New media content value assessment method, device and computer equipment

Abstract

The present invention discloses a new media content value assessment method, device and computer equipment. The method comprises: obtaining new media content to be assessed; preprocessing the new media content to be assessed to obtain a preprocessing result; extracting features from the preprocessing result to obtain an extraction result; inputting the extraction result into a counterfactual causal inference model to predict content design strategies to obtain a prediction result; the counterfactual causal inference model is obtained by defining intervention measures and results, combining new media content with consumer feedback data, using a causal representation learning algorithm, constructing feature representation and optimizing errors, and is used to predict changes in consumer feedback under different content design strategies. A value assessment report is generated based on the prediction results; and the value assessment report is output. By implementing the method of the present invention, the accuracy and reliability of the value assessment of knowledge-sharing bloggers can be improved.

Description

Technical Field

[0001] The present invention relates to new media content, and more specifically to a new media content value assessment method, device and computer equipment. Background Art

[0002] New media refers to media forms developed based on digital technology, especially Internet technology, including social media, blogs, video platforms, podcasts, online news websites, etc. These platforms not only provide information dissemination, but also encourage interaction between users and content creation; the content shared by knowledge-sharing bloggers is also a type of new media content.

[0003] Knowledge-sharing bloggers are becoming increasingly important in new media content marketing, especially in the field of medical services. With their expertise and influence, they have become an important bridge between brands and consumers. Therefore, the value assessment of knowledge-sharing bloggers is particularly important. The current value assessment methods mainly rely on sales volume and fan interaction volume. Among them, sales volume is a direct indicator of the KOL's ability to bring goods and sales performance, and fan interaction includes the number of fans, likes, comments and sharing, which reflects the KOL's audience base and interactive activity. Although the existing evaluation methods reflect the market value of online influencers to a certain extent, knowledge-sharing bloggers often do not bring goods directly and lack available sales data. At the same time, data based on fan interaction may be manipulated by humans, resulting in distorted evaluation results. Moreover, the impact of the personalized characteristics of knowledge-sharing bloggers on their value, such as language expression style and emotional expression, is not fully considered.

[0004] Therefore, it is necessary to design a new method to improve the accuracy and reliability of the value assessment of knowledge sharing bloggers. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the prior art and provide a new media content value assessment method, device and computer equipment.

[0006] To achieve the above purpose, the present invention adopts the following technical solution: a new media content value assessment method, comprising:

[0007] Obtain new media content to be evaluated;

[0008] Preprocessing the new media content to be evaluated to obtain a preprocessing result;

[0009] Performing feature extraction on the preprocessing result to obtain an extraction result;

[0010] Inputting the extracted results into the counterfactual causal inference model to predict the content design strategy, so as to obtain the prediction results;

[0011] Generate a value assessment report based on the prediction results;

[0012] Outputting the value assessment report;

[0013] The counterfactual causal inference model is obtained by defining intervention measures and results, combining new media content and consumer feedback data, using a causal representation learning algorithm, constructing feature representation and optimizing errors, and is used to predict changes in consumer feedback under different content design strategies.

[0014] A further technical solution is: the preprocessing of the new media content to be evaluated to obtain a preprocessing result includes:

[0015] Video frame extraction and audio signal cleaning are performed on the new media content to be evaluated to obtain a preprocessing result.

[0016] A further technical solution is: performing feature extraction on the preprocessing result to obtain an extraction result, including:

[0017] Extracting basic attributes of the video in the preprocessing result and expressing them as a multi-dimensional feature vector to obtain video editing features;

[0018] Analyze the audio content of the video in the preprocessing result, extract Mel spectrum coefficients, Mel spectrum, chrominance frequency, spectral centroid, spectral contrast and tonality features, combine them into a feature matrix, calculate the cross-matching matrix between background music and vocal features, so as to obtain audio features;

[0019] Extracting dense optical flow and foreground motion area to represent dynamic features of the video in the preprocessing result, and extracting color features of the video in the preprocessing result by saturation, brightness, contrast and clarity of the video to obtain visual aesthetic features;

[0020] Extracting text information of the video in the preprocessing result through a pre-trained language model, and converting human voice information in the video into text to obtain content text features;

[0021] Extracting and learning emotion vectors from the audio features and the facial features in the video in the preprocessing results, and obtaining emotion features through supplementation and comparison integration;

[0022] The video clip features, the audio features, the visual aesthetic features, the content text features, and the emotion features are combined to obtain an extraction result.

[0023] A further technical solution is as follows: the training process of the counterfactual causal inference model includes:

[0024] define interventions and expected outcomes;

[0025] Obtain new media historical content and corresponding consumer feedback data to obtain initial data;

[0026] Learning the causal relationship representation of the initial data based on a causal representation learning algorithm to obtain a learning result;

[0027] Build a neural network model;

[0028] The neural network model is trained using the learning results to obtain a counterfactual causal inference model.

[0029] A further technical solution is: after obtaining the new media historical content and the corresponding consumer feedback data to obtain the initial data, it also includes:

[0030] The initial data is updated by using error minimization and regularization methods as well as adding penalty items.

[0031] Its further technical solution is: the neural network model adopts a fully connected ReLU layer structure and is optimized using the RMSProp algorithm to obtain a counterfactual causal inference model.

[0032] Its further technical solution is: the loss function used in the neural network model training process is optimized using the alternating sub-gradient descent method.

[0033] A further technical solution is that when the neural network model is trained, sparse re-weighting processing is performed on the input features.

[0034] The present invention also provides a new media content value assessment device, comprising:

[0035] A content acquisition unit, used to acquire new media content to be evaluated;

[0036] A preprocessing unit, used for preprocessing the new media content to be evaluated to obtain a preprocessing result;

[0037] A feature extraction unit, used for performing feature extraction on the preprocessing result to obtain an extraction result;

[0038] A prediction unit, used for inputting the extraction result into a counterfactual causal inference model to predict the content design strategy, so as to obtain a prediction result;

[0039] An evaluation report unit, used to generate a value evaluation report according to the prediction result;

[0040] An output unit is used to output the value assessment report.

[0041] The present invention further provides a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor implements the above method when executing the computer program.

[0042] The beneficial effects of the present invention compared with the prior art are as follows: the present invention collects all relevant new media content of the blogger to be evaluated, including articles, videos and interaction records, and organizes them to ensure the integrity and consistency of the data. The collected new media content is preprocessed to clean irrelevant information, remove noise, and convert the data into a format suitable for analysis to ensure data quality. Text analysis and sentiment analysis techniques are used to extract key features from the preprocessed data, such as content topics, audience engagement and sentiment tendencies, to obtain more representative feature data. The extracted features are input into a counterfactual causal inference model, and the model is used to analyze the potential impact of different content strategies on blogger performance, so as to obtain more accurate prediction results. A value assessment report is written based on the prediction results, providing a comprehensive analysis of the blogger's performance and corresponding improvement suggestions to ensure the accuracy and operability of the output results. The accuracy and reliability of the value assessment of knowledge-sharing bloggers are improved.

[0043] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0045] Figure 1 A schematic diagram of a flow chart of a new media content value assessment method provided by an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of a sub-process of a new media content value assessment method provided by an embodiment of the present invention;

[0047] Figure 3 A schematic diagram of a sub-process of a new media content value assessment method provided by an embodiment of the present invention;

[0048] Figure 4 A schematic block diagram of a new media content value assessment device provided by an embodiment of the present invention;

[0049] Figure 5 A schematic block diagram of a feature extraction unit of a new media content value assessment device provided by an embodiment of the present invention;

[0050] Figure 6A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0053] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0054] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0055] See also Figure 1 , Figure 1 A schematic flow chart of a new media content value assessment method provided in an embodiment of the present invention. The new media content value assessment method is applied in a server. New media content waiting for evaluation is collected through video and audio data of knowledge-sharing bloggers, and necessary preprocessing is performed. Then, video and audio features are extracted through computer vision and audio processing technology, and text features are extracted using natural language processing technology. Content design strategies with better feedback effects are learned in a causal relationship model based on counterfactual inference. Finally, a value assessment report is output and an intervention strategy is implemented, while the model and algorithm are continuously optimized to adapt to market changes. The overall process aims to improve the accuracy of value assessment of knowledge-sharing bloggers and support corporate digital marketing decisions.

[0056] Figure 1 Schematic diagram of the process of the new media content value assessment method provided by the embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S160.

[0057] S110: Obtain new media content to be evaluated.

[0058] In this embodiment, the new media content to be evaluated includes videos and live broadcast content of knowledge sharing bloggers, including video files, audio tracks, etc.

[0059] Specifically, you can determine the platform for data collection, such as YouTube, Bilibili, Douyin, Instagram, etc., and choose social media and video platforms that meet the target audience; use web crawler tools (such as Scrapy, Beautiful Soup, etc.) to automatically collect video and live content. Crawlers can extract metadata such as video URLs, titles, descriptions, and release dates; if the platform provides an API, you can use the API interface to obtain blogger content data, including direct links to videos and audios, comments, likes, and other information; if you cannot use crawlers or APIs, manually download videos and live content to ensure the integrity of the content.

[0060] S120: preprocessing the new media content to be evaluated to obtain a preprocessing result.

[0061] In this embodiment, the preprocessing result refers to the result obtained after preliminary processing of the collected new media content to be evaluated.

[0062] Specifically, video frames are extracted and audio signals are cleaned for the new media content to be evaluated to obtain a preprocessing result.

[0063] In addition, ensure that the collected data file formats are compatible, such as MP4 (video), AAC / WAV (audio), etc. When saving, consider the requirements of different quality and resolution. Record the metadata of each collected video, including blogger name, video title, upload date, number of views, interactive data (comments, likes, etc.), and related tags and classification information. Use a database or file system (such as HDFS, MySQL, MongoDB, etc.) to store the collected data for subsequent processing and analysis.

[0064] In one embodiment, the collected data is deduplicated to ensure that the same content is not collected multiple times. Video and audio files of different formats are converted to a unified format for subsequent processing (such as transcoding using FFmpeg). For live content, it may be necessary to edit and extract specific clips to retain valuable information. Ensure that the collected video and audio files are intact and can be played normally. Check the video quality and audio clarity to ensure the accuracy of subsequent analysis and feature extraction. Ensure compliance with the usage agreements and copyright laws of the relevant platforms to avoid unauthorized content collection. When collecting and using data, ensure compliance with laws and regulations on data privacy protection to avoid infringement of the personal information of bloggers and users.

[0065] S130: Perform feature extraction on the preprocessing result to obtain an extraction result.

[0066] In this embodiment, the extraction results refer to the video clip features, the audio features, the visual aesthetic features, the content text features, and the emotion features.

[0067] Specifically, video editing features refer to information about shot switching, editing style, and duration in a video. These features help analyze the rhythm, structure, and fluency of a video. For example, fast shot switching may indicate a dynamic scene, and slow editing may be used to render emotions.

[0068] Audio features: These include various properties of the audio signal in the video, such as frequency, volume, pitch, timbre, and rhythm. These features can be used to analyze the impact of background music, dialogue quality, or sound effects, and thus affect the audience's emotional response.

[0069] Visual aesthetic features: These relate to the visual elements of a video, such as color, composition, lighting, and special effects. These features help determine the aesthetic quality and visual appeal of a video. For example, bright colors and good composition often enhance the viewer's viewing experience.

[0070] Content text features: refers to the text information in the video, such as subtitles, text in images, or any visual data. These features help extract the theme, keywords, and plot development of the video, making it easier to index and search the content.

[0071] Emotional features: involve emotional information extracted from the video, including the actor's expression, tone of voice, and scene atmosphere. These features are used to analyze the audience's possible emotional reactions, such as happiness, sadness, tension, etc., which in turn affects the dissemination effect and acceptance of the video.

[0072] The extraction of these features is usually achieved through technologies such as machine learning, computer vision, and natural language processing, and can be applied to multiple fields such as video recommendation, sentiment analysis, and content understanding.

[0073] In addition, computer vision techniques are used for face detection and expression recognition to extract biometric features related to facial expressions and movements from video frames.

[0074] Acoustic features such as speaking rate, intonation, and volume are extracted from audio signals through audio processing techniques such as short-time Fourier transform (STFT).

[0075] Use speech recognition technology to convert audio content into text, and then use natural language processing technology to extract features of language expression such as emotions, sentiments and themes.

[0076] In one embodiment, see Figure 2 The above-mentioned step S130 may include steps S131~S136.

[0077] S131, extracting basic attributes of the video in the preprocessing result and expressing them as a multi-dimensional feature vector to obtain video editing features.

[0078] In this embodiment, the video clip features include the following aspects:

[0079] Video Length: The length of the entire video.

[0080] Average number of scenes and length: The average number of scenes in the video and the duration of each scene.

[0081] Speech rate: The speaking speed of the voice information in the video.

[0082] Portrait ratio: the proportion of portraits in the video.

[0083] Slice position and proportion: The position of different slices in the video clip and their proportion in the overall video.

[0084] The above features are integrated into a 7-dimensional feature vector. Through this representation, the basic structure and content dynamics of the video can be effectively captured.

[0085] S132, analyzing the audio content of the video in the preprocessing result, and extracting Mel spectrum coefficients, Mel spectrum, chrominance frequency, spectral centroid, spectral contrast and tonality features, combining them into a feature matrix, and calculating a cross-matching matrix between background music and vocal features to obtain audio features.

[0086] In this embodiment, the audio features mainly include multiple audio features of background music and human voice:

[0087] Mel-spectrogram coefficients (MFCCs): used to describe the characteristics of audio signals, especially for human voice.

[0088] Mel Spectrogram: It represents the spectral characteristics of audio and can effectively capture the pitch information in the audio.

[0089] Chroma STFT: Used to describe the harmonic characteristics in audio.

[0090] Spectral Centroid: Indicates the center-of-gravity frequency of the audio signal.

[0091] Spectral Contrast: A measure of the contrast between the frequency spectra of an audio signal.

[0092] Tonnetz: describes the harmonic relationships in music.

[0093] The above audio features are combined into a 5*7 feature matrix, and the cross-matching matrix of background music and vocal features is calculated. This can achieve deep learning of audio features and better understand the sound content of the video.

[0094] S133, extracting dense optical flow and foreground motion area to represent dynamic features of the video in the preprocessing result, and extracting color features of the video in the preprocessing result through saturation, brightness, contrast and clarity of the video to obtain visual aesthetic features.

[0095] In this embodiment, the visual aesthetic features include dynamic features and color features in the video:

[0096] Dynamic features: capture motion information in the video by extracting dense optical flow and foreground motion area.

[0097] Color features: extract the saturation, brightness, contrast, and clarity of the video to reflect the visual quality of the video.

[0098] The deep learning model is used to obtain the representation vectors of dynamic and color features respectively, ensuring a comprehensive understanding of the visual performance of the video.

[0099] S134. Extracting text information of the video in the preprocessing result through a pre-trained language model, and converting the human voice information in the video into text to obtain content text features.

[0100] In this embodiment, there are two aspects of extracting content text features:

[0101] Image text information extraction: Use pre-trained language models to extract text information from images in videos.

[0102] Voice to text: Convert the voice information in the video into text.

[0103] These two parts of text information complement each other and together constitute the text content feature representation of the video.

[0104] S135, extracting and learning emotion vectors from the audio features and the facial features in the video in the preprocessing results, and obtaining emotion features through supplementation and comparison integration.

[0105] In this embodiment, the extraction of emotional features is also multifaceted:

[0106] Extracting emotion vectors from vocal features.

[0107] Extracting emotional information from facial features.

[0108] The two parts of the emotion vectors are supplemented, contrasted and integrated to form an overall emotion feature representation vector.

[0109] S136, combining the video clip feature, the audio feature, the visual aesthetic feature, the content text feature and the emotion feature to obtain an extraction result.

[0110] In this embodiment, by comprehensively considering multiple dimensions such as video editing, audio, vision and text, the video content and quality can be understood more comprehensively. Using deep learning technology for feature representation learning can capture complex patterns and relationships, and improve the accuracy and effectiveness of feature analysis. The extraction of emotional features helps to deeply understand the audience's emotional response to the video, which is convenient for subsequent emotional analysis and emotion-driven recommendations. Through cross-matching between audio and video, the correlation between audio and visual content can be revealed, and the grasp of the overall information of the video can be enhanced. Through comprehensive analysis and understanding of the video, personalized recommendations can be provided to the audience to enhance the viewing experience. The extracted multiple features can be applied to a variety of tasks such as content classification, sentiment analysis, and automatic summary generation, and have broad application potential.

[0111] Through the above methods, video data can be deeply analyzed to provide a rich information basis for subsequent causal relationship model input, thereby improving the intelligence level of analysis and decision-making.

[0112] S140, inputting the extraction result into the counterfactual causal inference model to predict the content design strategy to obtain the prediction result.

[0113] In this embodiment, the prediction result refers to a content design strategy that can generate better feedback.

[0114] In this embodiment, the counterfactual causal inference model is obtained by defining intervention measures and results, combining new media content and consumer feedback data, using a causal representation learning algorithm, constructing feature representation and optimizing errors, and is used to predict changes in consumer feedback under different content design strategies.

[0115] In one embodiment, see Figure 3 The training process of the above-mentioned counterfactual causal inference model includes steps S141 to S145.

[0116] S141. Define interventions and expected outcomes.

[0117] In this example, interventions refer to different content design strategies, such as the editing style, audio effects, and visual elements of the video.

[0118] The results are various types of consumer feedback, such as viewing completion rate, comment content, number of likes, etc.

[0119] S142: Acquire new media historical content and corresponding consumer feedback data to obtain initial data.

[0120] In this embodiment, the sources of initial data include:

[0121] New media platforms: Select multiple new media platforms (such as social media, video sharing websites, blogs, etc.) as data sources to ensure the diversity and representativeness of the data.

[0122] Content Type: Gather different types of content, such as videos, articles, images, live broadcasts, etc. to ensure that a wide range of content design strategies are covered.

[0123] Feedback metrics: Identify various forms of consumer feedback, such as views, likes, comments, shares, completion rates, etc., to obtain comprehensive feedback data.

[0124] After determining the source, data can be extracted in the following ways:

[0125] Crawling technology: Use web crawler tools to extract historical content and relevant consumer feedback data from the selected platform. Make sure to follow the platform's usage policy and data acquisition specifications.

[0126] API call: If the platform provides an open API interface, content and feedback data can be directly obtained through the API to ensure the efficiency and accuracy of data extraction.

[0127] Preprocess the extracted data:

[0128] Clean the extracted data to remove duplicate, invalid or erroneous data to ensure data quality; convert data from different sources into a unified format for subsequent analysis and processing; annotate and classify the content, such as by content type, theme, sentiment, etc., for subsequent analysis.

[0129] The cleaned historical content and its corresponding consumer feedback data are integrated into the initial data set as the basis for model training and analysis.

[0130] S143, updating the initial data by using error minimization and regularization methods and adding penalty items.

[0131] In this embodiment, the first goal of causal representation learning can be achieved by conventional methods of error minimization and regularization on the training set to achieve a good generalization error. In order to achieve the second goal of causal representation learning, a penalty term is designed to make the counterfactual prediction close to the closest observation in the corresponding control group. This method ensures that the model can effectively make predictions even when some observations are missing.

[0132] Specifically, the specific implementation steps of error minimization include:

[0133] Target definition: Set the prediction target of the model, such as predicting specific indicators of consumer feedback (such as number of likes, viewing time, etc.).

[0134] Loss function: Choose an appropriate loss function (such as mean square error, cross entropy, etc.) to measure the difference between the model prediction value and the actual feedback value.

[0135] Optimization algorithm: Use optimization algorithms (such as gradient descent, Adam, etc.) to adjust model parameters to minimize the loss function and achieve better prediction results.

[0136] The specific implementation steps of regularization include:

[0137] Prevent overfitting: In order to solve the problem that the model performs well on the training set but poorly on the validation set, regularization techniques (such as L1 and L2 regularization) are applied to constrain the model complexity.

[0138] Regularization term: Add a regularization term to the loss function to penalize excessive weights, prompting the model to learn more concise representations, thereby improving generalization ability.

[0139] The specific steps to add penalty items include:

[0140] Define penalty terms: Design appropriate penalty terms based on the specific needs of the model (such as sparsity requirements and similarity requirements for certain features).

[0141] Combined with loss function: Integrate the penalty term into the loss function so that the model not only focuses on minimizing the error during optimization, but also takes into account the complexity of the model and other constraints. The penalty term is in the form of , the goal is to make the counterfactual predictions close to the closest observations in the corresponding control group. Representation learning vectors for video clip features, audio features, visual aesthetic features, content text features, and emotional features after video parsing. are different intervention strategies. It is an indicator of the completion rate within the first 5 seconds observed under different intervention strategies. is a hyperparameter to control the strength of the penalty term. n is the number of samples. is the representation learning result to be trained. h is the mapping function, whose purpose is to learn the representation , to ensure that the counterfactual representation learned based on actual observations has low prediction error.

[0142] Weight adjustment: According to the characteristics and importance of the feedback data, the weight of the penalty term is adjusted to ensure that the model can balance the relationship between accuracy and complexity when learning.

[0143] By obtaining historical content from new media and its consumer feedback data, and updating the initial data with error minimization and regularization methods, the model can be trained in a more realistic environment, improving its prediction performance and generalization ability. This process not only ensures the quality and representativeness of the data, but also makes the prediction results more reliable and practical through appropriate adjustments to the model.

[0144] S144. Learning the causal relationship representation of the initial data based on a causal representation learning algorithm to obtain a learning result.

[0145] In this example, the goal of causal representation learning is:

[0146] Goal 1: Ensure that the observed consumer feedback has low error in the representation of facts. This means that we want the model to accurately reflect the actual situation.

[0147] Goal 2: Introduce a relevant factual outcome so that unobserved counterfactual predictions have low error. The model should be able to make reasonable inferences even in unobserved situations.

[0148] Goal 3: Ensure that the distributions of intervention groups are similar. By reducing the differences in the distribution of feedback data under different strategies, the model can better compare and predict.

[0149] For goal one, the model is designed to accurately capture and reflect real consumer feedback. This requires the use of high-quality data sets that include a variety of consumer behaviors and their diverse responses; through appropriate data preprocessing and feature selection, ensure that the data reflects consumer intentions and preferences as realistically as possible. Improve the credibility and effectiveness of the model, so that decision makers can make more accurate judgments based on the model output. Reduce business decision errors caused by inaccurate models and enhance responsiveness to the market.

[0150] For the second objective, in addition to the observed results, unobserved counterfactuals should also be considered during the modeling process. This means that the model needs to be able to predict possible results under different circumstances. By setting reasonable assumptions and using relevant variables, the model can make reasonable inferences about unobserved situations. This enables the model to have stronger generalization capabilities and be able to handle the impact of more unknown variables in practical applications. This enhances the foresight of decision-making and enables companies to prepare in advance in a changing market environment.

[0151] For the third objective, in order to improve the comparability of the model under different strategies, the distribution of the intervention group and the control group needs to be matched or adjusted to reduce potential bias. Through data weighting and resampling techniques, the characteristic distribution between different groups is made as similar as possible.

[0152] By improving the similarity between groups, the reliability of causal inference is enhanced, enabling the model to more accurately identify causal relationships. This improves the robustness of the model and reduces the prediction error of the model under different intervention conditions, thereby improving overall performance.

[0153] S145. Construct a neural network model.

[0154] In this embodiment, the neural network model adopts a fully connected ReLU layer structure and is optimized using the RMSProp algorithm to obtain a counterfactual causal inference model to improve the convergence speed and prediction accuracy of the model.

[0155] Specifically, RMSProp is an adaptive learning rate optimization algorithm that can dynamically adjust the learning rate according to the change of gradient. The ReLU activation function can effectively avoid the gradient vanishing problem and improve the learning efficiency and accuracy of the model. The adaptive learning rate of the RMSProp algorithm can speed up the convergence speed and reduce the dependence on manual adjustment of hyperparameters.

[0156] S146. Using the learning results to train the neural network model to obtain a counterfactual causal inference model.

[0157] In this embodiment, the loss function used in the neural network model training process is optimized using the alternating sub-gradient descent method, and the model parameters are adjusted during the training process to minimize the overall goal.

[0158] In addition, the neural network model performs sparse reweighting on the input features during training. Specifically, a sparse reweighting design is introduced to weight the input features to reduce the impact of inconsistent distribution features on result prediction. The similarity between the intervention group and the control group is maximized to achieve goal three.

[0159] Specifically, a sparse reweighting design (i.e., a diagonal matrix W) is used to reduce the impact of distribution-inconsistent features on the prediction results.

[0160] The data set of the training process includes the content information of the new media and the corresponding consumer feedback data, and the diversity and representativeness of the content and feedback labels need to be ensured. The training set is input into the model after feature extraction in step S130.

[0161] Specifically, the alternating sub-gradient descent method can effectively handle the optimization problem of high-dimensional data and accelerate the model training process. The sparse reweighted design helps the model focus on important features, reduce the impact of noise, and improve the accuracy of prediction, thereby achieving the maximum similarity between the intervention group and the control group mentioned in the third goal.

[0162] In this embodiment, the training process of the entire counterfactual causal inference model uses discH to represent the difference distance, that is, the loss function.

[0163] Specifically, the counterfactual causal inference model can be expressed as: , the overall goal is to minimize the expression in the formula and function , is a counterfactual causal inference model; where the input of the model is Representation learning vectors for video clip features, audio features, visual aesthetic features, content text features, and emotional features after video parsing. It is the indicator of the completion rate in the first 5 seconds of use; are different intervention strategies. It is an indicator of the completion rate within the first 5 seconds observed under different intervention strategies. is a hyperparameter to control the strength of the penalty term. n is the number of samples. is the representation learning result to be trained. h is the mapping function, whose purpose is to learn the representation , to ensure that the counterfactual representation learned based on actual observations has low prediction error. The difference distance is defined to reduce the impact of inconsistent distribution features on prediction errors. and They represent the empirical factual distribution and empirical counterfactual distribution of sample characteristics respectively.

[0164] Counterfactuals are estimated by nearest neighbor matching and a sparse reweighting design (i.e., a diagonal matrix W) is introduced to reduce the influence of distribution-inconsistent features on the predicted outcomes. To minimize the variance, features k with large variance between intervention groups will be given smaller weights wk. Can write ; where p represents the expectation of intervention , and represents the mean of the characteristics under intervention and without intervention, and represents the mean of the characteristics under intervention and without intervention.

[0165] Minimizing the overall goal involves a trade-off between maximizing balance (solving the problem of inconsistent distribution) and prediction accuracy, and the alternating sub-gradient descent method is used to achieve minimization. During the training process, the neural network architecture in the model consists of a fully connected ReLU layer and is trained using RMSProp. Based on the learning, causal representations of video clip features, audio features, visual aesthetic features, content text features, and emotional features can be obtained.

[0166] The counterfactual causal inference model is used to predict how consumer feedback will change if different content design strategies are adopted, thereby helping bloggers optimize content design.

[0167] S150, generating a value assessment report according to the prediction results;

[0168] In this embodiment, the value assessment report includes value scoring, influencing factor analysis, etc.

[0169] Based on the value assessment report, implement targeted blogger training and marketing strategies.

[0170] Specifically, the prediction results include the prediction score, as well as the influencing factors and feature scores based on causal analysis under each prediction score. Based on the prediction results, for example, for a video with a predicted playback volume of 150,000 times, the key features and impact of the video can be output at the same time, providing qualitative and quantitative support for video content optimization.

[0171] By generating a value assessment report, companies can not only gain a deeper understanding of the market potential of bloggers, but also develop more effective marketing strategies. The core of this process is data-driven analysis, combined with market trends and consumer demand, to ensure that companies maintain their advantages in a highly competitive market.

[0172] S160: Output the value assessment report.

[0173] The method of this embodiment reduces data noise by deeply analyzing the video content of knowledge-sharing bloggers (including facial expressions, voice features, and emotional expressions), provides more accurate value assessment, extracts personalized features through video and audio analysis technology, and establishes a causal relationship model between these features and consumer feedback behavior to achieve more refined value assessment. In order to deeply understand how the personal characteristics of knowledge-sharing bloggers affect consumer feedback, the method of this embodiment constructs a causal inference model based on counterfactual inference to reveal the intrinsic connection between the personal characteristics of knowledge-sharing bloggers and consumer feedback.

[0174] The above-mentioned new media content value assessment method collects all relevant new media content of the blogger to be evaluated, including articles, videos and interaction records, and organizes them to ensure the integrity and consistency of the data. The collected new media content is preprocessed to clean irrelevant information, remove noise, and convert the data into a format suitable for analysis to ensure data quality. Text analysis and sentiment analysis techniques are used to extract key features from the preprocessed data, such as content topics, audience engagement, and sentiment tendencies, to obtain more representative feature data. The extracted features are input into the counterfactual causal inference model, and the model is used to analyze the potential impact of different content strategies on blogger performance, so as to obtain more accurate prediction results. A value assessment report is written based on the prediction results, providing a comprehensive analysis of the blogger's performance and corresponding improvement suggestions to ensure the accuracy and operability of the output results. The accuracy and reliability of the value assessment of knowledge-sharing bloggers are improved.

[0175] Figure 4 is a schematic block diagram of a new media content value assessment device 300 provided in an embodiment of the present invention. Figure 4 As shown, corresponding to the above new media content value assessment method, the present invention also provides a new media content value assessment device 300. The new media content value assessment device 300 includes a unit for executing the above new media content value assessment method, and the device can be configured in a server. Figure 4 The new media content value assessment device 300 includes a content acquisition unit 301, a preprocessing unit 302, a feature extraction unit 303, a prediction unit 304, an assessment report unit 305 and an output unit 306.

[0176] The content acquisition unit 301 is used to acquire new media content to be evaluated; the preprocessing unit 302 is used to preprocess the new media content to be evaluated to obtain a preprocessing result; the feature extraction unit 303 is used to extract features from the preprocessing result to obtain an extraction result; the prediction unit 304 is used to input the extraction result into the counterfactual causal inference model to predict the content design strategy to obtain a prediction result; the evaluation report unit 305 is used to generate a value evaluation report according to the prediction result; the output unit 306 is used to output the value evaluation report.

[0177] In one embodiment, the preprocessing unit 302 is used to extract video frames and clean audio signals of the new media content to be evaluated to obtain a preprocessing result.

[0178] In one embodiment, if Figure 5 As shown, the feature extraction unit 303 includes a video clip feature determination subunit 3031, an audio feature determination subunit 3032, a visual aesthetic feature determination subunit 3033, a content text feature determination subunit 3034, an emotion feature determination subunit 3035 and a combination subunit 3036.

[0179] The video editing feature determination subunit 3031 is used to extract the basic attributes of the video in the preprocessing result and express them as a multi-dimensional feature vector to obtain the video editing features; the audio feature determination subunit 3032 is used to analyze the audio content of the video in the preprocessing result, and extract the Mel spectrum coefficient, Mel spectrum, chromaticity frequency, spectral centroid, spectral contrast and tonality features, combine them into a feature matrix, and calculate the cross-matching matrix between the background music and the vocal features to obtain the audio features; the visual aesthetic feature determination subunit 3033 is used to extract the dense optical flow and the foreground motion area to represent the dynamic features of the video in the preprocessing result, and to obtain the saturation, brightness, contrast and clarity of the video. The color features of the video in the preprocessing result are extracted by using the clarity to obtain the visual aesthetic features; the content text feature determination subunit 3034 is used to extract the text information of the video in the preprocessing result through the pretrained language model, and convert the human voice information in the video into text to obtain the content text features; the emotion feature determination subunit 3035 is used to extract and learn the emotion vector from the audio features and the facial features in the video in the preprocessing result, and obtain the emotion features through supplementation and comparison integration; the combination subunit 3036 is used to combine the video editing features, the audio features, the visual aesthetic features, the content text features and the emotion features to obtain the extraction result.

[0180] In one embodiment, the device further includes a model training unit, and the model training unit further includes:

[0181] The definition subunit is used to define intervention measures and expected results; the data acquisition subunit is used to obtain the historical content of new media and the corresponding consumer feedback data to obtain initial data; the update subunit is used to update the initial data by minimizing the error and regularizing the method and adding penalty terms. The learning representation subunit is used to learn the causal relationship representation of the initial data based on the causal representation learning algorithm to obtain the learning result; the model construction subunit is used to construct a neural network model; the training subunit is used to train the neural network model using the learning result to obtain a counterfactual causal inference model.

[0182] It should be noted that technicians in the relevant field can clearly understand that the specific implementation process of the above-mentioned new media content value assessment device 300 and each unit can refer to the corresponding description in the aforementioned method embodiment, and for the convenience and brevity of description, it will not be repeated here.

[0183] The new media content value evaluation device 300 can be implemented in the form of a computer program. Figure 6 Runs on the computer device shown.

[0184] See also Figure 6 , Figure 6 5 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 500 may be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0185] See also Figure 6 The computer device 500 includes a processor 502 , a memory and a network interface 505 connected via a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .

[0186] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions, and when the program instructions are executed, the processor 502 can execute a new media content value assessment method.

[0187] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500 .

[0188] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a new media content value assessment method.

[0189] The network interface 505 is used to communicate with other devices over the network. Figure 6 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0190] The processor 502 is used to run the computer program 5032 stored in the memory to implement the following steps:

[0191] Acquire new media content to be evaluated; preprocess the new media content to be evaluated to obtain a preprocessing result; perform feature extraction on the preprocessing result to obtain an extraction result; input the extraction result into a counterfactual causal inference model to predict content design strategy to obtain a prediction result; generate a value assessment report based on the prediction result; and output the value assessment report.

[0192] Among them, the counterfactual causal inference model is obtained by defining intervention measures and results, combining new media content and consumer feedback data, using a causal representation learning algorithm, constructing feature representation and optimizing errors, and is used to predict changes in consumer feedback under different content design strategies.

[0193] In one embodiment, when the processor 502 implements the step of preprocessing the new media content to be evaluated to obtain a preprocessing result, the processor 502 specifically implements the following steps:

[0194] Video frame extraction and audio signal cleaning are performed on the new media content to be evaluated to obtain a preprocessing result.

[0195] In one embodiment, when the processor 502 implements the step of performing feature extraction on the preprocessing result to obtain the extraction result, the processor 502 specifically implements the following steps:

[0196] Extract the basic attributes of the video in the preprocessing result and express them as a multidimensional feature vector to obtain video editing features; analyze the audio content of the video in the preprocessing result, and extract Mel spectrum coefficients, Mel spectrum, chromaticity frequency, spectral centroid, spectral contrast and tonality features, combine them into a feature matrix, and calculate the cross-matching matrix between background music and vocal features to obtain audio features; extract dense optical flow and foreground motion area to represent the dynamic features of the video in the preprocessing result, and extract the color features of the video in the preprocessing result by the saturation, brightness, contrast and clarity of the video to obtain visual aesthetic features; extract the text information of the video in the preprocessing result through a pre-trained language model, and convert the vocal information in the video into text to obtain content text features; extract and learn emotion vectors from audio features and facial features in the video in the preprocessing result, and obtain emotional features through supplementation and comparison integration; combine the video editing features, the audio features, the visual aesthetic features, the content text features and the emotional features to obtain extraction results.

[0197] In one embodiment, when implementing the training step of the counterfactual causal inference model, the processor 502 specifically implements the following steps:

[0198] Define intervention measures and expected results; obtain new media historical content and corresponding consumer feedback data to obtain initial data; learn the causal relationship representation of the initial data based on a causal representation learning algorithm to obtain a learning result; build a neural network model; use the learning result to train the neural network model to obtain a counterfactual causal inference model.

[0199] The neural network model adopts a fully connected ReLU layer structure and is optimized using the RMSProp algorithm to obtain a counterfactual causal inference model.

[0200] The loss function used in the neural network model training process is optimized using the alternating sub-gradient descent method.

[0201] The neural network model performs sparse reweighting processing on the input features during training.

[0202] In one embodiment, after implementing the step of acquiring new media historical content and corresponding consumer feedback data to obtain initial data, the processor 502 further implements the following steps:

[0203] The initial data is updated by using error minimization and regularization methods as well as adding penalty items.

[0204] It should be understood that in the embodiment of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0205] It can be understood by those skilled in the art that all or part of the processes in the method for implementing the above embodiment can be completed by instructing the relevant hardware through a computer program. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiment of the above method.

[0206] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor executes the following steps:

[0207] Acquire new media content to be evaluated; preprocess the new media content to be evaluated to obtain a preprocessing result; perform feature extraction on the preprocessing result to obtain an extraction result; input the extraction result into a counterfactual causal inference model to predict content design strategy to obtain a prediction result; generate a value assessment report based on the prediction result; and output the value assessment report.

[0208] Among them, the counterfactual causal inference model is obtained by defining intervention measures and results, combining new media content and consumer feedback data, using a causal representation learning algorithm, constructing feature representation and optimizing errors, and is used to predict changes in consumer feedback under different content design strategies.

[0209] In one embodiment, when the processor executes the computer program to implement the step of preprocessing the new media content to be evaluated to obtain a preprocessing result, the processor specifically implements the following steps:

[0210] Video frame extraction and audio signal cleaning are performed on the new media content to be evaluated to obtain a preprocessing result.

[0211] In one embodiment, when the processor executes the computer program to implement the step of extracting features from the preprocessing results to obtain extraction results, the processor specifically implements the following steps:

[0212] Extract the basic attributes of the video in the preprocessing result and express them as a multidimensional feature vector to obtain video editing features; analyze the audio content of the video in the preprocessing result, and extract Mel spectrum coefficients, Mel spectrum, chromaticity frequency, spectral centroid, spectral contrast and tonality features, combine them into a feature matrix, and calculate the cross-matching matrix between background music and vocal features to obtain audio features; extract dense optical flow and foreground motion area to represent the dynamic features of the video in the preprocessing result, and extract the color features of the video in the preprocessing result by the saturation, brightness, contrast and clarity of the video to obtain visual aesthetic features; extract the text information of the video in the preprocessing result through a pre-trained language model, and convert the vocal information in the video into text to obtain content text features; extract and learn emotion vectors from audio features and facial features in the video in the preprocessing result, and obtain emotional features through supplementation and comparison integration; combine the video editing features, the audio features, the visual aesthetic features, the content text features and the emotional features to obtain extraction results.

[0213] In one embodiment, when the processor executes the computer program to implement the training step of the counterfactual causal inference model, the processor specifically implements the following steps:

[0214] Define intervention measures and expected results; obtain new media historical content and corresponding consumer feedback data to obtain initial data; learn the causal relationship representation of the initial data based on a causal representation learning algorithm to obtain a learning result; build a neural network model; use the learning result to train the neural network model to obtain a counterfactual causal inference model.

[0215] The neural network model adopts a fully connected ReLU layer structure and is optimized using the RMSProp algorithm to obtain a counterfactual causal inference model.

[0216] The loss function used in the neural network model training process is optimized using the alternating sub-gradient descent method.

[0217] The neural network model performs sparse reweighting processing on the input features during training.

[0218] In one embodiment, after executing the computer program to implement the step of acquiring new media historical content and corresponding consumer feedback data to obtain initial data, the processor further implements the following steps:

[0219] The initial data is updated by using error minimization and regularization methods as well as adding penalty items.

[0220] The storage medium may be a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk, etc., which are computer-readable storage media that can store program codes.

[0221] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0222] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of each unit is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0223] The steps in the method of the embodiment of the present invention can be adjusted in order, combined and deleted according to actual needs. The units in the device of the embodiment of the present invention can be combined, divided and deleted according to actual needs. In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0224] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, terminal, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0225] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A new media content value assessment method, characterized in that: include: Obtain new media content to be evaluated; Preprocessing the new media content to be evaluated to obtain a preprocessing result; Performing feature extraction on the preprocessing result to obtain an extraction result; Inputting the extracted results into the counterfactual causal inference model to predict the content design strategy, so as to obtain the prediction results; Generate a value assessment report based on the prediction results; Outputting the value assessment report; The counterfactual causal inference model is obtained by defining intervention measures and results, combining new media content and consumer feedback data, using a causal representation learning algorithm, constructing feature representation and optimizing errors, and is used to predict changes in consumer feedback under different content design strategies; The training process of the counterfactual causal inference model includes: define interventions and expected outcomes; Obtain new media historical content and corresponding consumer feedback data to obtain initial observable data; Based on the causal representation learning algorithm, combined with the initial observable data, the causal relationship representation learning of sample features is carried out to solve the representation learning bias caused by the inconsistent distribution of factual and counterfactual samples, so as to obtain the learning results; Build a neural network model; Using the learning results to train the neural network model to obtain a counterfactual causal inference model; After obtaining the historical content of new media and the corresponding consumer feedback data, the overall causal representation and inference model also includes: Degrading the influence of non-causal features through sparse reweighting; Balancing factual and counterfactual outcomes through differential distance design; The initial observable data is updated by using error minimization and regularization methods and adding penalty items; The neural network model adopts a fully connected ReLU layer structure and is optimized using the RMSProp algorithm to obtain a counterfactual causal inference model; The loss function used in the neural network model training process is optimized using the alternating sub-gradient descent method.

2. The new media content value assessment method according to claim 1, characterized in that: The preprocessing of the new media content to be evaluated to obtain a preprocessing result includes: Video frame extraction and audio signal cleaning are performed on the new media content to be evaluated to obtain a preprocessing result.

3. The new media content value assessment method according to claim 1, characterized in that: The performing feature extraction on the preprocessing result to obtain an extraction result includes: Extracting basic attributes of the video in the preprocessing result and expressing them as a multi-dimensional feature vector to obtain video editing features; Analyze the audio content of the video in the preprocessing result, extract Mel spectrum coefficients, Mel spectrum, chrominance frequency, spectral centroid, spectral contrast and tonality features, combine them into a feature matrix, calculate the cross-matching matrix between background music and vocal features, so as to obtain audio features; Extracting dense optical flow and foreground motion area to represent dynamic features of the video in the preprocessing result, and extracting color features of the video in the preprocessing result by saturation, brightness, contrast and clarity of the video to obtain visual aesthetic features; Extracting text information of the video in the preprocessing result through a pre-trained language model, and converting human voice information in the video into text to obtain content text features; Extracting and learning emotion vectors from the audio features and the facial features in the video in the preprocessing results, and obtaining emotion features through supplementation and comparison integration; The video clip features, the audio features, the visual aesthetic features, the content text features, and the emotion features are combined to obtain an extraction result.

4. A new media content value assessment device, characterized in that: include: A content acquisition unit, used to acquire new media content to be evaluated; A preprocessing unit, used for preprocessing the new media content to be evaluated to obtain a preprocessing result; A feature extraction unit, used for performing feature extraction on the preprocessing result to obtain an extraction result; A prediction unit, used for inputting the extraction result into a counterfactual causal inference model to predict the content design strategy, so as to obtain a prediction result; An evaluation report unit, used to generate a value evaluation report according to the prediction result; An output unit, used for outputting the value assessment report; The counterfactual causal inference model is obtained by defining intervention measures and results, combining new media content and consumer feedback data, using a causal representation learning algorithm, constructing feature representation and optimizing errors, and is used to predict changes in consumer feedback under different content design strategies; The device also includes a model training unit, and the model training unit also includes: A definition subunit is used to define intervention measures and expected results; a data acquisition subunit is used to obtain new media historical content and corresponding consumer feedback data to obtain initial observable data; an update subunit is used to update the initial observable data by using error minimization and regularization methods and adding penalty terms; a learning representation subunit uses a causal representation learning algorithm to learn sample feature causal relationship representations in combination with initial observable data to solve representation learning biases caused by inconsistent distributions of factual and counterfactual samples to obtain learning results; a model construction subunit is used to construct a neural network model; a training subunit is used to train the neural network model using the learning results to obtain a counterfactual causal inference model; The model training unit is also used for: after obtaining the new media historical content and the corresponding consumer feedback data, the overall causal representation and inference model also includes: degrading the influence of non-causal features through sparse reweighting; balancing the factual results and counterfactual results through difference distance design; updating the initial observable data through error minimization and regularization methods and adding penalty terms; the neural network model adopts a fully connected ReLU layer structure and is optimized using the RMSProp algorithm to obtain a counterfactual causal inference model; the loss function used in the neural network model training process is optimized using the alternating sub-gradient descent method.

5. A computer device, characterized in that: The computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 3 when executing the computer program.

Images