranking as hot words corresponding to the current time period;

[0127] The text information of the video file to be scored is also segmented to obtain multiple words, and the multiple words obtained are matched with the hot words corresponding to the current time period obtained above to obtain the hot word information contained in the video file to be scored. The hot word information can be the ranking rank and number of the hot words contained, the ranking and number of synonyms of the hot words contained, and other information.

[0128] The obtained hot word information of the video file to be rated is input into the rating model, and the rating model matches the hot word information of the video file to be rated with the preset score value in the model to obtain the score value corresponding to the hot word information of the video file to be rated;

[0129] The product of the score value corresponding to the hot word information of the video file to be rated and the preset second weight value is determined to obtain the above-mentioned second rating value.

[0130] Furthermore, the above-mentioned determining the scoring result by using the scoring model and according to the first scoring value and the second scoring value may include:

[0131] The sum of the first scoring value and the second scoring value is determined to obtain a scoring result.

[0132] This application sets score values ​​and weight values ​​for the two dimensions of hot word information and video category information. The set score values ​​can be adjusted through the model training process to make the model's scoring results more and more accurate.

[0133] In one embodiment, the advertisement delivery data of each historical advertisement video file also includes the credit data of each historical advertisement video file, and the click rate, hot word information and video category information of each historical advertisement video file are used as model training data to perform model training to obtain a scoring model, including:

[0134] The click-through rate, hot word information, video category information and credit data of each historical advertising video file are used as model training data to train the model and obtain a scoring model;

[0135] According to the predicted score value corresponding to the model training data and the click-through rate of the historical advertising video file to which the model training data belongs, the target loss function is determined, including:

[0136] The target loss function is determined based on the predicted score value corresponding to the model training data, the click-through rate of the historical advertising video file to which the model training data belongs, and the credit data.

[0137] In this application, the above-mentioned advertising data may include video category information of the advertisement, exposure rate, click-through rate, registration rate, first-time rate, review rate, and credit data. Among them, the credit data may include the number of users who are granted credit and the credit rate. Credit refers to being granted a certain credit limit.

[0138] In this application, the quality of the model can be evaluated by the click-through rate, the quality of the model can be evaluated by the credit data, and the quality of the model can be evaluated by the click-through rate and the credit data. In one embodiment, the model can also be evaluated by other data such as exposure rate, click-through rate, registration rate, first-time rate, review completion rate and other data, and different models are obtained by selecting different data to evaluate the model. For example, the click-through rate is selected to evaluate the model, and the resulting scoring model can make the click-through rate of the screened video files to be rated higher and the drainage effect is better. The credit data is selected to evaluate the model, and the resulting scoring model can make the credit rate of the users brought by the screened video files to be rated higher. Then the model is evaluated based on the click-through rate and the credit data, and the click-through rate of the video files to be rated screened by the scoring model is high and the credit rate of the users brought by the search is also correspondingly high.

[0139] The above-mentioned use of click-through rate to evaluate the model refers to determining the target loss function based on the click-through rate and the evaluation results obtained during model training, and determining the degree of model training, that is, whether the training is completed, based on the value of the target loss function.

[0140] In the present application, the objective function determined based on the predicted score value corresponding to the model training data, the click-through rate of the historical advertising video file to which the model training data belongs, and the credit data can measure the degree of mismatch between the predicted score value and the click-through rate and the credit data.

[0141] Please refer to Figure 4 In one embodiment, the above-mentioned determination of the target loss function based on the prediction score value corresponding to the model training data and the click-through rate of the historical advertisement video file to which the model training data belongs may include:

[0142] S41, converting the click rate of the historical advertisement video file to which the model training data belongs into a target score value;

[0143] S42. Construct a target loss function according to the predicted score value and the target score value corresponding to the model training data.

[0144] In the present application, the above-mentioned conversion of the click rate of the historical advertisement video file to which the model training data belongs into the target score value may include:

[0145] Compare the click-through rate of the historical advertising video file to which the model training data belongs with a preset range interval to determine the range interval corresponding to the click-through rate;

[0146] Get the preset score corresponding to the range interval corresponding to the click rate to obtain the target score value.

[0147] For example, the preset range interval and the corresponding preset score can be set as follows:

[0148] 0~0.01: low click rate, the default score is C1

[0149] Greater than 0.01: high click-through rate, the default score is C2

[0150] The target loss function in this application can be used to measure the inconsistency between the predicted score value and the target score value. Generally speaking, the smaller the target loss function, the better the robustness of the model.

[0151] In one embodiment, the above-mentioned controlling the advertisement delivery of the to-be-rated video file according to the rating result may include:

[0152] According to the scoring results, determine the bid price of the video file to be scored and the advertisement placement position;

[0153] Advertisements are delivered to the video file to be rated according to the bids of the video file to be rated and the advertisement delivery positions.

[0154] In the present application, the above-mentioned scoring result may be an evaluation score value, and the above-mentioned determination of the bid price and advertisement placement position of the video file to be scored according to the scoring result may include:

[0155] Determine the range of the evaluation score value and the preset advertisement bid corresponding to the range, and obtain the bid of the video file to be rated;

[0156] Determine the advertising placement position corresponding to the range of the evaluation score value.

[0157] In this application, the range of evaluation score values ​​can be divided, and the advertising bids corresponding to each range can be set. Generally speaking, the advertising bids corresponding to video files with low evaluation score values ​​are low, and the advertising bids corresponding to video files with high evaluation score values ​​are high.

[0158] Furthermore, the present application presets the advertising placement positions corresponding to each range interval. Generally, the advertising placement positions corresponding to video files with high evaluation scores are easier to be exposed, while the advertising placement positions corresponding to video files with low evaluation scores are more difficult to be exposed, or are not placed. The advertising placement positions in the present application can be the time period of playback or the video file of playback.

[0159] In one embodiment, Figure 5 As shown, a video file advertisement delivery control device is provided, comprising: an acquisition module 11, a scoring module 12 and a control module 13, wherein:

[0160] An acquisition module 11 is used to acquire a video file to be rated;

[0161] The scoring module 12 is used to determine the scoring result of the video file to be scored according to the video file to be scored and a pre-trained scoring model, wherein the scoring model is trained based on the advertising delivery data of each historical advertising video file, and the advertising delivery data includes the click rate of each historical advertising video file and the hot word information contained therein;

[0162] The control module 13 is used to control the advertisement delivery of the video file to be rated according to the rating result.

[0163] In one of the embodiments, the above-mentioned device may also include a training module (not shown), which can obtain the advertising delivery data of each historical advertising video file, the advertising delivery data also includes the video category information of the corresponding historical advertising video file, and the click rate, hot word information and video category information of each historical advertising video file are used as model training data to perform model training to obtain a scoring model.

[0164] In one of the embodiments, the above-mentioned training module can also input the click rate, hot word information and video category information of each historical advertising video file as model training data into the original scoring model in turn to obtain the predicted score value corresponding to the model training data. The predicted score value is determined based on the score value predicted by the original scoring model for the input hot word information and video category information. The target loss function is determined based on the predicted score value corresponding to the model training data and the click rate of the historical advertising video file to which the model training data belongs. The original scoring model is trained using the target loss function, and a classification model is obtained when the preset training stop condition is reached.

[0165] In one embodiment, the scoring module 12 can extract text information from each key frame in the video file to be scored, determine the video category information of the video file to be scored based on the text information, input the video category information of the video file to be scored into the scoring model, obtain a first scoring value predicted by the scoring model for the input video category information, extract hot word information from the text information, input the extracted hot word information into the scoring model, obtain a second scoring value predicted by the scoring model for the input hot word information, and determine the scoring result through the scoring model and based on the first scoring value and the second scoring value.

[0166] In one of the embodiments, the advertising placement data of each of the above-mentioned historical advertising video files also includes the credit data of each historical advertising video file. The above-mentioned training module can also use the click-through rate, hot word information, video category information and credit data of each historical advertising video file as model training data for model training to obtain a scoring model. The above-mentioned training module can also determine the target loss function based on the predicted score value corresponding to the model training data, the click-through rate and credit data of the historical advertising video file to which the model training data belongs.

[0167] In one of the embodiments, the training module may also convert the click rate of the historical advertisement video file to which the model training data belongs into a target score value, and construct a target loss function according to the predicted score value and the target score value corresponding to the model training data.

[0168] In one embodiment, the control module 13 may determine the bid price and advertisement placement position of the video file to be rated according to the rating result, and place advertisements on the video file to be rated according to the bid price and advertisement placement position of the video file to be rated.

[0169] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as the operation data of smart home devices. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a resource allocation method for compiling a virtual machine is implemented.

[0170] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program: obtaining a video file to be rated; determining a rating result of the video file to be rated based on the video file to be rated and a pre-trained rating model, wherein the rating model is trained based on advertising delivery data of each historical advertising video file, wherein the advertising delivery data includes the click-through rate of each historical advertising video file and hot word information contained therein; and controlling advertising delivery of the video file to be rated based on the rating result.

[0171] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0172] Acquire advertisement delivery data of each historical advertisement video file, the advertisement delivery data also including video category information of the corresponding historical advertisement video file;

[0173] The click rate, hot word information and video category information of each historical advertising video file are used as model training data for model training to obtain a scoring model.

[0174] In one embodiment, when the processor executes a computer program to implement the above-mentioned step of using the click rate, hot word information, and video category information of each historical advertisement video file as model training data to perform model training and obtain a scoring model, the following steps are specifically implemented:

[0175] Input the click rate, hot word information and video category information of each historical advertising video file into the original scoring model as model training data in turn, and obtain the predicted score value corresponding to the model training data. The predicted score value is determined according to the score value predicted by the original scoring model for the input hot word information and video category information;

[0176] Determine the target loss function according to the prediction score value corresponding to the model training data and the click-through rate of the historical advertising video file to which the model training data belongs;

[0177] The original scoring model is trained using the target loss function;

[0178] When the preset training stop condition is reached, the classification model is obtained.

[0179] In one embodiment, when the processor executes the computer program to implement the above-mentioned step of determining the scoring result of the video file to be scored according to the video file to be scored and the pre-trained scoring model, the following steps are specifically implemented:

[0180] Extracting text information from each key frame in the video file to be rated, and determining video category information of the video file to be rated based on the text information;

[0181] Inputting the video category information of the video file to be rated into the rating model, and obtaining a first rating value predicted by the rating model for the input video category information;

[0182] Extracting hot word information from the text information, inputting the extracted hot word information into a scoring model, and obtaining a second scoring value predicted by the scoring model for the input hot word information;

[0183] A scoring result is determined by the scoring model according to the first scoring value and the second scoring value.

[0184] In one embodiment, the advertisement delivery data of each of the above-mentioned historical advertisement video files also includes the credit data of each of the above-mentioned historical advertisement video files. When the processor executes the computer program to implement the above-mentioned step of using the click rate, hot word information and video category information of each of the historical advertisement video files as model training data for model training and obtaining the scoring model, the following steps are specifically implemented:

[0185] The click-through rate, hot word information, video category information and credit data of each historical advertising video file are used as model training data to train the model and obtain a scoring model;

[0186] The processor executes the computer program to implement the above-mentioned step of determining the target loss function according to the predicted score value corresponding to the model training data and the click-through rate of the historical advertising video file to which the model training data belongs, and specifically implements the following steps:

[0187] The target loss function is determined based on the predicted score value corresponding to the model training data, the click-through rate of the historical advertising video file to which the model training data belongs, and the credit data.

[0188] In one embodiment, the processor executes a computer program to implement the above-mentioned step of determining the target loss function based on the predicted score value corresponding to the model training data and the click-through rate of the historical advertising video file to which the model training data belongs, and specifically implements the following steps:

[0189] Convert the click-through rate of the historical advertising video file to which the model training data belongs into a target score value;

[0190] Construct the target loss function based on the predicted score value and target score value corresponding to the model training data.

[0191] In one embodiment, when the processor executes the computer program to implement the above-mentioned step of controlling the advertisement delivery of the to-be-rated video file according to the rating result, the following steps are specifically implemented:

[0192] According to the scoring results, determine the bid price of the video file to be scored and the advertisement placement position;

[0193] Advertisements are delivered to the video file to be rated according to the bids of the video file to be rated and the advertisement delivery positions.

[0194] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: obtaining a video file to be rated; determining a rating result of the video file to be rated based on the video file to be rated and a pre-trained rating model, wherein the rating model is trained based on advertising delivery data of each historical advertising video file, and the advertising delivery data includes the click rate of each historical advertising video file and the hot word information contained therein; and controlling the advertising delivery of the video file to be rated based on the rating result.

[0195] In one embodiment, when the computer program is executed by a processor, the following steps are specifically implemented:

[0196] Acquire advertisement delivery data of each historical advertisement video file, the advertisement delivery data also including video category information of the corresponding historical advertisement video file;

[0197] The click rate, hot word information and video category information of each historical advertising video file are used as model training data for model training to obtain a scoring model.

[0198] In one embodiment, the computer program is executed by a processor to implement the above-mentioned step of using the click rate, hot word information and video category information of each historical advertising video file as model training data to perform model training and obtain a scoring model, and specifically implements the following steps:

[0199] Input the click rate, hot word information and video category information of each historical advertising video file into the original scoring model as model training data in turn, and obtain the predicted score value corresponding to the model training data. The predicted score value is determined according to the score value predicted by the original scoring model for the input hot word information and video category information;

[0200] Determine the target loss function according to the prediction score value corresponding to the model training data and the click-through rate of the historical advertising video file to which the model training data belongs;

[0201] The original scoring model is trained using the target loss function;

[0202] When the preset training stop condition is reached, the classification model is obtained.

[0203] In one embodiment, when the computer program is executed by the processor to implement the above-mentioned step of determining the scoring result of the video file to be scored according to the video file to be scored and the pre-trained scoring model, the following steps are specifically implemented:

[0204] Extracting text information from each key frame in the video file to be rated, and determining video category information of the video file to be rated based on the text information;

[0205] Inputting the video category information of the video file to be rated into the rating model, and obtaining a first rating value predicted by the rating model for the input video category information;

[0206] Extracting hot word information from the text information, inputting the extracted hot word information into a scoring model, and obtaining a second scoring value predicted by the scoring model for the input hot word information;

[0207] A scoring result is determined by the scoring model according to the first scoring value and the second scoring value.

[0208] In one embodiment, the advertisement delivery data of each of the above historical advertisement video files also includes the credit data of each of the above historical advertisement video files. When the computer program is executed by the processor to implement the above step of using the click rate, hot word information and video category information of each of the historical advertisement video files as model training data for model training and obtaining the scoring model, the following steps are specifically implemented:

[0209] The click-through rate, hot word information, video category information and credit data of each historical advertising video file are used as model training data to train the model and obtain a scoring model;

[0210] The computer program is executed by the processor to implement the above-mentioned step of determining the target loss function based on the predicted score value corresponding to the model training data and the click-through rate of the historical advertising video file to which the model training data belongs, and specifically implements the following steps:

[0211] The target loss function is determined based on the predicted score value corresponding to the model training data, the click-through rate of the historical advertising video file to which the model training data belongs, and the credit data.

[0212] In one embodiment, the computer program is executed by a processor to implement the above-mentioned step of determining the target loss function according to the predicted score value corresponding to the model training data and the click-through rate of the historical advertising video file to which the model training data belongs, and specifically implements the following steps:

[0213] Convert the click-through rate of the historical advertising video file to which the model training data belongs into a target score value;

[0214] Construct the target loss function based on the predicted score value and target score value corresponding to the model training data.

[0215] In one embodiment, when the computer program is executed by the processor to implement the above-mentioned step of controlling the advertisement delivery of the to-be-rated video file according to the rating result, the following steps are specifically implemented:

[0216] According to the scoring results, determine the bid price of the video file to be scored and the advertisement placement position;

[0217] Advertisements are delivered to the video file to be rated according to the bids of the video file to be rated and the advertisement delivery positions.

[0218] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0219] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0220] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for controlling advertisement delivery of a video file, characterized in that: The method comprises: Get the video file to be rated; Determine the scoring result of the video file to be scored according to the video file to be scored and a pre-trained scoring model, wherein the scoring model is trained based on the advertisement delivery data of each historical advertisement video file, and the advertisement delivery data includes the click rate of each historical advertisement video file, the included hot word information, the video category information and the credit data; Controlling advertisement delivery of the video file to be rated according to the rating result; The training process of the scoring model includes: The click rate, hot word information, and video category information of each of the historical advertising video files are sequentially input as model training data into the original scoring model to obtain a predicted score value corresponding to each of the historical advertising video files, wherein the original scoring model includes model parameters, and the model parameters include a preset score value corresponding to the hot word information and a preset score value corresponding to the video category information, and the predicted score value is predicted based on the preset score value corresponding to the hot word information and the preset score value corresponding to the video category information; Determine a target loss function according to the predicted score value corresponding to each of the historical advertising video files, the click-through rate of each of the historical advertising video files, and the credit data, wherein the target loss function is used to measure the degree of mismatch between the scoring result and the click-through rate and the credit data; Performing model training, and continuously adjusting the preset score values ​​corresponding to the hot word information and the preset score values ​​corresponding to the video category information during the training process to adjust the value of the target loss function; When the value of the objective loss function reaches a preset training stop condition, the scoring model is obtained.

2. The method according to claim 1, characterized in that Determining the scoring result of the video file to be scored according to the video file to be scored and the pre-trained scoring model includes: Extracting text information from each key frame in the to-be-rated video file, and determining video category information of the to-be-rated video file according to the text information; Inputting the video category information of the to-be-rated video file into the rating model, and obtaining a first rating value predicted by the rating model for the input video category information; Extracting hot word information from the text information, inputting the extracted hot word information into the scoring model, and obtaining a second scoring value predicted by the scoring model for the input hot word information; The scoring result is determined by the scoring model and according to the first scoring value and the second scoring value.

3. The method according to claim 2, characterized in that The determining of the target loss function according to the predicted score value corresponding to the model training data and the click rate of the historical advertisement video file to which the model training data belongs includes: Converting the click-through rate of the historical advertisement video file to which the model training data belongs into a target score value; The target loss function is constructed according to the predicted score value corresponding to the model training data and the target score value.

4. The method according to claim 1, characterized in that: The controlling the advertisement delivery of the to-be-rated video file according to the rating result comprises: Determining the bid price and advertisement placement position of the video file to be rated according to the rating result; Advertisements are placed on the video file to be rated according to the bid price of the video file to be rated and the advertisement placement position.

5. A video file advertising control device, characterized in that: The device comprises: An acquisition module is used to obtain the video file to be rated; A scoring module, used to determine the scoring result of the video file to be scored according to the video file to be scored and a pre-trained scoring model, wherein the scoring model is trained based on the advertisement delivery data of each historical advertisement video file, and the advertisement delivery data includes the click rate of each historical advertisement video file, the included hot word information, the video category information and the credit data; A control module is used to control the advertising delivery of the video file to be scored according to the scoring result, wherein the click rate, hot word information, video category information and credit data of each of the historical advertising video files are sequentially input into the original scoring model as model training data to obtain the predicted score value corresponding to each of the historical advertising video files, the original scoring model includes model parameters, the model parameters include a preset score value corresponding to the hot word information and a preset score value corresponding to the video category information, the predicted score value is predicted based on the preset score value corresponding to the hot word information and the preset score value corresponding to the video category information, according to the predicted score value corresponding to each of the historical advertising video files, the click rate of each of the historical advertising video files and the credit data, determine the target loss function, the target loss function is used to measure the mismatch between the scoring result and the click rate and the credit data, perform model training, continuously adjust the preset score value corresponding to the hot word information and the preset score value corresponding to the video category information during the training process to adjust the value of the target loss function, and obtain the scoring model when the value of the target loss function reaches the preset training stop condition.

6. A computer 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, the steps of the method according to any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images