Content interception and model training method and device and electronic equipment

Abstract

The invention provides a content interception and model training method and device and electronic equipment, and relates to the technical field of artificial intelligence, in particular to the technical fields of computer vision, deep learning, large models and the like. According to the specific scheme, the method comprises the steps of obtaining to-be-processed multimedia content and an interception demand text; inputting the multimedia content and the interception demand text into a content interception model, and obtaining an interception result text output by the content interception model; and performing interception control processing on the multimedia content according to the interception result text.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, specifically to technical fields such as computer vision, deep learning, and large models, and in particular to a content interception and model training method, device, and electronic device. Background Art

[0002] With the rapid development of the Internet, especially the popularity of social platforms, multimedia content on the Internet has exploded. This phenomenon has not only enriched the way of information dissemination, but also provided users with more channels for expression and interaction. In order to create a positive and friendly online environment and protect the physical and mental health of minors, how to ensure the compliance of multimedia content has become an urgent problem to be solved. Summary of the invention

[0003] The present disclosure provides a content interception and model training method, device and electronic device.

[0004] According to one aspect of the present disclosure, a content interception method is provided, the method comprising:

[0005] Obtain multimedia content to be processed and intercept demand text;

[0006] Inputting the multimedia content and the interception requirement text into a content interception model, and obtaining an interception result text output by the content interception model;

[0007] The multimedia content is intercepted and controlled according to the interception result text.

[0008] According to another aspect of the present disclosure, a method for training a content interception model is provided, the method comprising:

[0009] Acquire training data; the training data includes: sample multimedia content, and sample interception requirement text and sample interception result text corresponding to the sample multimedia content;

[0010] Get the initial content blocking model;

[0011] Inputting the sample multimedia content and the sample interception requirement text into the content interception model, and obtaining the predicted interception result text output by the content interception model;

[0012] The content interception model is trained according to the predicted interception result text and the sample interception result text to obtain a trained content interception model.

[0013] According to another aspect of the present disclosure, a content interception device is provided, the device comprising:

[0014] A first acquisition module, used to acquire multimedia content to be processed and intercept demand text;

[0015] A second acquisition module, configured to input the multimedia content and the interception requirement text into a content interception model, and obtain an interception result text output by the content interception model;

[0016] The interception module is used to perform interception control processing on the multimedia content according to the interception result text.

[0017] According to another aspect of the present disclosure, a training device for a content interception model is provided, the device comprising:

[0018] A first acquisition module is used to acquire training data; the training data includes: sample multimedia content, and sample interception requirement text and sample interception result text corresponding to the sample multimedia content;

[0019] A second acquisition module is used to acquire an initial content interception model;

[0020] An input module, used for inputting the sample multimedia content and the sample interception requirement text into the content interception model, and obtaining the predicted interception result text output by the content interception model;

[0021] The training module is used to train the content interception model according to the predicted interception result text and the sample interception result text to obtain a trained content interception model.

[0022] According to another aspect of the present disclosure, there is provided an electronic device, including:

[0023] at least one processor; and

[0024] a memory communicatively connected to the at least one processor; wherein,

[0025] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in one aspect of the present disclosure; or perform the method described in another aspect of the present disclosure.

[0026] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute the method described in one aspect of the present disclosure; or, to execute the method described in another aspect of the present disclosure.

[0027] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, wherein when the computer program is executed by a processor, the computer program implements the method described in one aspect of the present disclosure; or executes the method described in another aspect.

[0028] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0030] Figure 1 A flowchart of a content interception method provided in the first embodiment of the present disclosure;

[0031] Figure 2 A schematic diagram of the content interception principle provided by an embodiment of the present disclosure;

[0032] Figure 3 A flowchart of a content interception method provided in Embodiment 2 of the present disclosure;

[0033] Figure 4 A flowchart of a content interception method provided in Embodiment 3 of the present disclosure;

[0034] Figure 5 A flowchart of a method for training a content interception model provided in Embodiment 4 of the present disclosure;

[0035] Figure 6 A flowchart of a method for training a content interception model provided in Embodiment 5 of the present disclosure;

[0036] Figure 7 A schematic diagram of determining candidate interception reasons provided by an embodiment of the present disclosure;

[0037] Figure 8 A flowchart of a method for training a content interception model provided in Embodiment 6 of the present disclosure;

[0038] Fig. 9 This is a schematic diagram of the structure of a content interception device provided in Embodiment 7 of the present disclosure;

[0039] Fig.10 A schematic diagram of the structure of a training device for a content interception model provided in Embodiment 8 of the present disclosure;

[0040] Fig.11 A schematic block diagram of an example electronic device that can be used to implement embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0041] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0042] In the related technologies, manual review or a single-mode interception model is used to intercept content. However, manual review is costly and slow, which is difficult to meet the needs of the modern network environment. The risk interception model has the following shortcomings:

[0043] (1) Low interception accuracy and recall rate: Low interception accuracy may lead to the mistaken deletion of harmless content, affecting user experience, while low recall rate may allow harmful or prohibited content to escape review;

[0044] (2) Poor interpretability: Neural network models are often viewed as “black boxes” and their decision-making process is not transparent. When false positives or negative positives occur, it is difficult to analyze the causes of model errors.

[0045] (3) Lack of personalized interception capabilities: Different platforms may have different requirements for content interception. For example, online education platforms where minors are the main users need to strictly intercept related prohibited content, while platforms that are mainly for adults can moderately relax the review standards;

[0046] (4) Insufficient ability to respond to emergencies: After an emergency (such as a major incident), there is an urgent need to quickly intercept relevant content, but traditional models cannot immediately identify untrained content, and training new models requires time and cost, resulting in missing the best interception opportunity;

[0047] (5) Single-modal interception capability: Existing models can usually only process information in a single modality, such as text or pictures. However, when risky content appears in the form of a combination of text and pictures or multiple pictures, existing models are difficult to effectively intercept.

[0048] In view of the above problems, the present disclosure proposes a content interception and model training method, device and electronic device.

[0049] Figure 1 This is a flow chart of the content interception method provided in the first embodiment of the present disclosure. It should be noted that the content interception method in the embodiment of the present disclosure can be applied to a content interception device, which can be configured in an electronic device so that the electronic device can perform a content interception function.

[0050] Among them, the electronic device can be any device with computing capabilities, such as a personal computer (PC), a mobile terminal, a server, etc. The mobile terminal can be, for example, a vehicle-mounted device, a mobile phone, a tablet computer, a personal digital assistant, a wearable device, a smart speaker, a server, a server cluster, and other hardware devices with various operating systems, touch screens and / or display screens.

[0051] The content interception device may also be software in the electronic device, such as content interception software, etc. In the following embodiments, the execution subject is an electronic device as an example for description.

[0052] like Figure 1 As shown, the content interception method may include the following steps:

[0053] Step 101, obtaining the multimedia content to be processed and the interception requirement text.

[0054] In the embodiments of the present disclosure, the multimedia content to be processed may be digital content in various forms obtained from a network platform or uploaded by a user, including but not limited to pictures, videos, audios, documents, etc.

[0055] In order to achieve personalized interception, the interception requirement text can be used to guide the content interception model. The interception requirement text is used to guide the content interception model which content or behaviors need to be identified and intercepted, as well as specific interception standards, etc. For example, the interception requirement text indicates at least one of the following: intercepting risk elements, providing reasons for interception, and determining whether the first risk element exists in the multimedia content.

[0056] Among them, in order to improve the pertinence and accuracy of interception, the first risk element can be determined according to the interception scenario; for example, the first risk element can be the key factor that is most likely to cause risks or problems in the scenario, such as the first risk element is sensitive information, a prohibited behavior, etc.; or, the first risk element is determined according to the emergency, and the first risk element can be a key factor directly related to the emergency, for example, the first risk element is a social security incident, etc.

[0057] As an example, an interception requirement text adapted to the user's requirement may be obtained from a setting rule library.

[0058] As another example, relevant personnel may manually input interception requirement text through the interface.

[0059] Step 102: input the multimedia content and the interception requirement text into the content interception model, and obtain the interception result text output by the content interception model.

[0060] Furthermore, in order to improve the interpretability of interception, the multimedia content and interception requirement text are input into the content interception model, and the content interception model automatically identifies potential risk elements in the multimedia content, and generates a corresponding interception result text according to the interception criteria provided by the interception requirement text; wherein the interception result text is used to indicate at least one of the following: whether risk elements are included, whether interception reasons are included, and whether the first risk element exists in the multimedia content.

[0061] For example, Figure 2 As shown, according to the user-defined interception content (first risk element) and whether the interception reason needs to be explained, an interception requirement text (generated question) 201 is generated, and the interception requirement text indicates whether there is a first risk element in the multimedia content. Then, the multimedia content to be processed and the interception requirement text are input into the content interception model 202 to obtain the interception result text (model answer) output by the content interception model. It should be noted that if the user-defined interception content is empty, the interception requirement text is a general text "whether there is a risk element in the multimedia content".

[0062] Step 103: perform interception control processing on the multimedia content according to the interception result text.

[0063] In the disclosed embodiment, the multimedia content including prohibited content is automatically intercepted and controlled based on the interception result text, for example, the multimedia content including prohibited content is deleted, blocked, marked, etc.

[0064] The content interception method of the disclosed embodiment obtains the multimedia content to be processed and the interception requirement text; inputs the multimedia content and the interception requirement text into the content interception model to obtain the interception result text output by the content interception model; and performs interception control processing on the multimedia content according to the interception result text. Thus, the content interception model is used to intercept and identify the multimedia content based on the interception requirement text with personalized configuration to obtain the interception result text, and the interception control processing is performed based on the interception result text. This not only realizes automated content review, reduces manual intervention, and improves the efficiency and accuracy of content review, but also can realize the identification and interception of various prohibited contents, enhances the flexibility and adaptability of content review, and improves the health and safety of the network environment.

[0065] In order to clearly explain how to intercept and control multimedia content according to the interception result text in any embodiment of the present disclosure, the present disclosure proposes another content interception method.

[0066] Figure 3 This is a flow chart of the content interception method provided in the second embodiment of the present disclosure.

[0067] like Figure 3As shown, the content interception method may include the following steps:

[0068] Step 301, obtaining the multimedia content to be processed and the interception requirement text.

[0069] Step 302: Input the multimedia content and the interception requirement text into the content interception model, and obtain the interception result text output by the content interception model.

[0070] Step 303: When the interception result text includes risk elements, the multimedia content is intercepted.

[0071] In order to effectively intercept various types of prohibited content, as an example, in response to the interception requirement text indicating interception of risk elements, when the interception result text includes risk elements, it indicates that the multimedia content includes prohibited or harmful content. In order to improve the health and safety of the network environment, the multimedia content can be intercepted.

[0072] It should be noted that, in order to further improve the security of multimedia content, when the interception request text also indicates to provide reasons for interception, content adjustment processing can also be performed on risk elements in the intercepted multimedia content.

[0073] Step 304: If the interception result text does not include any risk elements, the multimedia content is released.

[0074] As another example, in response to the interception requirement text indicating interception of risk elements, if the interception result text does not include risk elements, it means that the multimedia content is healthy or safe content and the multimedia content can be released.

[0075] It should be noted that step 303 and step 304 are parallel solutions, and only one is executed during actual execution.

[0076] In addition, it should be noted that the execution process of steps 301 to 302 can be implemented in any way in the embodiments of the present disclosure, and the embodiments of the present disclosure do not limit this and will not be repeated.

[0077] The content interception method of the disclosed embodiment responds to the interception requirement text indicating the interception of risk elements, and intercepts the multimedia content when the risk elements are included in the interception result text; or releases the multimedia content when the risk elements are not included in the interception result text. Thus, based on the judgment of the risk elements in the interception result text, it is automatically decided whether to intercept or release the content, thereby reducing manual intervention and improving content review efficiency.

[0078] In order to clearly explain how to intercept and control multimedia content according to the interception result text in any embodiment of the present disclosure, the present disclosure proposes another content interception method.

[0079] Figure 4 This is a flow chart of the content interception method provided in the third embodiment of the present disclosure.

[0080] like Figure 4 As shown, the content interception method may include the following steps:

[0081] Step 401, obtaining the multimedia content to be processed and the interception requirement text.

[0082] Step 402: Input the multimedia content and the interception requirement text into the content interception model, and obtain the interception result text output by the content interception model.

[0083] Step 403: When the interception result text indicates that the multimedia content contains the first risk element, the multimedia content is intercepted.

[0084] As an example, the interception requirement text indicates determining whether there is a first risk element in the multimedia content, and if the interception result text indicates that there is the first risk element in the multimedia content, the multimedia content is intercepted.

[0085] For example, the multimedia content is a picture, and the interception requirement text is "Is there a game screen with a certain risk element in the picture?", and the interception result text is "Yes", which means that there is a game screen with the certain risk element in the picture, and then the picture is intercepted.

[0086] Step 404: When the interception result text indicates that the multimedia content does not contain the first risk element, the multimedia content is released.

[0087] As another example, when the interception result text clearly indicates that the first risk element does not exist in the multimedia content, the content can be released directly without further reviewing other risk elements, thereby reducing the time and resource consumption of the review.

[0088] It should be noted that step 403 and step 404 are parallel solutions, and only one is executed during actual execution.

[0089] In addition, it should be noted that the execution process of steps 401 to 402 can be implemented in any way in the embodiments of the present disclosure, and the embodiments of the present disclosure do not limit this and will not be repeated.

[0090] The content interception method of the disclosed embodiment determines whether there is a first risk element in the multimedia content in response to an interception requirement text indication, and intercepts the multimedia content when the interception result text indicates that the first risk element exists in the multimedia content; or releases the multimedia content when the interception result text indicates that the first risk element does not exist in the multimedia content. Thus, by processing the multimedia content based on the interception requirement text and the interception result text, it is possible to efficiently and accurately determine whether the multimedia content contains the first risk element, so that when the multimedia content contains the first risk element, the multimedia content is timely intercepted, and when the multimedia content does not contain the first risk element, the multimedia content is released, effectively improving the efficiency and security of content processing.

[0091] In order to implement the above embodiment, the present disclosure proposes a training method for a content interception model.

[0092] Figure 5 A flowchart of a content interception model training method provided in Embodiment 4 of the present disclosure.

[0093] like Figure 5 As shown, the training method of the content blocking model may include the following steps:

[0094] Step 501, obtaining training data.

[0095] The training data includes: sample multimedia content, sample interception requirement text and sample interception result text corresponding to the sample multimedia content.

[0096] In the embodiments of the present disclosure, training data can be extracted from historical audit records, or manually annotated data can be used as training data, such as manually annotating multimedia content to obtain sample multimedia content, manually checking each multimedia content according to the requirements of the interception requirement text (sample interception requirement text), marking the risk elements therein, and generating corresponding interception result text (sample interception result text); or, synthetic or simulated data can be used to supplement the training data, for example, by using image generation technology to create a picture containing specific risk elements, or by using text generation technology to create a sample interception result text containing sensitive words.

[0097] Among them, it should be noted that in order to realize multimodal processing of content interception, the sample multimedia content includes sample images and / or sample texts; in order to realize personalized interception and accuracy of interception, the sample interception requirement text indicates at least one of the following: intercepting risk elements, providing reasons for interception, and judging whether there is a first risk element in the multimedia content, the first risk element is a designated risk element among the risk elements, and the first risk element is, for example: prohibited behavior, harmful behavior or prohibited content, etc. The reason for interception is used to indicate the basis and reason for interception, etc.

[0098] Step 502: Obtain an initial content interception model.

[0099] It should be understood that pre-trained deep learning models (such as BERT, GPT, CLIP, etc.) have extensive natural language understanding, image recognition, audio analysis and other capabilities. Therefore, in the embodiments of the present disclosure, the initial content interception model can be a large model.

[0100] Step 503: Input the sample multimedia content and the sample interception requirement text into the content interception model, and obtain the predicted interception result text output by the content interception model.

[0101] In order to achieve effective training of the content interception model, in the embodiment of the present disclosure, sample multimedia content and sample interception requirement text are input into the initial content interception model, and the content interception model can output the predicted interception result text.

[0102] Step 504: training the content interception model according to the predicted interception result text and the sample interception result text to obtain a trained content interception model.

[0103] Furthermore, a loss function value is generated according to the difference between the predicted interception result text and the sample interception result text, and the content interception model is trained using the loss function value.

[0104] The training method of the content interception model of the embodiment of the present disclosure is through obtaining training data; the training data includes: sample multimedia content, and sample interception requirement text and sample interception result text corresponding to the sample multimedia content; obtaining an initial content interception model; inputting the sample multimedia content and the sample interception requirement text into the content interception model, and obtaining the predicted interception result text output by the content interception model; training the content interception model according to the predicted interception result text and the sample interception result text to obtain the trained content interception model, thereby achieving effective training of the content interception model, so that the trained content interception model can effectively intercept and identify the multimedia content based on the personalized configured interception requirement text, and obtain the interception result text, so as to accurately intercept and control the multimedia content based on the interception result text.

[0105] In order to clearly illustrate how to obtain training data in any embodiment of the present disclosure, the present disclosure proposes another training method for a content interception model.

[0106] Figure 6 A flowchart of a content interception model training method provided in Embodiment 5 of the present disclosure.

[0107] like Figure 6 As shown, the training method of the content blocking model may include the following steps:

[0108] Step 601: Obtain sample multimedia content, sample risk elements in the sample multimedia content, and sample interception requirement text corresponding to the sample multimedia content.

[0109] In the disclosed embodiment, sample multimedia content is extracted from historical data, or multimedia content captured from a network platform is used as sample multimedia content; then, sample risk elements are identified from the sample multimedia content, or sample risk elements in the sample multimedia content are manually annotated; according to the sample risk elements, relevant regulations are cited, and the reasons for intercepting the corresponding sample multimedia content are explained, and specific problems existing in the sample multimedia content are pointed out, etc., thereby generating a sample interception requirement text based on at least one of the relevant regulations corresponding to the sample risk elements, the reasons for intercepting the sample multimedia content, and the specific problems existing in the sample multimedia content. The sample interception requirement text indicates at least one of the following: intercepting risk elements, providing reasons for interception, and judging whether there is a first risk element in the multimedia content.

[0110] Step 602: Determine the description text of the sample multimedia content.

[0111] In the embodiment of the present disclosure, the sample multimedia content is identified and described to generate a description text, wherein the description text is used to explain or summarize the sample multimedia content, and the description text includes: image description text of the sample image and / or sample text.

[0112] As an example, taking the sample multimedia content as a picture, the sample multimedia content is input into a trained picture description model, and a description text output by the picture description model can be obtained.

[0113] As another example, taking the sample multimedia content as text, natural language processing (NLP) technology is used to extract keywords in the text, and text summary content is generated according to the extracted keywords, and the generated text summary content is used as the description text.

[0114] As another example, taking the example that the sample multimedia content includes images and texts, for the content including images and texts, the description text can be generated by combining the visual elements in the images and the semantic information in the texts.

[0115] Step 603: Determine the risk warning text corresponding to the sample multimedia content according to the sample risk element.

[0116] In the disclosed embodiment, the sample risk element is a prohibited element in the sample multimedia content, and the risk warning text is used to explain why the sample multimedia content is risky.

[0117] As an example, fill the sample risk elements into the risk warning template to generate the corresponding risk warning text.

[0118] As another example, risk warning information of multiple risk elements is generated in advance, and a query is performed based on sample risk elements to obtain corresponding risk warning texts.

[0119] Step 604: input the description text and the risk warning text into a reason generation model to obtain candidate interception reasons output by the reason generation model.

[0120] In order to enable users to clearly understand the reasons why content is blocked and improve the explainability of content blocking, in the embodiment of the present disclosure, the description text and the risk warning text are input into the reason generation model to obtain the candidate interception reasons output by the reason generation model.

[0121] In the disclosed embodiment, the candidate interception reasons may include, but are not limited to: specific problems existing in the sample multimedia content, reasons for interception, suggestions for improvement, etc.

[0122] Step 605: determine the candidate interception reasons and the sample risk elements as sample interception result text corresponding to the sample multimedia content.

[0123] In the embodiment of the present disclosure, the sample interception result text refers to the text of the final interception decision generated after reviewing the sample multimedia content. As an example, the candidate interception reasons and sample risk elements are combined to generate the sample interception result text corresponding to the sample multimedia content.

[0124] In order to improve the accuracy of the sample interception results, in the embodiment of the present disclosure, before the candidate interception reasons and sample risk elements are determined as the sample interception result text corresponding to the sample multimedia content, it is necessary to judge whether the candidate interception reasons are reasonable. When the candidate interception reasons are reasonable, the candidate interception reasons are retained; when the candidate interception reasons are unreasonable, the candidate interception reasons are deleted.

[0125] In order to improve the accuracy of judging whether the candidate interception reason is reasonable, as an example, the candidate interception reason and the description text are input into the reason judgment model to obtain the reason judgment result output by the reason judgment model; when the reason judgment result indicates that the candidate interception reason is reasonable, the candidate interception reason is retained; or, when the reason judgment result indicates that the candidate interception reason is unreasonable, the candidate interception reason is deleted.

[0126] Among them, in order to improve the accuracy of the reason judgment result, in the embodiment of the present disclosure, the relevant knowledge text corresponding to the description text is obtained, and the description text, the relevant knowledge text and the candidate interception reason are simultaneously input into the reason judgment model to obtain the reason judgment result output by the reason judgment model. Among them, the relevant knowledge text is a text used to indicate the legality, compliance and risk of the content, to help users understand why their content is intercepted, and to provide improvement suggestions and legal basis; the reason judgment result is an evaluation conclusion for each candidate interception reason, including a rationality score, an evaluation description and a recommended operation. Through the reason judgment result, the rationality of each candidate interception reason is determined, and the rationality rate of the candidate interception reason is calculated. When the rationality rate of the interception reason is less than or equal to the rationality rate threshold, the risk warning text is adjusted to improve the rationality of the candidate interception reason generated by the reason generation model.

[0127] For example, Figure 7 As shown, taking the sample multimedia content as a certain type of risk picture as an example, a certain type of risk picture is input into the picture description model 701 to obtain the picture description (description text), the description text and the risk warning text are input into the general multimodal model (reason generation model) 702 to obtain the candidate interception reason, the candidate interception reason and the description text are input into the large language model (reason judgment model) 703, and the rationality of the candidate interception reason is judged by the large language model 703 to obtain the judgment result 704 of the candidate interception reason. When the rationality rate of the interception reason is less than or equal to the rationality rate threshold, the manual adjustment prompt 705 is used to adjust the risk warning text; when the rationality rate of the interception reason is greater than the rationality rate threshold, the candidate interception reason is retained.

[0128] Similarly, taking the sample multimedia content as a certain type of risk text as an example, the certain type of risk text is directly input into the large language model (reason generation model) 711 to obtain candidate interception reasons, and then, the candidate interception reasons and the certain type of risk text are input into the large language model (reason judgment model) 712 to use the large language model to judge the rationality of the candidate interception reasons and obtain the judgment result 713 of the candidate interception reasons. When the rationality rate of the interception reasons is less than or equal to the rationality rate threshold, the manual adjustment prompt 714 is used to adjust the risk prompt text; when the rationality rate of the interception reasons is greater than the rationality rate threshold, the candidate interception reasons are retained.

[0129] Step 606: determine training data according to the sample multimedia content, the sample interception requirement text corresponding to the sample multimedia content, and the sample interception result text.

[0130] Furthermore, the sample multimedia content, the sample interception requirement text corresponding to the sample multimedia content, and the sample interception result text are used as training data.

[0131] Step 607: Obtain an initial content interception model.

[0132] Step 608: Input the sample multimedia content and the sample interception requirement text into the content interception model, and obtain the predicted interception result text output by the content interception model.

[0133] Step 609: training the content interception model according to the predicted interception result text and the sample interception result text to obtain a trained content interception model.

[0134] It should be noted that the execution process of steps 607 to 609 can be implemented in any way in the embodiments of the present disclosure, and the embodiments of the present disclosure do not limit this and will not be described in detail.

[0135] The training method of the content interception model of the embodiment of the present disclosure obtains sample multimedia content, sample risk elements in the sample multimedia content, and sample interception requirement text corresponding to the sample multimedia content; determines the description text of the sample multimedia content; determines the risk warning text corresponding to the sample multimedia content according to the sample risk elements; inputs the description text and the risk warning text into a reason generation model to obtain candidate interception reasons output by the reason generation model; determines the candidate interception reasons and the sample risk elements as sample interception result text corresponding to the sample multimedia content; determines the training data according to the sample multimedia content, the sample interception requirement text corresponding to the sample multimedia content, and the sample interception result text, thereby accurately determining the training data of the content interception model, and training the content interception model based on the training data, so that the trained content interception model can accurately intercept multimedia content containing risk elements while clarifying the interception reasons.

[0136] In order to clearly illustrate how to obtain training data in any embodiment of the present disclosure, the present disclosure proposes another training method for a content interception model.

[0137] Figure 8 A flowchart of a content interception model training method provided in Embodiment 6 of the present disclosure.

[0138] like Figure 8 As shown, the training method of the content blocking model may include the following steps:

[0139] Step 801: Obtain sample multimedia content and description text of the sample multimedia content.

[0140] Step 802: perform entity extraction processing on the description text to obtain candidate entities in the description text.

[0141] In the disclosed embodiment, entity recognition and extraction are performed on the description text to obtain candidate entities in the description text, wherein the candidate entities include: events, dates, times, people, places, etc.

[0142] Step 803: Determine a sample interception requirement text and a sample interception result text corresponding to the sample multimedia content according to the candidate entity in the description text and the interception requirement template text with the entity as a variable.

[0143] The interception requirement template text is used to indicate whether an entity exists in the sample multimedia content.

[0144] In order to accurately determine the sample interception requirement text and the sample interception result text corresponding to the sample multimedia content, in the embodiment of the present disclosure, the interception requirement template text is filled in according to the candidate entities in the description text to obtain the sample interception requirement text corresponding to the sample multimedia content, and the sample interception result text is a text generated based on the sample interception requirement text as the interception standard.

[0145] As a first possible implementation manner, the candidate entity is filled into the interception requirement template text to obtain a sample interception requirement text; and it is determined that the sample interception result text indicates that the candidate entity exists in the sample multimedia content.

[0146] That is to say, since the interception requirement template text is used to indicate whether an entity exists in the sample multimedia content, the candidate entity is filled into the interception requirement template text to obtain the sample interception requirement text. The sample interception requirement text indicates whether a candidate entity exists in the sample multimedia content. If a candidate entity exists in the description text, it means that the candidate entity exists in the sample multimedia content. Therefore, the sample interception result text indicates that a candidate entity exists in the sample multimedia content.

[0147] As a second possible implementation, non-candidate entities other than candidate entities are filled into the interception requirement template text to obtain a sample interception requirement text; and it is determined that the sample interception result text indicates that there is no non-candidate entity in the sample multimedia content.

[0148] In an embodiment of the present disclosure, non-candidate entities other than candidate entities are filled into an interception requirement template text to obtain a sample interception requirement text, which indicates whether there are non-candidate entities in the sample multimedia content. The description text contains candidate entities, that is, only candidate entities exist in the sample multimedia content. It can be determined that the sample interception result text indicates that there are no non-candidate entities in the sample multimedia content.

[0149] Step 804 , determining training data according to the sample multimedia content, the sample interception requirement text corresponding to the sample multimedia content, and the sample interception result text.

[0150] Furthermore, the sample multimedia content, the sample interception requirement text corresponding to the sample multimedia content, and the sample interception result text are used as training data.

[0151] Step 805: Obtain an initial content interception model.

[0152] Step 806: Input the sample multimedia content and the sample interception requirement text into the content interception model, and obtain the predicted interception result text output by the content interception model.

[0153] Step 807: training the content interception model according to the predicted interception result text and the sample interception result text to obtain a trained content interception model.

[0154] It should be noted that the execution process of step 801 and step 805 to step 807 can be implemented in any way in the embodiments of the present disclosure, and the embodiments of the present disclosure do not limit this and will not be repeated.

[0155] The training method of the content interception model of the embodiment of the present disclosure is as follows: obtaining sample multimedia content and a description text of the sample multimedia content; performing entity extraction processing on the description text to obtain candidate entities in the description text; determining a sample interception requirement text and a sample interception result text corresponding to the sample multimedia content according to the candidate entities in the description text and an interception requirement template text with the entity as a variable; wherein the interception requirement template text is used to indicate whether an entity exists in the sample multimedia content; determining training data according to the sample multimedia content, the sample interception requirement text corresponding to the sample multimedia content, and the sample interception result text, thereby accurately determining the training data of the content interception model, and training the content interception model based on the training data, so that the trained content interception model can accurately determine whether the multimedia content contains the first risk element, thereby effectively intercepting the multimedia content containing the first risk element.

[0156] With the above Figures 1 to 4 Corresponding to the content interception method provided in the embodiment, the present disclosure also provides a content interception device. Figures 1 to 4 The content interception method provided in the embodiment corresponds to the content interception method, so the implementation method of the content interception method is also applicable to the content interception device provided in the embodiment of the present disclosure, and will not be described in detail in the embodiment of the present disclosure.

[0157] Fig. 9 This is a schematic diagram of the structure of the content interception device provided in the seventh embodiment of the present disclosure.

[0158] like Fig. 9 As shown, the content interception device 900 includes: a first acquisition module 910 , a second acquisition module 920 and an interception module 930 .

[0159] Among them, the first acquisition module 910 is used to obtain the multimedia content to be processed and the interception requirement text; the second acquisition module 920 is used to input the multimedia content and the interception requirement text into the content interception model to obtain the interception result text output by the content interception model; the interception module 930 is used to intercept and control the multimedia content according to the interception result text.

[0160] As a possible implementation manner of the embodiment of the present disclosure, the interception requirement text indicates at least one of the following: intercepting a risk element, providing a reason for interception, and determining whether a first risk element exists in the multimedia content.

[0161] As a possible implementation method of the embodiment of the present disclosure, the interception requirement text indicates intercepting the risk element; the interception module is specifically used to: intercept the multimedia content when the risk element is included in the interception result text; or release the multimedia content when the risk element is not included in the interception result text.

[0162] As a possible implementation manner of the embodiment of the present disclosure, the interception requirement text also indicates providing an interception reason; the device also includes: an adjustment module.

[0163] The adjustment module is used to adjust the multimedia content according to the interception reason when the interception result text includes the interception reason.

[0164] As a possible implementation method of the embodiment of the present disclosure, the interception requirement text indication determines whether the first risk element exists in the multimedia content; the interception module is specifically used to: when the interception result text indicates that the first risk element exists in the multimedia content, intercept the multimedia content; or when the interception result text indicates that the first risk element does not exist in the multimedia content, release the multimedia content.

[0165] As a possible implementation manner of the embodiment of the present disclosure, the first risk element is determined according to an interception scenario; or, the first risk element is determined according to an emergency event.

[0166] The content interception device of the disclosed embodiment obtains multimedia content to be processed and an interception requirement text; inputs the multimedia content and the interception requirement text into a content interception model to obtain an interception result text output by the content interception model; and intercepts and controls the multimedia content according to the interception result text. Thus, the content interception model is used to intercept and control the multimedia content based on the interception requirement text with personalized configuration, thereby improving the efficiency and accuracy of content review, realizing not only automated content review and reducing the need for manual intervention, but also real-time identification and interception of various prohibited contents, thereby enhancing the flexibility and adaptability of content review and improving the health and safety of the network environment.

[0167] With the above Figures 5 to 8 Corresponding to the training method of the content interception model provided in the embodiment, the present disclosure also provides a training device for the content interception model. Figures 5 to 8The training method of the content interception model provided in the embodiment corresponds to the embodiment, so the implementation method of the content interception model training method is also applicable to the training device of the content interception model provided in the embodiment of the present disclosure, and will not be described in detail in the embodiment of the present disclosure.

[0168] Fig.10 This is a structural diagram of a training device for a content interception model provided in Embodiment 8 of the present disclosure.

[0169] like Fig.10 As shown, the training device 1000 for the content interception model includes: a first acquisition module 1010 , a second acquisition module 1020 , an input module 1030 and a training module 1040 .

[0170] Among them, the first acquisition module 1010 is used to obtain training data; the training data includes: sample multimedia content, and sample interception requirement text and sample interception result text corresponding to the sample multimedia content; the second acquisition module 1020 is used to obtain the initial content interception model; the input module 1030 is used to input the sample multimedia content and the sample interception requirement text into the content interception model to obtain the predicted interception result text output by the content interception model; the training module 1040 is used to train the content interception model according to the predicted interception result text and the sample interception result text to obtain the trained content interception model.

[0171] As a possible implementation manner of the embodiment of the present disclosure, the sample interception requirement text indicates at least one of the following: intercepting risk elements, providing reasons for interception, and determining whether there is a first risk element in the multimedia content.

[0172] As a possible implementation method of the embodiment of the present disclosure, the sample interception requirement text indicates the interception of risk elements and provides reasons for interception; the first acquisition module is specifically used to: obtain sample multimedia content, sample risk elements in the sample multimedia content, and a sample interception requirement text corresponding to the sample multimedia content; determine the description text of the sample multimedia content; determine the risk warning text corresponding to the sample multimedia content based on the sample risk elements; input the text and the risk warning text into a reason generation model to obtain candidate interception reasons output by the reason generation model; determine the candidate interception reasons and the sample risk elements as sample interception result text corresponding to the sample multimedia content; determine training data based on the sample multimedia content, the sample interception requirement text corresponding to the sample multimedia content, and the sample interception result text.

[0173] As a possible implementation manner of the embodiment of the present disclosure, the sample multimedia content includes a sample image and / or a sample text; the description text includes: an image description text of the sample image and / or the sample text.

[0174] As a possible implementation of the embodiment of the present disclosure, the training device 1000 for the content interception model further includes: a third acquisition module and a processing module.

[0175] Among them, the third acquisition module is used to input the candidate interception reason and the description text into the reason judgment model to obtain the reason judgment result output by the reason judgment model; the processing module is used to retain the candidate interception reason when the reason judgment result indicates that the candidate interception reason is reasonable; or delete the candidate interception reason when the reason judgment result indicates that the candidate interception reason is unreasonable.

[0176] As a possible implementation method of the embodiment of the present disclosure, the third acquisition module is specifically used to: obtain the relevant knowledge text corresponding to the description text; input the description text, the relevant knowledge text and the candidate interception reason into the reason judgment model, and obtain the reason judgment result output by the reason judgment model.

[0177] As a possible implementation of the embodiment of the present disclosure, the training device 1000 for the content interception model further includes: a determination module and an adjustment module.

[0178] Among them, the determination module is used to determine the rationality rate of interception reasons based on the judgment results of various reasons; the adjustment module is used to adjust the risk warning text when the rationality rate of interception reasons is less than or equal to the rationality rate threshold.

[0179] As a possible implementation method of the embodiment of the present disclosure, the sample interception requirement text indicates whether a first risk element is present in the multimedia content; the first acquisition module is specifically used to: obtain the sample multimedia content and a description text of the sample multimedia content; perform entity extraction processing on the description text to obtain candidate entities in the description text; determine the sample interception requirement text and the sample interception result text corresponding to the sample multimedia content according to the candidate entities in the description text and an interception requirement template text with the entity as a variable; the interception requirement template text is used to indicate whether an entity is present in the sample multimedia content; determine the training data according to the sample multimedia content, the sample interception requirement text and the sample interception result text corresponding to the sample multimedia content.

[0180] As a possible implementation method of the embodiment of the present disclosure, the first acquisition module is specifically used to: fill the candidate entity into the interception requirement template text to obtain the sample interception requirement text; determine that the sample interception result text indicates that the candidate entity exists in the sample multimedia content; and / or fill non-candidate entities other than the candidate entities into the interception requirement template text to obtain the sample interception requirement text; determine that the sample interception result text indicates that the non-candidate entity does not exist in the sample multimedia content.

[0181] The training method of the content interception model of the embodiment of the present disclosure is through obtaining training data; the training data includes: sample multimedia content, and sample interception requirement text and sample interception result text corresponding to the sample multimedia content; obtaining an initial content interception model; inputting the sample multimedia content and the sample interception requirement text into the content interception model, and obtaining the predicted interception result text output by the content interception model; training the content interception model according to the predicted interception result text and the sample interception result text to obtain the trained content interception model, thereby achieving effective training of the content interception model, so that the trained content interception model can effectively intercept and identify the multimedia content based on the personalized configured interception requirement text, and obtain the interception result text, so as to accurately intercept and control the multimedia content based on the interception result text.

[0182] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information are all carried out with the user's consent, comply with the relevant laws and regulations, and do not violate public order and good morals.

[0183] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0184] Fig.11 A schematic block diagram of an example electronic device 1100 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0185] like Fig.11 As shown, the device 1100 includes a computing unit 1101, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1102 or a computer program loaded from a storage unit 1108 into a random access memory (RAM) 1103. In the RAM 1103, various programs and data required for the operation of the device 1100 can also be stored. The computing unit 1101, the ROM 1102, and the RAM 1103 are connected to each other via a bus 1104. An input / output (I / O) interface 1105 is also connected to the bus 1104.

[0186] A number of components in the device 1100 are connected to the I / O interface 1105, including: an input unit 1106, such as a keyboard, a mouse, etc.; an output unit 1107, such as various types of displays, speakers, etc.; a storage unit 1108, such as a disk, an optical disk, etc.; and a communication unit 1109, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1109 allows the device 1100 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0187] The computing unit 1101 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1101 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1101 performs the various methods and processes described above, such as a content interception method or a training method for a content interception model. For example, in some embodiments, a content interception method or a training method for a content interception model may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 1108. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 1100 via ROM 1102 and / or a communication unit 1109. When the computer program is loaded into RAM 1103 and executed by the computing unit 1101, one or more steps of the content interception method or the training method for the content interception model described above may be executed. Alternatively, in other embodiments, the computing unit 1101 may be configured to execute the content interception method or the content interception model training method in any other appropriate manner (for example, by means of firmware).

[0188] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0189] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0190] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0191] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0192] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0193] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0194] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0195] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A content interception method, the method comprising: Obtain multimedia content to be processed and intercept demand text; Inputting the multimedia content and the interception requirement text into a content interception model, and obtaining an interception result text output by the content interception model; The multimedia content is intercepted and controlled according to the interception result text.

2. The method according to claim 1, wherein: The interception requirement text indicates at least one of the following: intercepting a risk element, providing a reason for interception, and determining whether a first risk element exists in the multimedia content.

3. The method according to claim 1 or 2, wherein: The interception requirement text indicates interception of risk elements; The interception control processing of the multimedia content according to the interception result text includes: In the case where the interception result text includes risk elements, intercepting the multimedia content; or, In the case that the risk element is not included in the interception result text, the multimedia content is released.

4. The method according to claim 3, wherein: The interception requirement text also indicates to provide an interception reason; the method further includes: In a case where the interception result text includes an interception reason, content adjustment processing is performed on the multimedia content according to the interception reason.

5. The method according to claim 1 or 2, wherein: The interception requirement text indicates determining whether there is a first risk element in the multimedia content; The interception control processing of the multimedia content according to the interception result text includes: When the interception result text indicates that the first risk element exists in the multimedia content, intercepting the multimedia content; or, When the interception result text indicates that the first risk element does not exist in the multimedia content, the multimedia content is released.

6. The method according to claim 5, wherein: The first risk element is determined according to an interception scenario; or, the first risk element is determined according to an emergency event.

7. A method for training a content interception model, the method comprising: Get training data; The training data includes: sample multimedia content, and sample interception requirement text and sample interception result text corresponding to the sample multimedia content; Obtain an initial content blocking model; Inputting the sample multimedia content and the sample interception requirement text into the content interception model, and obtaining the predicted interception result text output by the content interception model; The content interception model is trained according to the predicted interception result text and the sample interception result text to obtain a trained content interception model.

8. The method according to claim 7, wherein: The sample interception requirement text indicates at least one of the following: intercepting risk elements, providing reasons for interception, and determining whether a first risk element exists in the multimedia content.

9. The method according to claim 7 or 8, wherein: The sample interception requirement text indicates interception of risk elements and provides reasons for interception; The obtaining of training data comprises: Acquire the sample multimedia content, sample risk elements in the sample multimedia content, and sample interception requirement text corresponding to the sample multimedia content; Determining a description text of the sample multimedia content; Determining, according to the sample risk element, a risk warning text corresponding to the sample multimedia content; Inputting the description text and the risk warning text into a reason generation model to obtain candidate interception reasons output by the reason generation model; Determine the candidate interception reason and the sample risk element as a sample interception result text corresponding to the sample multimedia content; The training data is determined according to the sample multimedia content, the sample interception requirement text corresponding to the sample multimedia content, and the sample interception result text.

10. The method according to claim 9, wherein: The sample multimedia content includes a sample image and / or a sample text; the description text includes: an image description text of the sample image and / or the sample text.

11. The method according to claim 9, wherein: Before determining the candidate interception reasons and the sample risk elements as sample interception result text corresponding to the sample multimedia content, the method further includes: Inputting the candidate interception reason and the description text into a reason judgment model, and obtaining a reason judgment result output by the reason judgment model; If the reason determination result indicates that the candidate interception reason is reasonable, retain the candidate interception reason; or, When the reason judgment result indicates that the candidate interception reason is unreasonable, the candidate interception reason is deleted.

12. The method according to claim 11, wherein: The step of inputting the candidate interception reason and the description text into a reason judgment model and obtaining a reason judgment result output by the reason judgment model comprises: Obtaining relevant knowledge text corresponding to the description text; The description text, the relevant knowledge text and the candidate interception reasons are input into the reason judgment model to obtain the reason judgment result output by the reason judgment model.

13. The method according to claim 11, wherein: The method further comprises: According to the judgment results of each of the reasons stated, determine the reasonableness rate of the interception reason; When the reasonable rate of the interception reason is less than or equal to the reasonable rate threshold, the risk warning text is adjusted.

14. The method according to claim 7 or 8, wherein: The sample interception requirement text indicates determining whether there is a first risk element in the multimedia content; The obtaining of training data comprises: Acquire the sample multimedia content and a description text of the sample multimedia content; Performing entity extraction processing on the description text to obtain candidate entities in the description text; Determine, according to the candidate entity in the description text and the interception requirement template text with the entity as a variable, a sample interception requirement text and a sample interception result text corresponding to the sample multimedia content; wherein the interception requirement template text is used to indicate whether the entity exists in the sample multimedia content; The training data is determined according to the sample multimedia content, the sample interception requirement text corresponding to the sample multimedia content, and the sample interception result text.

15. The method according to claim 14, wherein: The determining, according to the candidate entity in the description text and the interception requirement template text with the entity as a variable, the sample interception requirement text and the sample interception result text corresponding to the sample multimedia content includes: Filling the candidate entity into the interception requirement template text to obtain the sample interception requirement text; determining that the sample interception result text indicates that the candidate entity exists in the sample multimedia content; and / or, Fill the interception requirement template text with non-candidate entities other than the candidate entity to obtain the sample interception requirement text; and determine that the sample interception result text indicates that the non-candidate entity does not exist in the sample multimedia content.

16. A content interception device, the device comprising: A first acquisition module, used to acquire multimedia content to be processed and intercept demand text; A second acquisition module, configured to input the multimedia content and the interception requirement text into a content interception model, and obtain an interception result text output by the content interception model; The interception module is used to perform interception control processing on the multimedia content according to the interception result text.

17. A training device for a content interception model, the device comprising: A first acquisition module, used to acquire training data; The training data includes: sample multimedia content, and sample interception requirement text and sample interception result text corresponding to the sample multimedia content; A second acquisition module is used to acquire an initial content interception model; An input module, used for inputting the sample multimedia content and the sample interception requirement text into the content interception model, and obtaining the predicted interception result text output by the content interception model; The training module is used to train the content interception model according to the predicted interception result text and the sample interception result text to obtain a trained content interception model.

18. The device according to claim 17, wherein: The sample interception requirement text indicates interception of risk elements and provides reasons for interception; the first acquisition module is specifically used to: Acquire the sample multimedia content, sample risk elements in the sample multimedia content, and sample interception requirement text corresponding to the sample multimedia content; Determining a description text of the sample multimedia content; Determining, according to the sample risk element, a risk warning text corresponding to the sample multimedia content; Inputting the description text and the risk warning text into a reason generation model to obtain candidate interception reasons output by the reason generation model; Determine the candidate interception reason and the sample risk element as a sample interception result text corresponding to the sample multimedia content; The training data is determined according to the sample multimedia content, the sample interception requirement text corresponding to the sample multimedia content, and the sample interception result text.

19. The device according to claim 17, wherein: The sample interception requirement text indicates determining whether there is a first risk element in the multimedia content; the first acquisition module is specifically used to: Acquire the sample multimedia content and a description text of the sample multimedia content; Performing entity extraction processing on the description text to obtain candidate entities in the description text; Determine, according to the candidate entities in the description text and the interception requirement template text with the entities as variables, a sample interception requirement text and a sample interception result text corresponding to the sample multimedia content; The interception requirement template text is used to indicate whether the entity exists in the sample multimedia content; The training data is determined according to the sample multimedia content, the sample interception requirement text corresponding to the sample multimedia content, and the sample interception result text.

20. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 6; Alternatively, the method according to any one of claims 7 to 15 is performed.

21. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 6; or, to execute the method according to any one of claims 7 to 15.

22. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method according to any one of claims 1 to 6; or implements the method according to any one of claims 7 to 15.

Images