Method and system for quickly slicing short videos with high complete playing rate in batches

Abstract

The invention relates to the technical field of short videos, and discloses a method and a system for quickly slicing short videos in batches with a high complete playing rate, and the method comprises the steps: obtaining hotspot keywords of an Internet platform according to a fixed time interval in a first preset time period, and obtaining hotspot videos corresponding to the hotspot keywords; constructing a video material library, wherein the video material library comprises a text unit, a dubbing unit and a dubbing unit; the text unit is used for generating texts for the hot videos, the dubbing unit is used for generating dubbles for the hot videos, and the dubbing unit is used for generating dubbles according to the generated texts; based on the hotspot video, constructing a random population and an optimization population in sequence, and obtaining a target individual according to the optimization population; a plurality of target individuals and corresponding target function values are obtained, the target individuals and the corresponding hotspot keywords are combined to serve as training samples, the corresponding target function values serve as sample labels, and an individual prediction model is obtained through training; and in a second preset time period, obtaining the hotspot keyword.

Description

Technical Field

[0001] The present invention relates to the field of short video technology, and more specifically, to a method and system for quickly batch-forming short videos with a high completion rate. Background Art

[0002] With the rapid development of the new media industry, short videos have become an important tool for traffic monetization due to their low production costs, fast dissemination speed, and diverse content forms. Companies are increasingly relying on the number of short video views and completion rates as the core means of traffic monetization.

[0003] Existing methods for batch-generating short videos mainly use templates and automation technologies to simply splice video, images, text, audio and other materials to quickly produce a large number of short videos. However, this fixed template approach is difficult to adapt to the needs of the target market, and the generated content lacks personalization and targeting, and cannot effectively attract target users, resulting in low completion rates and unsatisfactory traffic monetization effects. Especially in the fierce market traffic competition, this approach is difficult to meet the traffic monetization needs of new media companies. Summary of the invention

[0004] The present invention provides a method and system for quickly batch-forming short videos with a high completion rate, which solves the technical problems raised in the background technology.

[0005] The present invention provides a method for quickly batch-forming short videos with a high completion rate, comprising:

[0006] Step 1, within a first preset time period, hot keywords of the Internet platform are obtained at fixed time intervals, and hot videos corresponding to the hot keywords are obtained;

[0007] Step 2, constructing a video material library, which includes a text unit, a music unit and a dubbing unit; the text unit is used to generate text for hot videos, the music unit is used to generate music for hot videos, and the dubbing unit is used to generate dubbing according to the generated text;

[0008] Step 3: Based on the hot videos, a random population and an optimal population are constructed in sequence, and the target individual is obtained according to the optimal population;

[0009] Step 4: loop step 3 to obtain several target individuals and corresponding objective function values, use the target individuals and corresponding hot keywords as training samples, and use the corresponding objective function values ​​as sample labels to train and obtain an individual prediction model;

[0010] Step 5, within the second preset time period, obtain hot keywords, and based on the genetic algorithm and the individual prediction model, obtain the editing method of the hot video to generate the corresponding short video.

[0011] Further, the text unit, including video parsing tools and insertion tools, includes:

[0012] Video parsing tool, used to input hot videos and extract the dialogue text of hot videos;

[0013] Insert tool, used to insert connecting text between dialogue texts. Connecting text includes:

[0014] Step 21, obtaining the hot text corresponding to the hot keyword, and disassembling the hot text based on the language model to obtain the event cause text, process text and result text of the hot text;

[0015] Step 22, obtaining the timestamp corresponding to the dialogue text of the hot video, if the time interval between adjacent dialogue texts is greater than a preset time, the interval between adjacent dialogue texts is used as an interval to be inserted;

[0016] Step 23, mapping the cause text, process text and result text, as well as the dialogue text inserted at both ends of the interval to be selected, into word vectors respectively, and calculating the relevance of the cause text, process text and result text to the interval to be selected based on the word vectors, the calculation formula is as follows: ;

[0017] in, Indicates the degree of association between the cause text, process text or result text and the interval to be inserted. Word vectors representing cause text, process text, or result text, The word vector representing the dialogue text inserted in front of the selected interval, The word vector representing the dialogue text inserted after the selected interval, , and represent the first weight parameter, the second weight parameter and the third weight parameter respectively, represents the Euclidean norm, represents the inner product. It means to find the square of Euclidean norm;

[0018] Step 24, respectively determine the insertion candidate intervals with the maximum correlation of the cause text, the process text and the result text, and assign the cause text, the process text and the result text as the connection texts to the insertion candidate intervals with the maximum correlation.

[0019] Furthermore, the dubbing unit and the music unit are as follows:

[0020] Obtain open source music materials and dubbing timbres, and cluster the music materials and dubbing timbres based on preset emotional styles to obtain several emotional music materials and several emotional dubbing timbres;

[0021] Emotional recognition is performed on hot keywords to obtain the emotional features of the hot keywords, and the corresponding emotional music materials and emotional dubbing timbres are matched based on the emotional features.

[0022] Furthermore, a random population is constructed, including:

[0023] The open-source soundtrack materials and dubbing timbres are represented by real number codes;

[0024] The connecting text is converted into a new style through a language model, and the style is represented by a real number code;

[0025] According to the hot videos, initialize several random individuals that meet the constraints. The random individuals are encoded as follows:

[0026] ;in, represents the i-th style of writing, represents the i-th soundtrack material, represents the i-th dubbing timbre;

[0027] The constraints include: Emotional music materials corresponding to hot videos, The emotional dubbing tone corresponding to the hot videos;

[0028] Construct an objective function, which is used to evaluate the short videos obtained by randomly editing hot videos. The formula of the objective function is as follows: ; ; ;

[0029] in, represents the objective function value, represents the time weight, It represents the lifeline of short videos. , and Respectively represent the lifeline in , and A vector of time points, It represents the ratio of the number of likes to the number of views at the hth time point. It represents the ratio of the number of completed broadcasts to the number of broadcasts at the hth time point. represents the cross product of vectors, , and denote curvature weight, weighted playback weight and interaction weight respectively, , and are not 0, and , and The sum is 1, Indicates the preset end time point of the lifeline. Indicates the starting time point of the lifeline. represents the hth time point of the lifeline, The gth time point of the lifeline, Indicates the number of lifeline time points, The first index representing the number of lifeline time points, A second index indicating the number of lifeline time points, Not equal to , The vector representing the lifeline at the nth time point;

[0030] Set the target threshold and calculate the objective function value for each random individual. If the objective function value of the random individual is greater than the target threshold, the corresponding random individual will be regarded as a high-quality individual.

[0031] Furthermore, the optimal population is constructed, including:

[0032] The optimal population is generated based on the encoding initialization of several high-quality individuals, as follows:

[0033] The writing style codes, music material codes and dubbing timbre codes of high-quality individuals are combined respectively to obtain a writing style coding library, a music material coding library and a dubbing timbre coding library;

[0034] The style codes of the optimal individuals of the optimal population are randomly selected from the style code library, the music material codes are randomly selected from the music material code library, and the dubbing timbre codes are randomly selected from the dubbing timbre code library, so as to obtain a number of optimal individuals;

[0035] The objective function values ​​of several optimizing individuals are calculated based on the objective function, and the optimizing individual with the largest objective function value is taken as the target individual.

[0036] Furthermore, individual prediction models are trained, including:

[0037] Mapping the hot keywords obtained in the first preset time period into hot word vectors, and merging the hot word vectors and the corresponding target individual codes to obtain feature vectors;

[0038] The feature vector is input into the individual prediction model, and the output of the individual prediction model represents the prediction evaluation score of the short video obtained by editing the hot video corresponding to the hot keyword according to the target individual coding;

[0039] The mean error variance of the predicted evaluation score and the objective function value corresponding to the target individual is used as the loss function of the individual prediction model, and the hyperparameters of the hidden layer of the individual prediction model are updated through back propagation.

[0040] Furthermore, the hidden layer of the individual prediction model is as follows: ; ;

[0041] in, Indicates that the eigenvector is The mapping output of the hidden layer is Indicates that the eigenvector is The mapping output of the hidden layer is express Activation function, represents element-wise multiplication, Indicates based on The left diagonal matrix constructed by As the left diagonal element of the matrix, the remaining elements are all 0 sparse matrix, Indicates based on The right diagonal matrix constructed by As the right diagonal elements of the matrix, the remaining elements are all 0 sparse matrix, Indicates The feature space of hidden layers, Indicates The weight parameters of the feature space of the hidden layers, Indicates The bias parameters of the feature space of the hidden layers, Represents a transpose operation.

[0042] Furthermore, based on the genetic algorithm and the individual prediction model, the editing method of the hot videos is obtained, including:

[0043] Step 81: Map the hot keywords obtained in the second preset time period into hot word vectors ;

[0044] Step 82, initialize a number of genetic individuals, the genetic individuals are represented as: ;in, , and Respectively represent Type of writing style Music material and A kind of dubbing tone;

[0045] Step 83, using the individual prediction model as the genetic fitness function of the genetic population, and obtaining the genetic fitness function value of each genetic individual through the individual prediction model;

[0046] Step 84, based on the genetic fitness function value, sort each genetic individual from large to small to obtain a feature sorting;

[0047] Step 85, retaining a preset number of genetic individuals in the feature sorting from front to back, and updating the remaining individuals as parents, the updating includes:

[0048] For each parent generation, a random number X is drawn within a fixed value range. If X is greater than a first preset threshold value P1, a parent generation is randomly selected for one-way crossover;

[0049] If X is less than the second preset threshold value P2, the parent generation is mutated;

[0050] If P2<X<P1, then suspend updating the parent generation;

[0051] Step 86, when the number of steps 84 and 85 is repeated for a preset number of times, the updating is stopped, and the genetic individual with the largest genetic fitness function value is output. According to the genetic individual, the editing method of the hot video corresponding to the hot keyword is obtained, and the corresponding short video is generated.

[0052] A system for quickly batch-forming short videos with a high completion rate, applied to the method described, comprising:

[0053] A data collection module, used to obtain hot keywords of the Internet platform at fixed time intervals within a first preset time period, and obtain hot videos corresponding to the hot keywords;

[0054] The clip library module is used to build a video material library, which includes a text unit, a music unit and a dubbing unit; the text unit is used to generate text for hot videos, the music unit is used to generate music for hot videos, and the dubbing unit is used to generate dubbing according to the generated text;

[0055] The population optimization module is used to construct random populations and optimal populations in sequence based on hot videos, and obtain target individuals based on the optimal populations;

[0056] The individual prediction module is used to obtain several target individuals and corresponding objective function values, and use the target individuals and corresponding hot keywords as training samples, and use the corresponding objective function values ​​as sample labels to train and obtain an individual prediction model;

[0057] The editing planning module obtains hot keywords within a second preset time period, and obtains the editing method of the hot video based on the genetic algorithm and the individual prediction model to generate the corresponding short video.

[0058] The beneficial effect of the present invention is that by intelligently extracting Internet hot content and optimizing the editing method of short videos in combination with genetic algorithms and prediction models, personalized short videos that are highly consistent with user interests and hot trends can be quickly generated. This method not only improves the completion rate of short videos, but also significantly enhances their commercial conversion capabilities, adapts to the current market demand for fierce traffic competition, and provides strong support for enterprises to achieve efficient traffic monetization. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flow chart of a method for quickly batch-forming short videos with a high completion rate of the present invention. DETAILED DESCRIPTION

[0060] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0061] like Figure 1 As shown, a method for quickly batching short videos with a high completion rate includes:

[0062] Step 1, within a first preset time period, hot keywords of the Internet platform are obtained at fixed time intervals, and hot videos corresponding to the hot keywords are obtained;

[0063] Step 2, constructing a video material library, which includes a text unit, a music unit and a dubbing unit; the text unit is used to generate text for hot videos, the music unit is used to generate music for hot videos, and the dubbing unit is used to generate dubbing according to the generated text;

[0064] Step 3: Based on the hot videos, a random population and an optimal population are constructed in sequence, and the target individual is obtained according to the optimal population;

[0065] Step 4: loop step 3 to obtain several target individuals and corresponding objective function values, use the target individuals and corresponding hot keywords as training samples, and use the corresponding objective function values ​​as sample labels to train and obtain an individual prediction model;

[0066] Step 5, within the second preset time period, obtain hot keywords, and based on the genetic algorithm and the individual prediction model, obtain the editing method of the hot video to generate the corresponding short video.

[0067] In one embodiment of the present invention, the text unit, including the video parsing tool and the insertion tool, includes:

[0068] Video parsing tool, used to input hot videos and extract the dialogue text of hot videos;

[0069] Insert tool, used to insert connecting text between dialogue texts. Connecting text includes:

[0070] Step 21, obtaining the hot text corresponding to the hot keyword, and disassembling the hot text based on the language model to obtain the event cause text, process text and result text of the hot text;

[0071] Step 22, obtaining the timestamp corresponding to the dialogue text of the hot video, if the time interval between adjacent dialogue texts is greater than a preset time, the interval between adjacent dialogue texts is used as an interval to be inserted;

[0072] Step 23, mapping the cause text, process text and result text, as well as the dialogue text inserted at both ends of the interval to be selected, into word vectors respectively, and calculating the relevance of the cause text, process text and result text to the interval to be selected based on the word vectors, the calculation formula is as follows: ;

[0073] in, Indicates the degree of association between the cause text, process text or result text and the interval to be inserted. Word vectors representing cause text, process text, or result text, The word vector representing the dialogue text inserted in front of the selected interval, The word vector representing the dialogue text inserted after the selected interval, , and represent the first weight parameter, the second weight parameter and the third weight parameter respectively, represents the Euclidean norm, represents the inner product. It means to find the square of Euclidean norm;

[0074] Step 24, respectively determine the insertion candidate intervals with the maximum correlation of the cause text, the process text and the result text, and assign the cause text, the process text and the result text as the connection texts to the insertion candidate intervals with the maximum correlation.

[0075] It should be noted that the dubbing tone is configured in the connecting text, that is, the connecting text is played through the dubbing tone.

[0076] Specifically, in the dissemination of short videos, hot videos often lack sufficient information background, and it is difficult for the audience to fully understand the cause, process and result of the event only by pictures and scattered dialogues. This may lead to insufficient understanding of the video content by the audience, thereby affecting the viewing experience and the completion rate of the video. To solve this problem, this embodiment generates and inserts connecting texts to supplement and improve the logic of the video content, providing the audience with a clearer context of events. The dialogue text is the core part for the audience to directly obtain information from the video, and its extraction and timestamp annotation provide a basis for the subsequent planning of text insertion. Obtain hot texts related to hot keywords (such as news reports, event introductions, etc.). Based on the language model, these texts are structured and disassembled into three parts: "cause of the event", "process" and "result". Each part of the text corresponds to the background, main process and final result of the event, ensuring clear logic and complete information. According to the dialogue text timestamp obtained by the video parsing tool, the time interval between adjacent dialogues is analyzed. If the interval is greater than the set threshold (that is, there is an obvious breakpoint between the dialogues), the interval is marked as a potential insertion point. The generated cause text, process text, result text, and the dialogue text at both ends of the interval to be inserted are converted into word vectors respectively. The semantic correlation between the linking text and the context dialogue text is calculated using a mathematical model to ensure that the inserted linking text is consistent with the semantic logic before and after. The correlation is calculated for the cause, process, and result texts respectively, and the interval with the highest correlation is selected as the insertion point. According to the time logic of the video, the three types of linking text (cause, process, result) are inserted into the best position respectively to make the video logic clear and the transition natural.

[0077] It should be noted that the video parsing tool can be Whisper, and the language model can be Chatgpt.

[0078] For example, if there are 10 dialogue texts, then the intervals of 9 dialogue texts are obtained. If it is determined that the intervals of 4 dialogue texts are greater than 10 seconds, then 4 insertion candidate intervals are obtained. The cause text, process text and result text are inserted into the intervals of the 4 dialogue texts. Based on the calculation of the correlation degree, it is obtained that the cause text matches the first insertion candidate interval, the process text matches the third insertion candidate interval, and the result text matches the fourth insertion candidate interval, then the cause text, process text and result text are inserted according to the matching results.

[0079] In one embodiment of the present invention, the dubbing unit and the music unit are as follows:

[0080] Obtain open source music materials and dubbing timbres, and cluster the music materials and dubbing timbres based on preset emotional styles to obtain several emotional music materials and several emotional dubbing timbres;

[0081] Emotional recognition is performed on hot keywords to obtain the emotional features of the hot keywords, and the corresponding emotional music materials and emotional dubbing timbres are matched based on the emotional features.

[0082] Specifically, the dubbing and soundtrack of short videos are important factors that affect the user's viewing experience. Good sound effects can not only enhance the emotional appeal of the video, but also effectively improve the audience's attention and participation. The present invention achieves a high degree of emotional fit between short video content and sound effects through intelligent emotional classification of soundtrack materials and dubbing timbre, and accurate matching with the emotional characteristics of hot keywords, thereby enhancing the attractiveness and completion rate of the video. By collecting open source soundtrack material libraries (such as audio clips provided by open source platforms) and dubbing timbre (such as sound samples of different voices and intonations in the timbre library), rich sound effect resources are provided. These materials are classified and standardized to ensure that subsequent models can be used efficiently. Emotional classification standards, music emotion classification: such as happy, sad, passionate, calm and other categories. Dubbing timbre classification: such as lively, solemn, humorous, sensational and other categories. Use emotion recognition algorithms (which can be based on deep learning Autoencoder) to extract and cluster the features of soundtrack materials and dubbing timbre. Feature extraction can be based on audio signal processing (such as pitch, rhythm, timbre) and emotional labeling (such as manual or algorithmic emotional labeling of audio). Several emotional soundtrack material groups and dubbing timbre groups are obtained, and these groups can be directly used for emotional matching with video content. Emotional recognition is performed on hot keywords, and keywords extracted from short videos or hot events are input. Emotional analysis tools use the sentiment analysis model (Chatgpt language model) based on NLP (natural language processing) to classify keywords by sentiment. The classification results can be multiple emotional labels (such as positive emotions, negative emotions, neutral emotions), or more fine-grained emotional types (such as anger, joy, sadness, etc.). Emotional feature extraction uses the emotional features of keywords (such as positive and negative tendencies or specific emotional types) as matching basis to generate the most appropriate sound effects for video content. According to the emotional features of keywords, emotionally consistent materials are selected from pre-clustered soundtrack materials and dubbing timbres. For example, if the emotional feature of the keyword is "joy", then "brisk music" and "lively voiceover timbre" are matched; if the emotional feature is "sadness", then "soft music" and "deep voiceover timbre" are matched. Add the matching music material and voiceover timbre to the short video to make the emotional atmosphere of the overall audio and video content more consistent.

[0083] In one embodiment of the present invention, constructing a random population includes:

[0084] The open-source soundtrack materials and dubbing timbres are represented by real number codes;

[0085] The connecting text is converted into a new style through a language model, and the style is represented by a real number code;

[0086] According to the hot videos, initialize several random individuals that meet the constraints. The random individuals are encoded as follows:

[0087] ;in, represents the i-th style of writing, represents the i-th soundtrack material, represents the i-th dubbing timbre;

[0088] The constraints include: Emotional music materials corresponding to hot videos, The emotional dubbing tone corresponding to the hot videos;

[0089] Construct an objective function, which is used to evaluate the short videos obtained by randomly editing hot videos. The formula of the objective function is as follows: ; ; ;

[0090] in, represents the objective function value, represents the time weight, It represents the lifeline of short videos. , and Respectively represent the lifeline in , and A vector of time points, It represents the ratio of the number of likes to the number of views at the hth time point. It represents the ratio of the number of completed broadcasts to the number of broadcasts at the hth time point. represents the cross product of vectors, , and denote curvature weight, weighted playback weight and interaction weight respectively, , and are not 0, and , and The sum is 1, Indicates the preset end time point of the lifeline. Indicates the starting time point of the lifeline. represents the hth time point of the lifeline, The gth time point of the lifeline, Indicates the number of lifeline time points, The first index representing the number of lifeline time points, A second index indicating the number of lifeline time points, Not equal to , The vector representing the lifeline at the nth time point;

[0091] Set the target threshold and calculate the objective function value for each random individual. If the objective function value of the random individual is greater than the target threshold, the corresponding random individual will be regarded as a high-quality individual.

[0092] Specifically, the construction of a random population of the insertion tool is an experimental means to explore the characteristics of hot video editing, which can cover a wide range of possibilities and verify the impact of editing parameters on the quality of short videos. The performance of random individuals is quantified through the objective function to provide an experimental basis for subsequent optimization and production. The experimental results of random individuals can be used as a solution set for the optimal population, which can be used for subsequent training of editing optimization models to promote the intelligence of the video production process. Since the random population of the insertion tool does not rely on editing experience, it provides more objective support for short video editing strategies.

[0093] In one embodiment of the present invention, constructing an optimal population includes:

[0094] The optimal population is generated based on the encoding initialization of several high-quality individuals, as follows:

[0095] The writing style codes, music material codes and dubbing timbre codes of high-quality individuals are combined respectively to obtain a writing style coding library, a music material coding library and a dubbing timbre coding library;

[0096] The style codes of the optimal individuals of the optimal population are randomly selected from the style code library, the music material codes are randomly selected from the music material code library, and the dubbing timbre codes are randomly selected from the dubbing timbre code library, so as to obtain a number of optimal individuals;

[0097] The objective function values ​​of several optimizing individuals are calculated based on the objective function, and the optimizing individual with the largest objective function value is taken as the target individual.

[0098] Specifically, in the previous stage of the present invention, by constructing a random population and using the objective function evaluation, several high-quality individuals that meet specific conditions have been screened out. However, the random population is inherently random. Although it covers a wide range of possible spaces, it does not fully explore the optimal editing combination. In order to further improve the quality of short video editing, an optimal population is constructed, aiming to find better target individuals based on high-quality individuals and through specific optimization strategies on the basis of existing high-quality combinations. In the experiment of the previous stage (random population), the random individuals with higher objective function values ​​screened out are marked as high-quality individuals. Three key codes are extracted from high-quality individuals: the style code corresponds to the style of the connecting text, such as humor, formality, narrative, etc. The music material code corresponds to the background music type. The dubbing timbre code corresponds to the timbre type of the dubbing. The style code, music material code, and dubbing timbre code extracted from the high-quality individuals are respectively constructed into three independent coding libraries. These coding libraries cover the parameter selections that perform better in high-quality individuals, ensuring that proven combinations are given priority in subsequent optimization. Each optimal individual in the optimal population is randomly generated from three coding libraries, so that the newly generated optimal population is based on the combination of high-quality individuals while retaining a certain degree of diversity. All optimal individuals are sorted by the objective function value, and the optimal individual with the largest objective function value is selected as the final target individual.

[0099] The target individual represents the optimal short video editing solution among the hot keywords.

[0100] In one embodiment of the present invention, training to obtain an individual prediction model includes:

[0101] Mapping the hot keywords obtained in the first preset time period into hot word vectors, and merging the hot word vectors and the corresponding target individual codes to obtain feature vectors;

[0102] The feature vector is input into the individual prediction model, and the output of the individual prediction model represents the prediction evaluation score of the short video obtained by editing the hot video corresponding to the hot keyword according to the target individual coding;

[0103] The mean error variance of the predicted evaluation score and the objective function value corresponding to the target individual is used as the loss function of the individual prediction model, and the hyperparameters of the hidden layer of the individual prediction model are updated through back propagation.

[0104] Specifically, in the process of short video generation and optimization, the quality of the editing scheme often needs to be measured by indicators such as the actual number of viewers playing and the interaction rate after the video is generated. This method is inefficient and has a high time cost. In order to improve efficiency, the present invention proposes an individual prediction model, which predicts the quality of short videos with given hot keywords and target individual codes by training historical data. This model can significantly reduce the cost of trial and error and provide prediction support for the intelligent editing of short videos. The hot keywords obtained in the first preset time period are converted into numerical representations, i.e., hot word vectors, through a word embedding model (such as Word2Vec, BERT or other word vector generation methods). Unstructured text data (keywords) are converted into numerical features suitable for processing by machine learning models. Each target individual (editing scheme) consists of a style code, a soundtrack material code, and a dubbing timbre code. The codes of the target individual are merged in a numerical form to form a complete target individual code. The hot word vector and the target individual code are spliced ​​to form a complete feature vector. The feature vector represents the combined characteristics of the hot keywords and the target individual editing scheme, and is the input of the prediction model. The individual prediction model can be a deep learning-based model (MLP), which can ensure that the model can accurately predict the actual quality of the short video corresponding to the target individual by minimizing the loss function. The model parameters are updated using the backpropagation algorithm. The optimizer can select a commonly used deep learning optimization algorithm (such as Adam, SGD, etc.). The backpropagation algorithm is used to adjust the weights and biases of the model's hidden layer so that the model gradually converges to a better state.

[0105] In one embodiment of the present invention, the hidden layer of the individual prediction model is as follows: ; ;

[0106] in, Indicates that the eigenvector is The mapping output of the hidden layer is Indicates that the eigenvector is The mapping output of the hidden layer is express Activation function, represents element-wise multiplication, Indicates based on The left diagonal matrix constructed by As the left diagonal element of the matrix, the remaining elements are all 0 sparse matrix, Indicates based on The right diagonal matrix constructed by As the right diagonal elements of the matrix, the remaining elements are all 0 sparse matrix, Indicates The feature space of hidden layers, Indicates The weight parameters of the feature space of the hidden layers, Indicates The bias parameters of the feature space of the hidden layers, Represents a transpose operation.

[0107] Specifically, in the individual prediction model, the hidden layer is designed to map the input feature vector (including the word vector of the hot keyword and the target individual code) to a high-dimensional feature space, extract key features, and then accurately predict the quality of short video clips. The hidden layer converts the feature vector into a mapping output of the feature space through a series of matrix operations and nonlinear activation functions, and dynamically adjusts the model parameters during the mapping process to optimize the prediction results. The input feature vector (composed of the word vector of the hot keyword and the genetic individual code) is mapped to a high-dimensional feature space after being processed by the hidden layer. The role of the left diagonal matrix and the right diagonal matrix The introduction of the left diagonal matrix and the right diagonal matrix is ​​to sparse the mapping process, that is, to perform feature transformation only in a specific dimension, while ignoring features that are not related to the prediction. This sparse operation helps the model to more accurately extract features that are highly correlated with hot keywords and genetic individual codes. In the hidden layer mapping process, the model will focus on the relationship between the word vector of the hot keyword and the target individual code, and then predict the evaluation score of the short video editing plan. When the genetic individual of the input hot keyword is significantly different from the target individual of the similar hot keyword in the first preset time period, the mapping result of the feature vector will deviate from the high-quality area in the feature space. Since the model has learned the best matching relationship between hot keywords and genetic individuals through historical data during training, when it deviates from this similarity, the model will give a lower prediction evaluation score. The mapping of the hidden layer has learned which feature combinations are of high quality. When it deviates from historical similar keywords, it is difficult for the model to determine the performance potential of the input scheme. However, when the genetic individual of the input hot keyword is closer to the target individual of the similar hot keyword in history, the model will not necessarily give a high prediction score directly. Although the similarity of the feature vector may improve the evaluation score, the model also examines other features, such as subtle changes in keywords, combinations of target individual encodings, and overall matching with the context. In the prediction process, the model not only relies on feature similarity, but also comprehensively considers the overall quality of the feature vector.

[0108] In one embodiment of the present invention, based on a genetic algorithm and an individual prediction model, a method of editing a hot video is obtained, including:

[0109] Step 81: Map the hot keywords obtained in the second preset time period into hot word vectors ;

[0110] Step 82, initialize a number of genetic individuals, the genetic individuals are represented as: ;in, , and Respectively represent Type of writing style Music material and A kind of dubbing tone;

[0111] Step 83, using the individual prediction model as the genetic fitness function of the genetic population, and obtaining the genetic fitness function value of each genetic individual through the individual prediction model;

[0112] Step 84, based on the genetic fitness function value, sort each genetic individual from large to small to obtain a feature sorting;

[0113] Step 85, retaining a preset number of genetic individuals in the feature sorting from front to back, and updating the remaining individuals as parents, the updating includes:

[0114] For each parent generation, a random number X is drawn within a fixed value range. If X is greater than a first preset threshold value P1, a parent generation is randomly selected for one-way crossover;

[0115] If X is less than the second preset threshold value P2, the parent generation is mutated;

[0116] If P2<X<P1, then suspend updating the parent generation;

[0117] Step 86, when the number of steps 84 and 85 is repeated for a preset number of times, the updating is stopped, and the genetic individual with the largest genetic fitness function value is output. According to the genetic individual, the editing method of the hot video corresponding to the hot keyword is obtained, and the corresponding short video is generated.

[0118] Specifically, in the optimization of short video editing, since multiple variables (writing style, music, dubbing, etc.) are involved, it is inefficient and infeasible to directly find the optimal combination through exhaustive method. Therefore, the present invention combines genetic algorithm with individual prediction model to simulate the selection and optimization process of biological evolution, and automatically generates high-quality short video editing schemes based on hot keywords. The genetic algorithm uses the method of population evolution to gradually optimize the initial genetic individuals (i.e., editing schemes) through operations such as selection, crossover, and mutation, and combines the individual prediction model as an adaptation function to guide the optimization direction, and finally obtains the best performing editing method.

[0119] A method system for quickly batch-forming short videos with a high completion rate, applied to the method described, comprising:

[0120] A data collection module, used to obtain hot keywords of the Internet platform at fixed time intervals within a first preset time period, and obtain hot videos corresponding to the hot keywords;

[0121] The clip library module is used to build a video material library, which includes a text unit, a music unit and a dubbing unit; the text unit is used to generate text for hot videos, the music unit is used to generate music for hot videos, and the dubbing unit is used to generate dubbing according to the generated text;

[0122] The population optimization module is used to construct random populations and optimal populations in sequence based on hot videos, and obtain target individuals based on the optimal populations;

[0123] The individual prediction module is used to obtain several target individuals and corresponding objective function values, and use the target individuals and corresponding hot keywords as training samples, and use the corresponding objective function values ​​as sample labels to train and obtain an individual prediction model;

[0124] The editing planning module obtains hot keywords within a second preset time period, and obtains the editing method of the hot video based on the genetic algorithm and the individual prediction model to generate the corresponding short video.

[0125] The embodiments of the present embodiment have been described, but the present embodiment is not limited to the above-mentioned specific implementation modes. The above-mentioned specific implementation modes are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, which are all within the protection of the present embodiment.

Claims

1. A method for quickly batch-forming short videos with a high completion rate, characterized in that: include: Step 1, within a first preset time period, obtaining hot keywords of the Internet platform at fixed time intervals, and obtaining hot videos corresponding to the hot keywords; Step 2, constructing a video material library, which includes a text unit, a music unit and a dubbing unit; the text unit is used to generate text for hot videos, the music unit is used to generate music for hot videos, and the dubbing unit is used to generate dubbing according to the generated text; Step 3: Based on the hot videos, a random population and an optimal population are constructed in sequence, and the target individual is obtained according to the optimal population; Step 4: loop step 3 to obtain several target individuals and corresponding objective function values, combine the target individuals and the corresponding hot keywords as training samples, and use the corresponding objective function values ​​as sample labels to train and obtain an individual prediction model; Step 5, within the second preset time period, obtain hot keywords, and based on the genetic algorithm and the individual prediction model, obtain the editing method of the hot video to generate the corresponding short video.

2. According to claim 1, a method for quickly batch-forming short videos with a high completion rate is characterized in that: Text unit, including video parsing tools and insertion tools, including: Video parsing tool, used to input hot videos and extract the dialogue text of hot videos; Insert tool, used to insert connecting text between dialogue texts. Connecting text includes: Step 21, obtaining the hot text corresponding to the hot keyword, and disassembling the hot text based on the language model to obtain the event cause text, process text and result text of the hot text; Step 22, obtaining the timestamp corresponding to the dialogue text of the hot video, if the time interval between adjacent dialogue texts is greater than a preset time, the interval between adjacent dialogue texts is used as an interval to be inserted; Step 23, mapping the cause text, process text and result text, as well as the dialogue text inserted at both ends of the interval to be selected, into word vectors respectively, and calculating the relevance of the cause text, process text and result text to the interval to be selected based on the word vectors, the calculation formula is as follows: ； in, Indicates the degree of association between the cause text, process text or result text and the interval to be inserted. Word vectors representing cause text, process text, or result text, The word vector representing the dialogue text inserted in front of the selected interval, The word vector representing the dialogue text inserted after the selected interval, , and represent the first weight parameter, the second weight parameter and the third weight parameter respectively, represents the Euclidean norm, represents the inner product. It means to find the square of Euclidean norm; Step 24, respectively determine the insertion candidate intervals with the maximum correlation of the cause text, the process text and the result text, and assign the cause text, the process text and the result text as the connection texts to the insertion candidate intervals with the maximum correlation.

3. The method for quickly batch-forming short videos with a high completion rate according to claim 1 is characterized in that: The dubbing unit and the music unit are as follows: Obtain open source music materials and dubbing timbres, and cluster the music materials and dubbing timbres based on preset emotional styles to obtain several emotional music materials and several emotional dubbing timbres; Emotional recognition is performed on hot keywords to obtain the emotional features of the hot keywords, and the corresponding emotional music materials and emotional dubbing timbres are matched based on the emotional features.

4. A method for quickly batch-forming short videos with a high completion rate according to claim 2 or 3, characterized in that: Construct a random population, including: The open-source soundtrack materials and dubbing timbres are represented by real number codes; The connecting text is converted into a new style through a language model, and the style is represented by a real number code; According to the hot videos, initialize several random individuals that meet the constraints. The random individuals are encoded as follows: ;in, represents the i-th style of writing, represents the i-th soundtrack material, represents the i-th dubbing timbre; The constraints include: Emotional music materials corresponding to hot videos, The emotional dubbing tone corresponding to the hot videos; Construct an objective function, which is used to evaluate the short videos obtained by randomly editing hot videos. The formula of the objective function is as follows: ； ； ； in, represents the objective function value, represents the time weight, It represents the lifeline of short videos. , and Respectively represent the lifeline in , and A vector of time points, It represents the ratio of the number of likes to the number of views at the hth time point. It represents the ratio of the number of completed broadcasts to the number of broadcasts at the hth time point. represents the cross product of vectors, , and denote curvature weight, weighted playback weight and interaction weight respectively, , and are not 0, and , and The sum is 1, Indicates the preset end time point of the lifeline. Indicates the starting time point of the lifeline. represents the hth time point of the lifeline, The gth time point of the lifeline, Indicates the number of lifeline time points, The first index representing the number of lifeline time points, A second index indicating the number of lifeline time points, Not equal to , Represents the vector of the lifeline at the nth time point; Set the target threshold and calculate the objective function value for each random individual. If the objective function value of the random individual is greater than the target threshold, the corresponding random individual will be regarded as a high-quality individual.

5. The method for quickly batch-forming short videos with a high completion rate according to claim 4 is characterized in that: Constructing an optimal population, including: The optimal population is generated based on the encoding initialization of several high-quality individuals, as follows: The writing style codes, music material codes and dubbing timbre codes of high-quality individuals are combined respectively to obtain a writing style coding library, a music material coding library and a dubbing timbre coding library; The style codes of the optimal individuals of the optimal population are randomly selected from the style code library, the music material codes are randomly selected from the music material code library, and the dubbing timbre codes are randomly selected from the dubbing timbre code library, so as to obtain a number of optimal individuals; The objective function values ​​of several optimizing individuals are calculated based on the objective function, and the optimizing individual with the largest objective function value is taken as the target individual.

6. The method for quickly batch-forming short videos with a high completion rate according to claim 5, characterized in that: The individual prediction model is trained, including: Mapping the hot keywords obtained in the first preset time period into hot word vectors, and merging the hot word vectors and the corresponding target individual codes to obtain feature vectors; The feature vector is input into the individual prediction model, and the output of the individual prediction model represents the prediction evaluation score of the short video obtained by editing the hot video corresponding to the hot keyword according to the target individual coding; The mean error variance of the predicted evaluation score and the objective function value corresponding to the target individual is used as the loss function of the individual prediction model, and the hyperparameters of the hidden layer of the individual prediction model are updated through back propagation.

7. The method for quickly batch-forming short videos with a high completion rate according to claim 6 is characterized in that: The hidden layers of the individual prediction model are as follows: ； ； in, Indicates that the feature vector is The mapping output of the hidden layer is Indicates that the eigenvector is The mapping output of the hidden layer is express Activation function, represents element-wise multiplication, Indicates based on The left diagonal matrix constructed by As the left diagonal element of the matrix, the remaining elements are all 0 sparse matrix, Indicates based on The right diagonal matrix constructed by As a sparse matrix with the right diagonal elements of the matrix and the rest of the elements are 0, Indicates The feature space of hidden layers, Indicates The weight parameters of the feature space of the hidden layers, Indicates The bias parameters of the feature space of the hidden layers, Represents a transpose operation.

8. The method for quickly batch-forming short videos with a high completion rate according to claim 7 is characterized in that: Based on genetic algorithms and individual prediction models, the editing methods of hot videos are obtained, including: Step 81: Map the hot keywords obtained in the second preset time period into hot word vectors ; Step 82, initialize several genetic individuals, the genetic individuals are represented as: ;in, , and Respectively represent Type of writing style Music material and A kind of dubbing tone; Step 83, using the individual prediction model as the genetic fitness function of the genetic population, and obtaining the genetic fitness function value of each genetic individual through the individual prediction model; Step 84, based on the genetic fitness function value, sort each genetic individual from large to small to obtain a feature sorting; Step 85, retaining a preset number of genetic individuals in the feature sorting from front to back, and updating the remaining individuals as parents, the updating includes: For each parent generation, a random number X is drawn within a fixed value range. If X is greater than a first preset threshold value P1, a parent generation is randomly selected for one-way crossover; If X is less than the second preset threshold value P2, the parent generation is mutated; If P2<X<P1, then suspend updating the parent generation; Step 86, when the number of steps 84 and 85 is repeated for a preset number of times, the updating is stopped, and the genetic individual with the largest genetic fitness function value is output. According to the genetic individual, the editing method of the hot video corresponding to the hot keyword is obtained, and the corresponding short video is generated.

9. A system for quickly batch-forming short videos with a high completion rate, characterized in that: The method according to any one of claims 1 to 8, comprising: A data collection module, used to obtain hot keywords of the Internet platform at fixed time intervals within a first preset time period, and obtain hot videos corresponding to the hot keywords; The clip library module is used to build a video material library, which includes a text unit, a music unit and a dubbing unit; the text unit is used to generate text for hot videos, the music unit is used to generate music for hot videos, and the dubbing unit is used to generate dubbing according to the generated text; The population optimization module is used to construct random populations and optimal populations in sequence based on hot videos, and obtain target individuals based on the optimal populations; The individual prediction module is used to obtain several target individuals and corresponding objective function values, and use the target individuals and corresponding hot keywords as training samples, and use the corresponding objective function values ​​as sample labels to train and obtain an individual prediction model; The editing planning module obtains hot keywords within a second preset time period, and obtains the editing method of the hot video based on the genetic algorithm and the individual prediction model to generate the corresponding short video.

Images