Video processing method and device

Abstract

The present invention provides a video processing method and device; the method comprises: obtaining at least one interactive information corresponding to a video, the interactive information being used to characterize the user's emotional tendency toward the video; determining the emotional category of the at least one interactive information according to the mapping relationship between the interactive information and the emotional category; obtaining the interactive information of the at least one interactive information whose emotional category is a negative category; when the interactive information of the negative category meets a preset condition, determining the video as a target video carrying negative content. Through the present invention, videos carrying negative content can be automatically identified.

Description

Technical Field

[0001] This invention relates to Internet technology, and more particularly to a method and apparatus for processing video. Background Art

[0002] With the development of the internet, videos carrying all kinds of content are constantly flooding users' view. Among them, there are many videos carrying negative content, such as videos carrying incorrect news content, which cause users a very poor visual experience.

[0003] In related technologies, the identification of videos carrying negative content mainly involves playing the video and manually reviewing it, which wastes a lot of human and material resources. Summary of the Invention

[0004] This invention provides a video processing method that can automatically identify videos carrying negative content.

[0005] The technical solution of the embodiment of the present invention is achieved as follows:

[0006] This invention provides a video processing method, including:

[0007] Obtain at least one interactive information of the corresponding video, the interactive information being used to characterize the user's emotional inclination towards the video;

[0008] The emotional category of the at least one type of interactive information is determined based on the mapping relationship between interactive information and emotional categories.

[0009] Obtain interactive information in which the emotional category of the at least one type of interactive information is negative;

[0010] When the interactive information of the negative category meets the preset conditions, the video is determined to be a target video carrying negative content.

[0011] This invention provides a video processing apparatus, comprising:

[0012] An information acquisition unit is used to acquire at least one interactive information of a corresponding video, wherein the interactive information is used to characterize the user's emotional tendency toward the video;

[0013] A category determination unit is used to determine the emotional category of the at least one type of interactive information based on the mapping relationship between interactive information and emotional categories;

[0014] An information filtering unit is used to obtain interactive information in the at least one type of interactive information whose emotional category is negative.

[0015] The video determination unit is used to determine that the video is a target video carrying negative content when the interactive information of the negative category meets preset conditions.

[0016] In the above scheme, the category determination unit is further configured to obtain the word vector sequence corresponding to the text interaction information when the interaction information includes text interaction information;

[0017] Input the word vector sequence corresponding to the text interaction information into the sentiment classification model, and output the sentiment category corresponding to the text interaction information.

[0018] In the above scheme, the device further includes a model training unit for constructing a training sample set, the training sample set including interaction information samples and reference sentiment categories corresponding to the interaction information samples;

[0019] Input the word vector sequence corresponding to the interactive information sample into the sentiment classification model, and output the target sentiment category corresponding to the interactive information sample;

[0020] The model parameters of the emotion classification model are updated based on the difference between the target emotion category and the reference emotion category.

[0021] In the above scheme, the category determination unit is further used to perform word segmentation on the text interaction information to obtain the word set of the interaction information;

[0022] Obtain the word vector corresponding to each word in the word set;

[0023] The word vectors corresponding to each word in the word set are combined into a word vector sequence corresponding to the text interaction information.

[0024] In the above scheme, the information filtering unit is further configured to obtain text information associated with the video when the interactive information includes multiple text interactive information;

[0025] Determine the semantic similarity between each of the text interaction messages and the text information;

[0026] Based on the semantic similarity, the multiple text interaction information is filtered to obtain a target text interaction information set.

[0027] In the above scheme, the video determination unit is further used to obtain text interaction information and non-text interaction information in the interaction information of the negative category;

[0028] Determine a first quantity of the text interaction information and a second quantity of the non-text interaction information;

[0029] When the first quantity satisfies the first quantity condition and the second quantity satisfies the second quantity condition, the video is determined to be a target video carrying negative content.

[0030] In the above scheme, the video determining unit is further used to determine the number of interactive information whose emotion category is negative;

[0031] When the number of interactive messages in the negative category meets a preset condition, the video is determined to be a target video carrying negative content.

[0032] In the above scheme, the video determining unit is further used to obtain the total number of interactive information corresponding to the video;

[0033] When the ratio of the number of negative category interactive information to the total number of interactive information reaches a preset threshold, the video is determined to be a target video carrying negative content.

[0034] In the above solution, the device further includes a video processing unit for deleting the target video; or,

[0035] A prompt message for the target video is sent to the terminal, the prompt message being used to indicate that the target video carries negative content.

[0036] This invention provides a server, comprising:

[0037] Memory, used to store executable instructions;

[0038] The processor, when executing executable instructions stored in the memory, implements the video processing method provided in this embodiment of the invention.

[0039] This invention provides a storage medium storing executable instructions that, when executed by a processor, implement the video processing method provided in this invention.

[0040] The embodiments of the present invention have the following beneficial effects:

[0041] By mapping interactive information to emotion categories, the emotion category of at least one interactive information is determined, and interactive information with a negative emotion category is obtained. When the interactive information with a negative emotion category meets a preset condition, the video is determined to be a target video carrying negative content. In this way, the interactive information with a negative emotion category can be automatically determined according to the mapping relationship between interactive information and emotion categories, and then the video can be determined to be a target video carrying negative content based on the interactive information with a negative emotion category, thereby realizing the automatic identification of videos carrying negative content. Attached Figure Description

[0042] Figure 1 A schematic diagram of the network architecture of the video processing system provided in an embodiment of the present invention;

[0043] Figure 2A schematic diagram of the composition structure of a video processing apparatus provided in an embodiment of the present invention;

[0044] Figure 3 This is a flowchart illustrating the video processing method provided in an embodiment of the present invention.

[0045] Figure 4 This is a flowchart illustrating the video processing method provided in an embodiment of the present invention.

[0046] Figure 5 A flowchart illustrating the video processing method provided in an embodiment of the present invention;

[0047] Figure 6 This is a schematic diagram of the composition structure of a video processing device provided in an embodiment of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0050] If the application documents contain similar descriptions such as "first" or "second", the following explanation shall be added: In the following description, the terms "first" or "second" are used only to distinguish similar objects and do not represent a specific order of objects. It is understood that "first" and "second" may be interchanged in a specific order or sequence where permitted, so that the embodiments of the present invention described herein can be implemented in an order other than that illustrated or described herein.

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0052] Before further explaining the embodiments of the present invention in detail, the nouns and terms involved in the embodiments of the present invention are explained. The nouns and terms involved in the embodiments of the present invention are subject to the following interpretations.

[0053] 1) Word segmentation refers to the process of recombining a continuous sequence of characters into a sequence of words according to certain rules;

[0054] 2) Pipeline: The solution to a complex problem is broken down into processing units, which are then processed sequentially. The result of the previous processing unit is also the input of the next module.

[0055] 3) Negative content refers to erroneous news, content that does not match the title, etc. Such content is prone to complaints or reports, which can affect the product's reputation.

[0056] First, the video processing system of this embodiment will be described. Figure 1 This is a schematic diagram of the network architecture of the video processing system provided in an embodiment of the present invention. See also: Figure 1 To support an exemplary application, the video processing system 100 includes a terminal (including terminal 400-1 and terminal 400-2) and a server 200. The terminal connects to the server 200 through a network 300, which can be a wide area network, a local area network, or a combination of both, and uses a wireless link to achieve data transmission.

[0057] Server 200 is used to acquire at least one interactive information of the corresponding video, the interactive information being used to characterize the user's emotional tendency toward the video; determine the emotional category of at least one interactive information based on the mapping relationship between interactive information and emotional category; acquire interactive information of the at least one interactive information whose emotional category is negative; when the interactive information of the negative category meets preset conditions, determine the video as a target video carrying negative content; and send a prompt message for the target video to the terminal, the prompt message being used to prompt that the target video carries negative content.

[0058] Terminals (terminal 400-1 and / or terminal 400-2) are used to receive prompts for the target video.

[0059] The video processing apparatus provided in the embodiments of the present invention will now be described. The video processing apparatus of the embodiments of the present invention can be implemented in various forms, such as by a single terminal like a smartphone, tablet, or desktop computer, or by a terminal and a server working together. The video processing apparatus of the embodiments of the present invention can be implemented in hardware or a combination of hardware and software. Various exemplary implementations of the video processing apparatus provided in the embodiments of the present invention will be described below.

[0060] The hardware structure of the video processing device according to an embodiment of the present invention will be described in detail below. Figure 2 This is a schematic diagram of the composition structure of a video processing apparatus provided in an embodiment of the present invention. It can be understood that... Figure 2 The diagram shows only an exemplary structure of the video processing apparatus, not the entire structure; implementation is possible as needed. Figure 2 The structure shown may be part or all of the structure.

[0061] The video processing apparatus provided in this embodiment of the invention includes at least one processor 201, a memory 202, a user interface 203, and at least one network interface 204. The various components in the video processing apparatus 20 are coupled together via a bus system 205. It is understood that the bus system 205 is used to implement communication between these components. In addition to a data bus, the bus system 205 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 2 The general labeled all buses as Bus System 205.

[0062] The user interface 203 may include a monitor, keyboard, mouse, trackball, click wheel, buttons, touchpad, or touch screen.

[0063] It is understood that memory 202 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), flash memory, etc. Volatile memory can be random access memory (RAM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM) and synchronous static random access memory (SSRAM). The memory 202 described in this embodiment is intended to include these and any other suitable types of memory.

[0064] The memory 202 in this embodiment of the invention is capable of storing data to support the operation of the server 200. Examples of this data include any computer programs used to operate on the server 200, such as operating systems and applications. The operating system includes various system programs, such as framework layers, core library layers, driver layers, etc., used to implement various basic business functions and handle hardware-based tasks. Applications can include various other applications.

[0065] As an example of the video processing apparatus provided in this embodiment of the invention, which is implemented using a combination of hardware and software, the video processing apparatus provided in this embodiment of the invention can be directly embodied as a combination of software modules executed by processor 201. The software modules can be located in a storage medium, which is located in memory 202. Processor 201 reads the executable instructions included in the software modules in memory 202 and combines them with necessary hardware (e.g., including processor 201 and other components connected to bus 205) to complete the video processing method provided in this embodiment of the invention.

[0066] As an example, processor 201 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0067] As an example of the hardware implementation of the video processing apparatus provided in this embodiment of the invention, the video processing apparatus provided in this embodiment of the invention can be directly executed by a processor 201 in the form of a hardware decoding processor. For example, it can be executed by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components to implement the video processing method provided in this embodiment of the invention.

[0068] In this embodiment of the invention, the memory 202 is used to store various types of data to support the operation of the video processing device 20. Examples of such data include any executable instructions for operation on the video processing device 20, such as executable instructions, which may contain a program implementing the video processing method of this embodiment of the invention.

[0069] The video processing method provided in this invention will be described in conjunction with exemplary applications and implementations of the server provided in this embodiment. Figure 3 This is a flowchart illustrating the video processing method provided in an embodiment of the present invention. See also... Figure 3 The video processing method provided in this embodiment of the invention includes:

[0070] Step 301: The server obtains at least one interactive information for the corresponding video, which is used to characterize the user's emotional tendency toward the video.

[0071] In practical applications, there are various types of interactive information for a video, such as comments, reposts, likes, dislikes, and shares. The server can obtain one or more types of interactive information for a video, and there can be multiple instances of each type of interactive information. For example, the server can obtain all comments for a video, or all comments and reposts for a video.

[0072] Here, emotional tendency refers to a user's subjective preference or internal evaluation of a video, such as liking, supporting, disliking, or being angry about the video.

[0073] Step 302: Determine the emotion category of at least one type of interactive information based on the mapping relationship between interactive information and emotion categories.

[0074] It should be noted that emotion categories include positive and negative categories. Positive categories include expressions of support and liking for the video, while negative categories include expressions of dislike and anger. Here, there are multiple interaction messages; the emotion category of each interaction message is determined based on the mapping relationship between the interaction messages and their corresponding emotion categories.

[0075] In some embodiments, when the interactive information includes non-text interactive information, the server can determine the sentiment category of the non-text interactive information based on the mapping relationship between the attributes of the non-text interactive information and the sentiment category. For example, when the interactive information is a like, the corresponding sentiment category is positive; when the interactive information is a dislike, the corresponding sentiment category is negative.

[0076] In some embodiments, when the interactive information includes text interactive information, the server can determine the sentiment category of the interactive information by: obtaining the word vector sequence corresponding to the text interactive information; inputting the word vector sequence corresponding to the text interactive information into the trained sentiment classification model, and outputting the sentiment category of the corresponding text interactive information.

[0077] In practice, the text interaction information is represented by comments. The word vector sequence corresponding to the comments is input into the trained sentiment classification model to obtain the positive or negative probability of the comments. A probability threshold is set: when the obtained probability is greater than the threshold, the sentiment category of the comment is determined to be negative; when the obtained probability is less than the threshold, the sentiment category of the comment is determined to be positive. For example, with a probability threshold of 0.5, if the word vector sequence of a certain comment is input into the trained sentiment classification model and the obtained probability is 0.8, the sentiment category of the comment is negative; if the obtained probability is 0.2, the sentiment category of the comment is positive.

[0078] In some embodiments, the server obtains word vectors of text interaction information by: performing word segmentation on the text interaction information to obtain a word set of the text interaction information; obtaining the word vector corresponding to each word in the word set; and combining the word vectors corresponding to each word in the word set into a word vector sequence of the text interaction information.

[0079] Here, word segmentation refers to the process of recombining a continuous sequence of characters into a sequence of words according to certain rules. For example, if the interactive message is "I like literature", the word segmentation results in "I", "like", and "literature".

[0080] In some embodiments, the server can use a string matching-based word segmentation method to segment the interactive text information, that is, matching the character sequences in the interactive text information with words in a machine dictionary. In other embodiments, the server can use an understanding-based word segmentation method to segment the interactive text information, that is, performing syntactic and semantic analysis simultaneously with word segmentation, using syntactic and semantic information to handle ambiguity. In still other embodiments, the server can use a statistical word segmentation method to segment the interactive text information, that is, using a large amount of already segmented text, learning the rules of word segmentation through a statistical machine learning model, thereby achieving the segmentation of unknown text. In practice, existing word segmentation tools can be used to segment the interactive text information, such as jieba, SnowNLP, THULAC, etc.

[0081] In practice, the server can utilize the open-source Chinese word vector corpus from Beijing Normal University and Renmin University to obtain the word vector corresponding to each word in the word set. This corpus includes word vectors trained on dozens of corpora from various fields (Baidu Encyclopedia, Wikipedia, People's Daily 1947-2017, Zhihu, Weibo, literature, finance, classical Chinese, etc.), covering various fields and including multiple training settings.

[0082] In some embodiments, word segmentation and word vector conversion can be performed using a pipeline model. This allows the previous module to be replaced and a new module to be used to build the model when a new word segmentation tool or word vector emerges.

[0083] In some embodiments, the sentiment classification model is trained by: constructing a training sample set, which includes interaction information samples and reference sentiment categories corresponding to the interaction information samples; inputting the word vector sequences corresponding to the interaction information samples into the sentiment classification model to obtain the target sentiment category corresponding to the interaction information samples; and updating the model parameters of the sentiment classification model based on the difference between the target sentiment category and the reference sentiment category.

[0084] In some embodiments, the server may update the sentiment classification model in the following manner:

[0085] Based on the target sentiment category and the reference sentiment category, the server determines the value of the loss function and checks whether the value of the loss function exceeds a preset threshold. When the value of the loss function exceeds the preset threshold, the server determines the error signal of the sentiment classification model based on the loss function, backpropagates the error information in the sentiment classification model, and updates the model parameters of each layer during the propagation process.

[0086] Here's an explanation of backpropagation: Training sample data is input into the input layer of the neural network model, passes through the hidden layer, and finally reaches the output layer to output the result. This is the forward propagation process of the sentiment classification model. Since there is an error between the output result of the sentiment classification model and the actual result, the error between the output result and the actual value is calculated and propagated backward from the output layer to the hidden layer until it reaches the input layer. During the backpropagation process, the values ​​of the model parameters are adjusted according to the error. The above process is iterated until convergence.

[0087] In some implementations, the loss function of the sentiment classification model is:

[0088]

[0089] in, Let y1 be the target sentiment category, y1 be the reference sentiment category, and ω1(θ) be the output model parameters for the current iteration.

[0090] In practice, the server can analyze content with a high volume of negative feedback each day, identify frequently posted negative comments by users, label them as samples of negative sentiment, and build rules based on these. By using these rules to extract more similar comments as new samples of negative sentiment, a large amount of sample data with negative sentiment can be obtained quickly.

[0091] In some embodiments, the server randomly shuffles the obtained sample set and divides it into a training sample set, a validation sample set, and a test sample set according to a certain ratio. Then, it uses the training sample set and the validation sample set to adjust the parameters on a single model, compares the prediction accuracy of different parameters longitudinally, and selects the parameter combination with the highest accuracy.

[0092] In some embodiments, the server can simultaneously train multiple different models, obtain the parameter combination with the highest accuracy among all models, and then perform a horizontal comparison among the models with the highest accuracy parameter combination to determine the optimal model. The optimal model is then used to determine the sentiment category of the interactive information. For example, the server can simultaneously train a text-convolutional neural network (text-CNN), a gated recurrent unit (GRU) model, and a long short-term memory (LSTM) model. Through horizontal comparison, the LSTM model is found to have the highest accuracy.

[0093] In practical implementation, sentiment classification models can be built using TensorFlow and Keras. TensorFlow is a framework for implementing and executing machine learning algorithms by allowing tensors to flow across a computational graph. TensorFlow is characterized by its flexibility, code portability, automatic differentiation, multi-language support, and high performance. Keras is an open-source artificial neural network library written in Python, which can serve as a high-level application programming interface (API) for TensorFlow, Microsoft-CNTK, and Theano, enabling the design, debugging, evaluation, application, and visualization of deep learning models. Therefore, sentiment classification models can be quickly built using TensorFlow and Keras.

[0094] Step 303: Based on the determined emotion category, obtain the interaction information in the at least one type of interaction information whose emotion category is negative.

[0095] In some embodiments, when the interactive information includes multiple text interactive information, the server obtains the text information associated with the video; determines the semantic similarity between each text interactive information and the text information; and filters the multiple text interactive information based on the semantic similarity to obtain a target text interactive information set.

[0096] In practical applications, the text information associated with the video can be the video's title, summary, tags, etc. When the text information includes multiple types, the semantic similarity between the text interaction information and each piece of text information is determined separately. For example, when the text information is a title and a summary, the semantic similarity between the text interaction information and the title, and the semantic similarity between the text interaction information and the summary, are obtained separately. When any similarity meets a preset threshold, the text interaction information is added to the target text interaction information set.

[0097] In practice, the server can obtain the semantic similarity between text interaction information and text information through a trained neural network model. The text interaction information and text information are input into the neural network model, and the semantic similarity between them is output.

[0098] In some embodiments, the neural network model can be trained as follows: a training sample set is constructed, including text information samples associated with the video, text interaction information samples corresponding to the video, and reference semantic similarity between the text interaction information and the text information; the word vector sequences corresponding to the text interaction information samples and the word vector sequences corresponding to the text information samples are input into the neural network model to obtain the target semantic similarity between the text interaction information samples and the text information samples; based on the output target semantic similarity and reference semantic similarity, the value of the loss function of the neural network model is determined; and based on the value of the loss function, the model parameters of the neural network model are updated.

[0099] The loss function of the neural network model is:

[0100]

[0101] in, y2 is the target semantic similarity of the output, y2 is the reference semantic similarity, and ω2(θ) is the current model parameter.

[0102] It should be noted that the semantic similarity between text interaction information and text information represents the degree of relevance between the text interaction information and the video. Interaction information is filtered by semantic similarity to remove text interaction information with low relevance to the video.

[0103] In some embodiments, the sentiment classification model can be a dual-output model. The input of the sentiment classification model includes not only text interaction information but also text information from the video. The text interaction information and the text information are input into the trained sentiment classification model, and the output is the sentiment category corresponding to the text interaction information and the similarity between the text interaction information and the text information.

[0104] In practice, since the sentiment classification model is a dual-output model, it has two loss functions, each corresponding to one of the two outputs. During training, the model updates its parameters separately based on the two loss functions.

[0105] Thus, this embodiment of the invention only needs to train one sentiment classification model to simultaneously obtain the sentiment category corresponding to the text interaction information, as well as the similarity between the text interaction information and the text information.

[0106] In some implementations, the server can obtain text interaction information with a negative sentiment category for the corresponding video in the following way: based on the determined sentiment category, determine the text interaction information with the sentiment category of the target category from the target text interaction information set.

[0107] In practice, the server obtains text interaction information that is related to the video content and has a negative emotional category. This can filter out some mindless and abusive text interaction information and improve the accuracy of identifying the video as a target video with negative content.

[0108] Step 304: When the interactive information of the negative category meets the preset conditions, the video is determined to be the target video carrying negative content.

[0109] In some embodiments, the server determines the number of interactive messages with a negative emotional category; when the number of interactive messages of a target category meets a preset condition, the video is determined to be a target video carrying negative content.

[0110] In some embodiments, the server can set a quantity threshold to determine whether the number of interactive messages for a target category exceeds a preset quantity threshold. If it does, the video is determined to be a target video carrying negative content; otherwise, the video is not a target video carrying negative content. For example, if the quantity threshold is set to 100 and the interactive message is comments, the video is determined to be a target video carrying negative content when the number of negative comments exceeds 100.

[0111] In some embodiments, the server obtains the number of interactive messages for the target category; obtains the total number of interactive messages for the corresponding video; and determines the video as the target video when the ratio of the number of interactive messages for the negative category to the total number of interactive messages reaches a preset ratio threshold. For example, if the ratio threshold is set to 30%, and the total number of comments for a video is 100, the number of negative comments is 40, and the ratio of the number of negative comments to the total number of comments is 40%, then the video is determined to be the target video carrying negative information.

[0112] In some embodiments, the server may determine whether a video is a target video carrying negative information by combining the number of interactive information of the target category and the ratio of the number of interactive information of the target category to the total number of interactive information: when the ratio of the number of interactive information of the target category to the total number of interactive information reaches a preset ratio threshold, and the number of interactive information of the target category meets a preset quantity threshold, the video is determined to be a target video.

[0113] For example, if the ratio threshold is set to 30% and the quantity threshold is set to 100, then when the number of negative comments is 5 and the total number of comments is 10, although the ratio meets the preset ratio threshold, it cannot be determined that the video is the target video.

[0114] In some embodiments, the server obtains text interaction information and non-text interaction information from the negative category of interaction information; determines a first quantity of text interaction information and a second quantity of non-text interaction information; and determines the video as the target video when the first quantity satisfies the first quantity condition and the second quantity satisfies the second quantity condition.

[0115] In practice, non-textual interactive information for negative categories can be "likes" or "dislikes." A video is determined to carry negative content when the number of negative comments meets a first condition and the number of dislikes meets a second condition. For example, if the first condition is greater than 100 comments and the second condition is greater than 500 comments, then a video is determined to carry negative content when the number of negative comments is 200 and the number of dislikes is 600.

[0116] In some embodiments, when the number of interactive messages for a target category meets a preset condition, the server will further determine whether the posterior data meets the preset condition. Here, the posterior data can be user page views (PVPageView), visit counts (VV), likes, comments, and shares, etc.

[0117] For example, when the interactive information is a comment, after determining that the number of comments meets the preset conditions, it is further determined whether the user's page views meet the conditions. Only when the user's page views also meet the conditions will the video be determined as the target video.

[0118] This invention improves the effectiveness of interactive data by determining whether the posterior data meets preset conditions, thereby improving the accuracy of identifying the video as the target video.

[0119] In some embodiments, when the server determines that a video is a target video carrying negative content, it deletes the target video; or, it sends a prompt message to the terminal regarding the target video, the prompt message being used to indicate that the target video carries negative content.

[0120] In practice, the server can delete the target video from its video resource library to prevent users from watching videos containing negative content; or when a user plays the target video, a prompt message for the target video can be sent to the terminal, indicating that the video contains negative content; or the target video can be reported to the administrator, who can then restrict the playback permissions of the target video.

[0121] This invention, through the mapping relationship between interactive information and emotion categories, determines the emotion category of at least one interactive information, thereby obtaining interactive information with a negative emotion category among the at least one interactive information. When the interactive information with the negative emotion category meets preset conditions, the video is determined to be a target video carrying negative content. In this way, the interactive information with the negative emotion category can be automatically determined according to the mapping relationship between interactive information and emotion categories, and then the video can be determined to be a target video carrying negative content based on the interactive information with the negative emotion category, thus realizing the automatic identification of videos carrying negative content.

[0122] The following uses interactive information as an example to illustrate the video processing method provided by this invention. Figure 4 This is a flowchart illustrating the video processing method provided in an embodiment of the present invention. See [link / reference]. Figure 4 The video processing method provided in this embodiment of the invention includes:

[0123] Step 401: The server retrieves the titles of the videos in the video resource library and all comments for the corresponding videos.

[0124] Step 402: Use the jieba word segmentation tool to segment the title and each comment separately to obtain multiple word sets for the corresponding title and each comment.

[0125] Step 403: Based on the Chinese word vector corpus, obtain the word vector corresponding to each word in each word set to obtain multiple word vector sequences for the corresponding title and each comment.

[0126] Step 404: Input the corresponding title and multiple word vector sequences of each comment into the sentiment classification model, and output the similarity between each comment and the title, as well as the sentiment category corresponding to each comment.

[0127] Here, each time the sentiment classification model is input with the word vector sequence of the title and the word vector sequence of a comment, the output is the similarity between the comment and the title and the sentiment category corresponding to the comment. By inputting multiple times, each comment is fed into the sentiment classification model to output the similarity between each comment and the title and the sentiment category corresponding to each comment.

[0128] In practice, the training can be achieved by: constructing a training sample set, which includes the titles of video samples, comments of video samples, sentiment categories of comments, and reference similarity between comments and titles; inputting the word vector sequences corresponding to comments and titles into the sentiment classification model to obtain the target sentiment category of comments and the similarity between comments and titles; and updating the model parameters of the sentiment classification model based on the differences between the output target sentiment category and reference sentiment category, and the differences between the output target similarity and reference similarity.

[0129] Step 405: Based on the similarity between each comment and the title, select comments from all comments whose similarity to the title exceeds a preset threshold to form a target comment set.

[0130] Step 406: Based on the corresponding sentiment category of each comment, filter out comments with a negative sentiment category from the target comment set.

[0131] Step 407: Obtain the number of comments with negative sentiment after filtering, and the total number of comments for the video.

[0132] Step 408: Calculate the proportion of negative comments to the total number of comments.

[0133] Step 409: When the percentage is greater than 30%, obtain the video's PV, VV, likes, comments, and shares.

[0134] Step 410: When the video's PV, VV, likes, comments, and shares are all greater than 500, the video is identified as the target video.

[0135] Step 411: Send the target video back to the administrator client.

[0136] Step 412: The administrator sends a command to delete the target video via the client.

[0137] Step 413: The server deletes the target video from the video resource library.

[0138] The video processing method provided in the embodiments of the present invention will be described below. Figure 5 This is a flowchart illustrating the video processing method provided in an embodiment of the present invention. In actual implementation, the video processing method includes a model training part and a part that applies the trained model to process the video.

[0139] First, let's explain the training part of the model.

[0140] In practice, the server extracts a subset of typical comments and labels them positively or negatively, using these extracted comments and their labels as a sample set. Simultaneously, it obtains the title or summary of the video to which the comment belongs, as well as the relevance of the comment to the video. This sample set includes the extracted video comments, their positive / negative labels, the video title or summary, and the relevance of the comment to the video. Specifically, the server analyzes content with high daily negative feedback, identifies frequently posted negative comments, labels them as negative samples, and uses this as a basis to build rules. These rules are then used to extract more similar comments as new negative samples, thus quickly obtaining a large number of negatively labeled samples.

[0141] In practice, the data in the sample set is randomly shuffled and divided into a training set, a validation set, and a test set according to a certain ratio.

[0142] Next, a sentiment classification model was built using TensorFlow and Keras, and the network structure of the model was adaptively adjusted to suit business needs. Here, the input to the sentiment classification model includes summaries or titles of comments and videos, and the output includes positive / negative predictions of comments and the relevance of comments to videos, in order to identify mindless abusive comments.

[0143] Here, multiple sentiment classification models were built. Parameters were adjusted on each model using training and validation sets for longitudinal comparison. The performance of each model was then evaluated using a test set. The best-performing parameter combination was selected from each model, and then the models with the optimal parameters were compared horizontally to determine the final optimal model. For example, text-CNN, GRU, and LSTM models could be built simultaneously. In practice, horizontal comparisons revealed that the LSTM model had the highest accuracy.

[0144] It should be noted that the input model consists of word vector sequences corresponding to comments and word vector sequences corresponding to titles or summaries, obtained through word segmentation tools and word vector conversion tools.

[0145] In practice, an open-source word segmentation tool and word vectors were used as the pipeline. The word segmentation tool used was "jieba," and the word vectors used were from the open-source Chinese word vector corpus from Beijing Normal University and Renmin University of China. The pipeline allows for the replacement of previous modules with new word segmentation tools or word vectors, enabling the adoption of new modules to build the model.

[0146] Next, the part of applying the trained model to process the video will be explained.

[0147] First, acquire massive amounts of comment data from the entire internet, along with the titles or summaries of the videos corresponding to each comment. Input each comment and its corresponding title or summary into a trained sentiment feature model, outputting the sentiment category of each comment and its relevance to the video. Divide the comments according to different videos to obtain the proportion of negative comments for each video. Here, when determining the proportion of negative comments for each video, irrelevant negative comments are removed. Videos with a proportion of negative comments exceeding a preset value are filtered out, and then a second filtering is performed based on posterior data, such as TV / VV rating, number of likes, number of comments, and number of shares. For example, the k videos with the most comments can be reported to the administrator.

[0148] The video processing apparatus provided in the embodiments of the present invention will be described next. In some embodiments, the video processing apparatus may be implemented using software modules. Figure 6 This is a schematic diagram of the composition of the video processing apparatus provided in an embodiment of the present invention. See also: Figure 6 The video processing device includes:

[0149] The information acquisition unit 601 is used to acquire at least one interactive information of the corresponding video, the interactive information being used to characterize the user's emotional tendency toward the video;

[0150] The category determination unit 602 is used to determine the emotional category of the at least one interactive information based on the mapping relationship between interactive information and emotional categories;

[0151] Information filtering unit 603 is used to obtain interactive information in the at least one type of interactive information whose emotional category is negative;

[0152] The video determination unit 604 is used to determine that the video is a target video carrying negative content when the interactive information of the negative category meets the preset conditions.

[0153] This invention also provides a server, comprising:

[0154] Memory, used to store executable instructions;

[0155] The processor is configured to execute executable instructions stored in the memory to implement the video processing method provided in the embodiments of the present invention.

[0156] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0157] As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborating files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0158] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0159] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of the present invention are included within the scope of protection of the present invention.

Claims

1. A video processing method, characterized in that: The method comprises: Acquire at least one kind of interactive information corresponding to the video, where the interactive information is used to characterize the user's emotional tendency toward the video; When the interactive information includes non-text interactive information, determining the emotional category of the non-text interactive information according to a mapping relationship between the attribute of the non-text interactive information and the emotional category; When the interactive information includes multiple text interactive information, obtain text information associated with the video; respectively determine the semantic similarity between each text interactive information and the text information; according to the semantic similarity, filter the multiple text interactive information to obtain a target text interactive information set; obtain a word vector sequence corresponding to the text interactive information in the target text interactive information set; input the word vector sequence corresponding to the text interactive information into a sentiment classification model, and output the sentiment category corresponding to the text interactive information; According to the determined emotion category, obtaining the interaction information whose emotion category is a negative category in the at least one interaction information; Obtaining the amount of interaction information of the negative category; When the amount of the interactive information of the negative category meets a preset amount threshold, obtaining text interactive information and non-text interactive information in the interactive information of the negative category; Determining a first amount of the text interaction information and a second amount of the non-text interaction information; When the first number is greater than a first number threshold, the second number is greater than a second number threshold, and the data volume of the video's posterior data is greater than a data volume threshold, the video is determined to be a target video carrying negative content, wherein the posterior data includes at least one of the following: page views, number of visits, number of likes, number of comments, and number of reposts.

2. The method according to claim 1, characterized in that The method further comprises: Constructing a training sample set, wherein the training sample set includes interactive information samples and reference emotion categories corresponding to the interactive information samples; Inputting the word vector sequence corresponding to the interactive information sample into the sentiment classification model, and outputting the target sentiment category corresponding to the interactive information sample; Based on the difference between the target emotion category and the reference emotion category, the model parameters of the emotion classification model are updated.

3. The method according to claim 1, characterized in that The step of obtaining a word vector sequence corresponding to the text interaction information in the target text interaction information set includes: Performing word segmentation processing on the text interaction information in the target text interaction information set to obtain a word set of the text interaction information; Obtain the word vector corresponding to each word in the word set; The word vectors corresponding to each word in the word set are combined into a word vector sequence corresponding to the text interaction information.

4. The method according to claim 1, characterized in that The method further comprises: Deleting the target video; or, Prompt information for the target video is sent to a terminal, where the prompt information is used to prompt that the target video carries negative content.

5. A video processing device, characterized in that: The device comprises: An information acquisition unit, used to acquire at least one kind of interactive information corresponding to a video, wherein the interactive information is used to characterize a user's emotional tendency toward the video; a category determination unit, configured to determine the emotion category of the non-text interaction information according to a mapping relationship between an attribute of the non-text interaction information and an emotion category when the interaction information includes non-text interaction information; An information screening unit is used to obtain text information associated with the video when the interactive information includes multiple text interactive information; respectively determine the semantic similarity between each text interactive information and the text information; screen the multiple text interactive information according to the semantic similarity to obtain a target text interactive information set; obtain a word vector sequence corresponding to the text interactive information in the target text interactive information set; input the word vector sequence corresponding to the text interactive information into a sentiment classification model, and output the sentiment category corresponding to the text interactive information; according to the determined sentiment category, obtain the interactive information whose sentiment category is a negative category in the at least one interactive information; A video determination unit is used to obtain the number of interactive information of the negative category; when the number of interactive information of the negative category meets a preset number threshold, obtain text interactive information and non-text interactive information in the interactive information of the negative category; determine a first number of the text interactive information and a second number of the non-text interactive information; when the first number is greater than a first number threshold, the second number is greater than a second number threshold, and the data volume of the posterior data of the video is greater than a data volume threshold, determine that the video is a target video carrying negative content, wherein the posterior data includes at least one of the following: page views, number of visits, number of likes, number of comments, and number of reposts.

6. A server, characterized in that: The server comprises: A memory for storing executable instructions; A processor, configured to implement the video processing method according to any one of claims 1 to 4 when running the executable instructions stored in the memory.

7. A computer-readable storage medium storing executable instructions, characterized in that: When the executable instructions are executed by a processor, the video processing method described in any one of claims 1 to 4 is implemented.

Images