Content distribution method and device, electronic equipment and computer readable storage medium

Abstract

The embodiment of the invention discloses a content distribution method and device, electronic equipment and a computer readable storage medium. According to the embodiment of the invention, after at least one target content is acquired and at least one time keyword is extracted from the target content, at least one first candidate distribution timeliness corresponding to the target content is identified from the time keyword, and the target content is subjected to timeliness classification to obtain at least one second candidate distribution timeliness; screening out a target event matched with the target content from a preset event set, performing event detection on the target event, determining a target distribution time limit of the target content based on a detection result, the first candidate distribution time limit and the second candidate distribution time limit, and distributing the target content according to the target distribution time limit; according to the scheme, the content distribution accuracy can be improved; the embodiment of the invention can be applied to various scenes such as cloud technology, artificial intelligence, intelligent traffic, auxiliary driving and the like.

Description

Technical Field

[0001] The present invention relates to the field of content distribution, and in particular to a content distribution method, apparatus, electronic device, and computer-readable storage medium. Background Art

[0002] In recent years, with the rapid development of Internet technology, a huge amount of content has been generated on the Internet. For this huge amount of content to reach the client, it is necessary to distribute this content. In order to ensure the timeliness of content distribution, current content distribution methods often distribute based on timestamps, timeliness classifications, or rule-based methods in the content.

[0003] In the process of researching and practicing the current technology, the inventors of the present application found that there are often no timestamps in some content, resulting in a very low coverage rate for distribution based on timestamps. In addition, the methods of distribution based on timeliness classification or rules are relatively single and cannot meet the timeliness requirements in different scenarios. Therefore, the accuracy of content distribution is relatively low. Summary of the Invention

[0004] Embodiments of the present application provide a content distribution method, apparatus, electronic device, and computer-readable storage medium, which can improve the accuracy of content distribution.

[0005] A content distribution method includes:

[0006] Obtain at least one target content, and extract at least one time keyword from the target content;

[0007] Identify at least one first candidate distribution timeliness corresponding to the target content from the time keywords;

[0008] Perform timeliness classification on the target content to obtain at least one second candidate distribution timeliness;

[0009] Screen out a target event matching the target content from a preset event set, and perform event detection on the target event;

[0010] Based on the detection result, the first candidate distribution timeliness, and the second candidate distribution timeliness, determine the target distribution timeliness of the target content, and distribute the target content according to the target distribution timeliness.

[0011] Correspondingly, an embodiment of the present application provides a content distribution apparatus, including:

[0012] An acquisition unit, configured to obtain at least one target content, and extract at least one time keyword from the target content;

[0013] An identification unit, configured to identify at least one first candidate distribution time limit corresponding to the target content from the time keywords;

[0014] A classification unit, configured to perform time limit classification on the target content to obtain at least one second candidate distribution time limit;

[0015] A matching unit, configured to filter out a target event matching the target content from a preset event set and perform event detection on the target event;

[0016] A distribution unit, configured to determine a target distribution time limit of the target content based on the detection result, the first candidate distribution time limit, and the second candidate distribution time limit, and distribute the target content according to the target distribution time limit.

[0017] In some embodiments, the identification unit may specifically be configured to query for time limit time words and preset time words in the time keywords; and determine at least one first candidate distribution time limit corresponding to the target content based on the query result.

[0018] In some embodiments, the identification unit may specifically be configured to, when there is at least one of the time limit time words, identify a first candidate distribution time limit corresponding to the target content from the time limit time words; and when there is at least one of the preset time words, filter out at least one type of first candidate distribution time limit from a preset first time limit set.

[0019] In some embodiments, the identification unit may specifically be configured to identify time information in the time limit time words and determine the time type of the time information; when the time information is a time limit expiration time, convert the time information into a distribution time limit to obtain a first candidate distribution time limit; when the time information is an event occurrence time, obtain attribute information of the target content, and determine a first candidate distribution time limit corresponding to the target content based on the attribute information and the time information.

[0020] In some embodiments, the identification unit may specifically be configured to query for a preset time limit corresponding to the preset time word in a first high-accuracy time limit set to obtain a first query result; query for a preset time limit corresponding to the preset time word in a first high-calling time limit set to obtain a second query result; and determine a first candidate distribution time limit corresponding to the target content based on the first query result and the second query result.

[0021] In some embodiments, the classification unit may specifically be configured to determine the content type of the target content, and perform time limit classification on the target content based on the content type to obtain at least one dimension of time limit types; and filter out a preset time limit corresponding to the time limit type from a preset second time limit set to obtain a second candidate distribution time limit.

[0022] In some embodiments, the classification unit may specifically be configured to, when the target content is graphic and text content, extract text features of at least one dimension from the target content to obtain a timeliness type corresponding to the text features; when the target content is video content, extract text features and visual features from the target content to obtain at least one dimension of timeliness type of the target content; when the target content is audio content, convert the target content into text content, and extract text features of at least one dimension from the text content to obtain a timeliness type corresponding to the text features.

[0023] In some embodiments, the classification unit may specifically be configured to extract high-precision text features from the target content by using a high-precision classification model, and determine a first timeliness type of at least one granularity based on the high-precision text features, where the high-precision classification model includes a classification model with a classification accuracy rate exceeding a first preset threshold; extract high-recall text features from the target content by using a high-recall classification model, and determine a second timeliness type of at least one granularity based on the high-recall text features, where the high-recall classification model includes a classification model with a classification recall rate exceeding a second preset threshold; and use the first timeliness type and the second timeliness type as the timeliness type corresponding to the text features.

[0024] In some embodiments, the classification unit may specifically be configured to extract high-precision text features and high-precision visual features from the target content by using a high-precision classification model to obtain a high-precision timeliness type of at least one granularity of the target content; extract high-recall text features and high-recall visual features from the target content by using a high-recall classification model to obtain a high-recall timeliness type of at least one granularity of the target content; and use the high-precision timeliness type and the high-recall timeliness type as the timeliness type of the target content.

[0025] In some embodiments, the classification unit may specifically be configured to extract high-precision text features and high-precision visual features from the target content by using a high-precision classification model; fuse the high-precision text features and the high-precision visual features to obtain high-precision content features; and determine a high-precision timeliness type of at least one granularity of the target content based on the high-precision content features.

[0026] In some embodiments, the matching unit may specifically be configured to perform text linking on the events in the preset event set and the target content to obtain at least one linked content; classify the linked content to obtain a matching type between the target content and the events in the preset event set; and filter out target events matching the target content from the preset event set based on the matching type.

[0027] In some embodiments, the distribution unit may be specifically configured to determine the event distribution timeliness of the target content when detecting a change in the target event; respectively determine the current timeliness types of the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness; and based on the current timeliness types, screen out the target distribution timeliness of the target content from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness.

[0028] In some embodiments, the distribution unit may be specifically configured to obtain the target time when the target event changes, and use the target time as the current stop distribution time of the target content; and based on the current stop distribution time, generate the event distribution timeliness of the target content.

[0029] In some embodiments, the distribution unit may be specifically configured to screen out the target distribution timeliness of the target content from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness when the current timeliness type includes at least one high-accuracy timeliness; and use the preset distribution timeliness as the target distribution timeliness of the target content when the current timeliness type does not include the high-accuracy timeliness and the high-calling timeliness.

[0030] In some embodiments, the distribution unit may be specifically configured to screen out at least one current distribution timeliness corresponding to the high-accuracy timeliness from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness; determine the timeliness length of each current distribution timeliness, and compare the timeliness lengths; and based on the comparison result, screen out the target distribution timeliness of the target content from the current distribution timeliness.

[0031] In some embodiments, the distribution unit may be specifically configured to send the target content to an audit server when the current timeliness type includes at least one of the high-calling timeliness, so that the audit server audits the target content.

[0032] In addition, an embodiment of the present application further provides an electronic device, including a processor and a memory, where the memory stores an application program, and the processor is configured to run the application program in the memory to execute the content distribution method provided by the embodiment of the present application.

[0033] In addition, an embodiment of the present application further provides a computer-readable storage medium, where the computer-readable storage medium stores multiple instructions, and the instructions are suitable for being loaded by a processor to execute the steps in any content distribution method provided by the embodiment of the present application.

[0034] In addition, an embodiment of the present application further provides a computer program product, including a computer program or instruction, which when executed by a processor, implements the steps in the content distribution method provided by the embodiment of the present application.

[0035] After obtaining at least one target content and extracting at least one time keyword from the target content, the embodiment of the present application identifies at least one first candidate distribution timeliness corresponding to the target content from the time keywords, classifies the timeliness of the target content to obtain at least one second candidate distribution timeliness, filters out a target event matching the target content from a preset event set, and performs event detection on the target event. Based on the detection result, the first candidate distribution timeliness, and the second candidate distribution timeliness, the target distribution timeliness of the target content is determined, and the target content is distributed according to the target distribution timeliness. Since this solution does not require a timestamp to determine the distribution timeliness of the target content, the coverage of content distribution is relatively wide. In addition, for different distribution scenarios, the distribution timeliness of the target content can be comprehensively judged through multiple dimensions such as time keywords, timeliness classification, and event detection, enabling a comprehensive judgment and identification of the timeliness of the target content. Therefore, the accuracy of content distribution can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative efforts.

[0037] Figure 1 is a schematic diagram of the scenario of the content distribution method provided by the embodiment of the present application;

[0038] Figure 2 is a schematic flowchart of the content distribution method provided by the embodiment of the present application;

[0039] Figure 3 is a schematic structural diagram of the feature extraction network provided by the embodiment of the present application;

[0040] Figure 4 is a schematic structural diagram of the high-precision classification model provided by the embodiment of the present application;

[0041] Figure 5 is a schematic diagram of determining the event distribution timeliness provided by the embodiment of the present application;

[0042] Figure 6 is a schematic overall diagram of distributing an article provided by the embodiment of the present application;

[0043] Figure 7It is another flowchart of the content distribution method provided by the embodiments of the present application;

[0044] Figure 8 It is a schematic structural diagram of the content distribution device provided by the embodiments of the present application;

[0045] Figure 9 It is a schematic structural diagram of the electronic device provided by the embodiments of the present application. Detailed implementation manners

[0046] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0047] The embodiments of the present application provide a content distribution method, device, electronic device, and computer-readable storage medium. Among them, the content distribution device can be integrated in the electronic device, and the electronic device can be a server or a terminal device, etc.

[0048] Among them, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, network acceleration services (Content Delivery Network, CDN), and big data and artificial intelligence platforms. The terminal includes, but is not limited to, mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, aircraft, etc. The terminal and the server can be directly or indirectly connected through wired or wireless communication methods, and the present application does not make any restrictions here. The embodiments of the present invention can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, intelligent transportation, assisted driving, etc.

[0049] For example, referring to Figure 1 , taking the content distribution device integrated in the electronic device as an example, after the electronic device obtains at least one target content and extracts at least one time keyword from the target content, it identifies at least one first candidate distribution timeliness corresponding to the target content in the time keyword, classifies the timeliness of the target content to obtain at least one second candidate distribution timeliness, screens out the target event that matches the target content in the preset event set, and performs event detection on the target event. Based on the detection result, the first candidate distribution timeliness, and the second candidate distribution timeliness, it determines the target distribution timeliness of the target content, and distributes the target content according to the target distribution timeliness, thereby improving the accuracy of content distribution.

[0050] Among them, the content distribution method provided in the embodiments of the present application involves computer vision, speech technology, and natural language processing (NLP) in artificial intelligence. The embodiments of the present application can extract at least one time keyword from the target content, identify at least one first candidate distribution timeliness from the time keywords, classify the timeliness of the target content to obtain at least one second candidate distribution timeliness, and can also screen out the target event that matches the target content from the preset event set and detect the target event, so that the target distribution timeliness of the target content can be determined through multiple dimensions such as time keywords, timeliness classification, and target events, and the content distribution of the target content can be performed based on the target distribution timeliness, thereby improving the accuracy of content distribution.

[0051] Among them, artificial intelligence (AI) is a theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a way similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines, enabling the machines to have the functions of perception, reasoning, and decision-making.

[0052] Artificial intelligence technology is an interdisciplinary subject with a wide range of fields, including both hardware-level technologies and software-level technologies. The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, pre-trained model technology, operation / interaction systems, and mechatronics. Among them, the pre-trained model is also called the large model or the basic model, and can be widely applied to downstream tasks in various major directions of artificial intelligence after fine-tuning. The software technologies of artificial intelligence mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0053] Among them, computer vision technology (CV) is a science that studies how to enable machines to "see". Further speaking, it refers to using cameras and computers to replace the human eye for tasks such as object recognition, detection, and measurement in machine vision, and further performing graphic processing to make the computer-processed images more suitable for human eye observation or transmission to instrument detection. As a scientific discipline, computer vision researches related theories and technologies, attempting to establish artificial intelligence systems that can obtain information from images or multi-dimensional data. Large model technology has brought important changes to the development of computer vision technology. Pretrained models in the visual field such as swin-transformer, ViT, V-MOE, and MAE can be quickly and widely applied to downstream specific tasks after fine-tuning. Computer vision technology usually includes technologies such as image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, 3D object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping, etc., and also includes common biometric recognition technologies such as face recognition and fingerprint recognition.

[0054] Among them, the key technologies of speech technology include automatic speech recognition technology (ASR), text-to-speech technology (TTS), and voiceprint recognition technology. Enabling computers to listen, see, speak, and feel is the future development direction of human-computer interaction, and among them, speech has become one of the most promising human-computer interaction methods. Large model technology has brought changes to the development of speech technology. Pretrained models that follow the Transformer architecture such as WavLM and UniSpeech have strong generalization and versatility and can excellently complete speech processing tasks in various directions.

[0055] Among them, natural language processing (NLP) is an important direction in the field of computer science and artificial intelligence. It studies various theories and methods that can achieve effective communication between humans and computers using natural language. Natural language processing involves natural language, that is, the language people use in daily life, and is closely related to linguistics research; at the same time, it involves computer science and mathematics... The important technology for model training in the field of artificial intelligence, the pretrained model, is developed from the large language model in the NLP field. After fine-tuning, the large language model can be widely applied to downstream tasks. Natural language processing technology usually includes technologies such as text processing, semantic understanding, machine translation, robot question answering, and knowledge graphs.

[0056] It can be understood that in the specific implementation of the present application, when it comes to relevant data such as the target content of an object, when the following embodiments of the present application are applied to specific products or technologies, permission or consent needs to be obtained, and the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions.

[0057] The following will be described in detail respectively. It should be noted that the description order of the following embodiments does not limit the preferred order of the embodiments.

[0058] This embodiment will be described from the perspective of a content distribution device. The content distribution device can be specifically integrated in an electronic device, which can be a server or a terminal device, etc. Among them, the terminal can include a tablet computer, a notebook computer, a personal computer (PC), a wearable device, a virtual reality device, or other intelligent devices that can perform content distribution.

[0059] A content distribution method includes:

[0060] Obtain at least one target content, extract at least one time keyword from the target content, identify at least one first candidate distribution timeliness corresponding to the target content in the time keyword, classify the timeliness of the target content to obtain at least one second candidate distribution timeliness, screen out a target event that matches the target content in a preset event set, and perform event detection on the target event. Based on the detection result, the first candidate distribution timeliness, and the second candidate distribution timeliness, determine the target distribution timeliness of the target content, and distribute the target content according to the target distribution timeliness.

[0061] As Figure 2 shown, the specific process of this content distribution method is as follows:

[0062] 101. Obtain at least one target content and extract at least one time keyword from the target content.

[0063] Among them, the target content can be understood as the content to be distributed, and the distribution status of the target content can include being distributed or to be distributed.

[0064] Among them, the time keyword can be understood as a text word related to time. The time keyword can include a preset time word and can also include a timeliness time word. The so-called timeliness time word can be understood as a text word related to or associated with timeliness.

[0065] Among them, there are various ways to obtain at least one target content. Specifically, it can be as follows:

[0066] For example, at least one content to be distributed uploaded by a terminal or a client can be received as the target content. Alternatively, at least one content to be distributed or already distributed can be obtained from a content platform or a content server as the target content. Or, at least one target content can be filtered out from a network or a content database. Or, a content distribution request carrying the storage address of at least one target content can also be received, and based on this storage address, at least one target content can be obtained, and so on.

[0067] After obtaining at least one target content, at least one time keyword can be extracted from the target content. There are various ways to extract at least one time keyword from the target content. For example, the content type of the target content can be determined, and based on the content type, at least one time keyword can be extracted from the target content.

[0068] Among them, there are various types of content types. For example, it can include text content, image content, animation content, audio content, or video content, and so on. Based on the content type, there are various ways to extract at least one time keyword from the target content. For example, when the target content is text content, a text recognition model is used to extract at least one time keyword from the text content. When the target content is other content other than text content, the target content is converted into text content, and a text recognition model is used to extract at least one time keyword from the text content, and so on.

[0069] Among them, the text content can include the title of the target content, the body text, or the content after converting other formats of content into text, and so on.

[0070] Among them, there are various ways to extract at least one time keyword from the text content using a text recognition model. For example, a text recognition model can be used to extract text features from the text content, and based on the text features, at least one text word of the text content can be determined, and at least one time keyword can be filtered out from the text words. Or, a text recognition model can also be used to extract time keyword features from the text content, and based on the time keyword features, at least one time keyword can be determined, and so on.

[0071] Among them, the network structure of the text recognition model can be various. For example, it can include BERT (a text processing model), CNN (convolutional neural network), or other models that can perform time keyword recognition, and so on.

[0072] 102. Identify at least one first candidate distribution time limit corresponding to the target content from the time keywords.

[0073] Among them, the first candidate distribution timeliness can be understood as the distribution timeliness corresponding to the time keyword. The so-called distribution timeliness can be understood as the time interval or range corresponding to the start time and the exit time (stop distribution time) of the target content during the distribution process. It should be noted that for a target that has already started distribution, the distribution timeliness may not include the start time, that is, it only includes the exit time.

[0074] Among them, there are various ways to identify at least one first candidate distribution timeliness corresponding to the target content in the time keyword, specifically as follows:

[0075] For example, the timeliness time word and the preset time word can be queried in the time keyword, and based on the query result, at least one first candidate distribution timeliness corresponding to the target content is determined.

[0076] Among them, the preset time word can be a preset time keyword. For example, it can include today, tomorrow, next week, upcoming, or other preset time keywords.

[0077] Among them, the timeliness time word can include time words related or associated with timeliness. There are various types of timeliness time words. For example, it can include time keywords of ordinary non-timeliness time, time keywords of event occurrence time, and time keywords of timeliness expiration time. The so-called ordinary non-timeliness time can be understood as that the time keyword has no timeliness meaning in the text content. The event occurrence time can be understood as the time when the event described in the text occurs, which is often used in fields such as news reports and historical stories. The timeliness expiration time can be understood as the time when the event described in the text expires, which is usually used in scenarios such as commodity promotions and flash sales.

[0078] Among them, based on the query result, there are various ways to determine at least one first candidate distribution timeliness corresponding to the target content. For example, when there is at least one timeliness time word, the first candidate distribution timeliness corresponding to the target content is identified in the timeliness time word. When there is at least one preset time word, at least one type of first candidate distribution timeliness is screened out from the preset first timeliness set. Specifically as follows:

[0079] (1) When there is at least one timeliness time word, the first candidate distribution timeliness corresponding to the target content is identified in the timeliness time word.

[0080] For example, when there is at least one timeliness time word, the time information can be identified in the timeliness time word, and the time type of the time information is determined. When the time information is the timeliness expiration time, the time information is converted into the distribution timeliness to obtain the first candidate distribution timeliness. When the time information is the event occurrence time, the attribute information of the target content is obtained, and based on the attribute information and the time information, the first candidate distribution timeliness corresponding to the target content is determined.

[0081] Among them, the time information can be the time-related information included in the time-of-effect words. For example, taking the time keyword with the time-of-effect word as the incident time as an example, at this time, the time information of the time-of-effect word can be the information of the incident time, and the time type of the time information can be the incident time. The time-of-effect words can include time keywords of ordinary non-time-of-effect, time keywords of incident time, and time keywords of expired time-of-effect. Therefore, the time information of the time-of-effect words can include information of ordinary non-time-of-effect, information of incident time, and information of expired time-of-effect, and the corresponding time types can include ordinary non-time-of-effect time, incident time, and expired time-of-effect, and so on.

[0082] Among them, when the time information is the expired time-of-effect, the time information can be converted into a distribution time-of-effect to obtain a first candidate distribution time-of-effect. There can be multiple ways to convert the time information into a distribution time-of-effect. For example, the expired time-of-effect can be identified in the time information, and the expired time-of-effect can be used as the stop distribution time. When the target content is the content that has been distributed, the start distribution time of the target content is obtained, and based on the start distribution time and the stop distribution time, the first candidate distribution time-of-effect of the target content is determined. When the target content is the content to be distributed, the current moment or the preset distribution moment can be used as the start distribution time, and based on the start distribution time and the stop distribution time, the first candidate distribution time-of-effect of the target content is determined.

[0083] Among them, when the time information is the incident time, the attribute information of the target content can be obtained, and based on the attribute information and the time information, the first candidate distribution time-of-effect corresponding to the target content is determined. The attribute information of the target content can include information about the release object of the target content, release time, release platform, content label, or content type, and so on. There can be multiple ways to determine the first candidate distribution time-of-effect corresponding to the target content based on the attribute information and the time information. For example, based on the attribute information and the time information, the first candidate distribution time-of-effect corresponding to the target content can be screened out from the preset time-of-effect set, or alternatively, based on the attribute information and the time information, the content time-of-effect type of the target content can be determined, and the time-of-effect corresponding to the content time-of-effect type can be screened out from the preset time-of-effect set to obtain the first candidate distribution time-of-effect corresponding to the target content, and so on.

[0084] Optionally, in some embodiments, when the time information is ordinary non-time-of-effect time, it can be determined that there is no first candidate distribution time-of-effect in the time-of-effect word.

[0085] (2) When there is at least one preset time word, at least one type of first candidate distribution time-of-effect is screened out from the preset first time-of-effect set.

[0086] The preset first time validity set may include a first high accuracy time validity set and a first high recall time validity set. The so-called first high accuracy time validity set may include at least one preset time validity whose accuracy exceeds a first preset accuracy threshold, and the first high recall time validity set may include at least one preset time validity whose recall rate exceeds a first preset recall rate threshold.

[0087] There are multiple ways to select at least one type of first candidate distribution timeliness from the preset first timeliness set, which may be as follows:

[0088] For example, you can query the preset time limit corresponding to the preset time word in the first high-precision time limit set to obtain a first query result, and query the preset time limit corresponding to the preset time word in the first high-call time limit set to obtain a second query result. Based on the first query result and the second query result, determine the first candidate distribution time limit corresponding to the target content.

[0089] Among them, based on the first query result and the second query result, there can be multiple ways to determine the first candidate distribution timeliness corresponding to the target content. For example, when the first query result indicates that there is at least one preset timeliness corresponding to a preset time word in the first high-precision timeliness set, the preset timeliness corresponding to the preset time word can be used as the first candidate distribution timeliness. When the second query result indicates that there is at least one preset timeliness corresponding to a preset time word in the first high-call timeliness set, the preset timeliness corresponding to the preset time word can be used as the first candidate distribution timeliness, and so on.

[0090] Among them, determining the first candidate distribution timeliness based on the preset time word can be regarded as determining the distribution timeliness based on the rule, and at least one first candidate distribution timeliness can be determined by the rule engine module. The so-called rule engine module is designed to solve the problem of preset time words (fixed time words and specific keywords) in the target content. Fixed time words can include today, tomorrow or next week, etc., and specific keywords appear in the form of "soon..." etc. Different preset time words are matched with different timeliness according to experience, and these preset time words can be manually pinned. When these preset time words appear in the target content, the rule engine module can determine or output the corresponding timeliness length (distribution timeliness), and can also perform an exit operation on the target content based on the distribution timeliness to determine the timeliness and accuracy of the target content. The rule engine module can determine or output the corresponding timeliness length for content such as news, thereby ensuring the timeliness and accuracy of the news.

[0091] 103. Classify the target content by timeliness to obtain at least one second candidate distribution timeliness.

[0092] For example, the content type of the target content can be determined, and based on the content type, the target content can be classified by timeliness to obtain timeliness types in at least one dimension. The preset timeliness corresponding to the timeliness type can be filtered out from the preset second timeliness set to obtain the second candidate distribution timeliness.

[0093] Among them, there can be multiple content types. For example, it can include graphic content, video content, audio content, and so on. There can be multiple ways to classify the target content by timeliness based on the content type. For example, when the target content is graphic content, at least one dimension of text features is extracted from the target content to obtain at least one dimension of timeliness type of the target content. When the target content is video content, text features and visual features are extracted from the target content to obtain at least one dimension of timeliness type of the target content. When the target content is audio content, the target content is converted into text content, and at least one dimension of text features is extracted from the text content to obtain the timeliness type corresponding to the text features. Specifically, it can be as follows:

[0094] (1) When the target content is graphic content, at least one dimension of text features is extracted from the target content to obtain at least one dimension of timeliness type of the target content.

[0095] Among them, graphic content can be understood as content containing images and / or text, that is, the image content can include image content or text content. When the target content is graphic content, the distribution of the target content is in the form of graphics and text.

[0096] For example, a high-precision classification model can be used to extract high-precision text features from the target content, and based on the high-precision text features, at least one granularity of the first timeliness type is determined. A high-recall classification model can be used to extract high-recall text features from the target content, and based on the high-recall text features, at least one granularity of the second timeliness type is determined. The first timeliness type and the second timeliness type are used as the timeliness types corresponding to the text features.

[0097] Among them, the high-precision classification model can be understood as a classification model with a classification accuracy exceeding the first preset threshold, and the high-recall classification model can be understood as a classification model with a classification recall rate exceeding the second preset threshold. The network structures of the high-precision classification model and the high-recall classification model can be the same or different. The high-precision classification model can include a feature extraction network and a timeliness classification network. There can be multiple ways to extract high-precision text features from the target content using the high-precision classification model. For example, the feature extraction network of the high-precision classification model can be used to extract features from the target content to obtain the text features of the target content.

[0098] Among them, the network structure of the feature extraction network can be of various types. For example, the BERT network can be used as the backbone of the feature extraction network, or it can also include longformer (a type of feature extraction network) or Roberta (a type of feature extraction network), etc. Taking the BERT network as the feature extraction network as an example, the network structure of the feature extraction network can be as Figure 3 shown, and the classification feature (cls token) output by BERT is used as the text feature of the target content. The BERT network is a deep learning model that can process natural language text. Its input consists of word vectors and can encode and understand the text content, thus providing strong support for the classification of graphic and text content (graphic and text media).

[0099] After extracting the high-precision text features, at least one granularity of the first timeliness type can be determined based on the high-precision text features. There are various ways to determine at least one granularity of the first timeliness type. For example, the high-precision text features can be classified by a timeliness classification network to obtain at least one granularity of the first timeliness type.

[0100] Among them, the network structure of the timeliness classification network can be of various types. For example, it can include a fully connected layer (FC), or other networks that can perform timeliness classification, etc.

[0101] Among them, the granularity of timeliness classification can include coarse-grained and fine-grained. The timeliness types of different granularities correspond to different distribution timeliness. There are various coarse-grained timeliness types. For example, it can include pre-market, during market, after market, pre-game, during game, after game, other game-related, pre-holiday, during holiday, after holiday or other holiday-related, etc. There are various fine-grained timeliness types. For example, it can include weakly related to the event periphery, event interpretation, official release, regular consultation content, flash news, digital bulletin, morning report, evening report, breaking news, heavyweight news or preview content, etc. The distribution timeliness corresponding to these different granularity timeliness types can be the same or different.

[0102] Among them, the method of using the high-precision classification model to extract high-precision text features from the target content and determining at least one granularity of the second timeliness type based on the high-precision text features can be similar to the method of determining at least one first timeliness type. For details, see the above description and will not be elaborated here one by one.

[0103] (2) When the target content is video content, text features and visual features are extracted from the target content to obtain at least one dimension of the timeliness type of the target content.

[0104] For example, when the target content is video content, a high-accuracy classification model can be used to extract high-accuracy text features and high-accuracy visual features from the target content to obtain at least one granularity of high-accuracy timeliness types of the target content. A high-precision classification model can be used to extract high-precision text features and high-precision visual features from the target content to obtain at least one granularity of high-precision timeliness types of the target content. The high-accuracy timeliness type and the high-precision timeliness type are used as the timeliness types of the target content.

[0105] Among them, there are various ways to use a high-accuracy classification model to extract high-accuracy text features and high-accuracy visual features from the target content to obtain at least one granularity of high-accuracy timeliness types. For example, a high-accuracy classification model can be used to extract high-accuracy text features and high-accuracy visual features from the target content, and the high-accuracy text features and high-accuracy visual features can be fused to obtain high-accuracy content features. Based on the high-accuracy content features, at least one granularity of high-accuracy timeliness types is determined.

[0106] Among them, for the case where the target content is text content, the feature extraction network of the high-accuracy classification model can use a two-stream network as the backbone. At this time, the feature extraction network can include a text feature extraction network and a visual feature extraction network. There are various ways to use a high-accuracy classification model to extract high-accuracy text features and high-accuracy visual features from the target content. For example, the target content can be framed to obtain at least one video frame, the video frame can be text-recognized to obtain video text content (video ASR), the content title of the target content can be obtained, the text feature extraction network can be used to extract features from the content title and the video text content to obtain high-accuracy text features, and the visual feature extraction network can be used to extract features from the target content or the video frame to obtain high-accuracy visual features, and so on.

[0107] Among them, the network structure of the text feature extraction network can be various. For example, it can include BERT or other networks that can extract text features, and so on. The network structure of the visual feature extraction network can be various. For example, it can include Swin3D Tiny (a visual feature extraction network) or other networks that can extract visual features, and so on.

[0108] After using a high-accuracy classification model to extract high-accuracy text features and high-accuracy visual features from the target content, the high-accuracy text features and high-accuracy visual features can be fused to obtain high-accuracy content features. There are various ways to fuse the high-accuracy text features and high-accuracy visual features. For example, the encoder (Encode) of a two-layer Transformer (an attention model) can be used to fuse the high-accuracy text features and high-accuracy visual features to obtain a richer and more accurate feature representation, and this feature representation is used as the high-accuracy content features.

[0109] After fusing the high-precision text features and high-precision visual features, at least one granularity of high-precision timeliness types of the target content can be determined based on the fused high-precision content features. There are various ways to determine at least one granularity of high-precision timeliness types of the target content based on the fused high-precision content features. For example, a fully connected layer (FC) can be used to classify the high-precision content features to obtain at least one granularity of high-precision timeliness types of the target content. Or, other classification networks can also be used to classify the high-precision content features to obtain at least one granularity of high-precision timeliness types of the target content, and so on.

[0110] Among them, taking the text feature extraction network as BERT and the visual feature extraction network as Swin Transformer (a kind of feature extraction network) as an example, the network structure of the high-precision classification model can be as Figure 4 shown. Use Swin Transformer to extract visual features in the video content of the target content as high-precision visual features, use BERT to extract text features in the text content (title and video ASR) of the target content as high-precision text features, and then use Cross-attention (cross-attention network) to fuse the high-precision text features and high-precision visual features to obtain high-precision content features. Use a fully connected layer (FC) to classify the high-precision content features for visual types to obtain at least one granularity of high-precision timeliness types. During the training process of the high-precision classification model, the high-precision timeliness types output by the high-precision classification model can be compared with the timeliness type labels to obtain the classification loss, and the high-precision classification model can be converged based on the classification loss to obtain the trained high-precision classification model.

[0111] Among them, the method of extracting high-precision text features and high-precision visual features in the target content using the high-precision classification model can be similar to the method of extracting high-precision text features and high-precision visual features in the target content using the high-precision classification model. For details, see the above description and will not be elaborated here one by one.

[0112] (3) When the target content is audio content, convert the target content into text content and extract at least one dimension of text features from the text content to obtain the timeliness type corresponding to the text features.

[0113] Among them, there are various ways to convert the target content into text content, specifically as follows:

[0114] For example, at least one audio frame can be extracted from the audio content, and text recognition is performed on the audio frame to obtain the text content corresponding to the target content.

[0115] After converting the target content into text content, at least one dimension of text features can be extracted from the text content, so as to obtain the timeliness type corresponding to the text features. The method of extracting at least one dimension of text features from the text content can be similar to the method of extracting at least one dimension of text features from the target content. For details, please refer to the above description and will not be elaborated here one by one.

[0116] 104. Screen out target events that match the target content from the preset event set and perform event detection on the target events.

[0117] Among them, an event can be understood as a certain thing that happens in real life. The event can include an event object. Therefore, the content of the event can also be understood as what happens to the event object. For example, taking the event object as a certain star, the events corresponding to the event object can include the departure of the star, the arrival of the star, or things related to the star, and so on.

[0118] Among them, there are various ways to screen out target events that match the target content from the preset event set. Specifically, it can be as follows:

[0119] For example, link the events in the preset event set with the target content to obtain at least one link content, classify the link content, obtain the matching type between the target content and the events in the preset event set, and based on the matching type, screen out the target events that match the target content from the preset event set.

[0120] Among them, there are various ways to link the events in the preset event set with the target content. For example, the events in the preset event set can be text-linked with the target content through a separator to obtain at least one link content, or other connection methods can also be used to text-link the events in the preset event set with the target content to obtain at least one link content, and so on.

[0121] Among them, there are various types of separators. For example, it can include [sep] or other symbols that can be used for connection, and so on.

[0122] After text-linking the events in the preset event set with the target content, the link content of the link can be classified to obtain the matching type between the target content and the events in the preset event type set. There are various ways to classify the link content. For example, a classification model can be used to classify the link content, and based on the classification result, the matching type between the target content and the events in the preset event set can be determined.

[0123] Among them, the classification model may include the BERT model or other models that can classify the matching types of link content, and so on. Taking the BERT model as an example of the classification model, the link content can be input into the BERT model and classified through the CLS token. In this classification process, the classification result 0 can indicate non - matching, and 1 can indicate matching, so as to determine the matching type between each event in the preset event set and the target content.

[0124] After classifying the link content, based on the determined matching type after classification, the target events that match the target content can be filtered out from the preset event set. There are various ways to filter out the target events that match the target content from the preset event set. For example, the events with a matching type of "matching" can be filtered out from the preset event set to obtain the target events.

[0125] After filtering out the target events that match the target content from the preset event set, event detection can be performed on the target events. There are various ways to perform event detection on the target events. For example, the event of the target event can be detected to obtain the detection result of the target event. The detection result here can include that the target event has changed or has not changed.

[0126] Among them, it should be noted that the detection of the target event can be real - time or continuous detection. The key point of the detection is whether the event of the target event has changed.

[0127] 105. Based on the detection result, the first post - candidate distribution timeliness and the second candidate distribution timeliness, determine the target distribution timeliness of the target content, and distribute the target content according to the target distribution timeliness.

[0128] Among them, there are various ways to determine the target distribution timeliness of the target content based on the detection result, the first candidate distribution timeliness and the second candidate distribution timeliness. Specifically, it can be as follows:

[0129] For example, when it is detected that the target event has changed, determine the event distribution timeliness of the target content, respectively determine the current timeliness types of the event distribution timeliness, the first candidate distribution timeliness and the second candidate distribution timeliness, and based on the current timeliness types, filter out the target distribution timeliness of the target content from the event distribution timeliness, the first candidate distribution timeliness and the second candidate distribution timeliness.

[0130] Among them, the event distribution timeliness can be understood as the distribution timeliness determined based on the target event. When it is detected that the target event changes, there are various ways to determine the event distribution timeliness of the target content. For example, when it is detected that the event event of the target event changes, it is determined that the target event has changed, and the target time when the target event changes is obtained, and the target time is used as the current stop distribution time of the target content. Based on the current stop distribution time, the event distribution timeliness of the target content is generated.

[0131] Among them, the stop distribution time can be understood as the time to stop distributing the target content, that is, the exit time of the target content. There are various ways to generate the event distribution timeliness of the target content based on the current stop distribution time. For example, the start distribution time of the target content can be obtained, and the timeliness composed of the start distribution time and the stop distribution time is used as the event distribution timeliness, and so on.

[0132] Among them, it should be noted that for the event distribution timeliness, the key lies in the change of the target event. Once the event event of the target event changes, the target content matching the previous event will be eliminated, that is, the change time is used as the exit time (stop distribution time) of the target content. This process aims to determine that the matching can be adjusted flexibly to adapt to the change of the event. Taking the target event as the departure of a certain star as an example, the process of determining the event distribution timeliness based on the event change can be as Figure 5 described. The target event at time t1 is "a certain star departs". At this time, the target content matching the target event is "a certain star takes the bullet train to XXX City" or "a certain star will arrive in XXX City on XX month XX day". When the event event of the target event changes at time t2 and becomes "a certain star arrives", at this time, the target content matching the previous target event can stop distributing and exit. Therefore, time t2 can be the stop distribution timeliness of the target content, and the target content matching the target event "a certain star arrives in XX City" or "the fans are very excited to see a certain star" can continue to be distributed until the target event changes, and the change time of the target event can be used as the stop distribution time of the above target content, and then the event distribution timeliness is determined, and so on, so as to be able to respond to the changing situation in a timely and effective manner, so as to maintain the efficiency and practicality of the distribution of the target content.

[0133] After determining the event distribution timeliness of the target content, the current timeliness types of the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness can be determined respectively. The current timeliness type can include one of high-accuracy timeliness, high-recall timeliness, or other timeliness. It should be noted that the first candidate distribution timeliness determined based on the timeliness time word can include high-accuracy timeliness or other timeliness, the first candidate distribution timeliness determined based on the preset keyword can include high-accuracy timeliness, high-recall timeliness, or other timeliness, the event distribution timeliness determined based on the target event can include high-accuracy timeliness or other timeliness, the second candidate distribution timeliness determined by classifying the timeliness of the target content can include high-accuracy timeliness with different granularities, high-recall timeliness with different granularities, or other timeliness, and so on. High-accuracy timeliness can include a distribution timeliness with an accuracy exceeding the preset accuracy threshold, and high-recall timeliness can include a distribution timeliness with a recall rate exceeding the preset recall rate threshold.

[0134] After determining the current timeliness types of the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness, the target distribution timeliness of the target content can be screened out from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness based on the current timeliness type. There can be multiple ways to screen out the target distribution timeliness of the target content from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness. For example, when the current timeliness type includes at least one high-accuracy timeliness type, the target distribution timeliness of the target content is screened out from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness. When the current timeliness type does not include high-accuracy timeliness and high-recall timeliness, the preset distribution timeliness is used as the target distribution timeliness of the target content.

[0135] Among them, when the current timeliness type includes at least one high-accuracy timeliness, the target distribution timeliness of the target content can be screened out from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness. There can be multiple ways to screen out the target distribution timeliness of the target content from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness. For example, at least one current distribution timeliness corresponding to the high-accuracy timeliness is screened out from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness, the timeliness length of each current distribution timeliness is determined, and the timeliness lengths are compared. Based on the comparison result, the target distribution timeliness of the target content is screened out from the current distribution timeliness.

[0136] Among them, the aging length can be understood as the time length between the start time and the end time in the distribution aging. For example, if the start time is 8 am and the end time is 8 pm, the aging length at this time can be 12 hours. After determining the aging length of each current distribution aging, the aging lengths can be compared. Then, based on the comparison result, the target distribution aging of the target content can be screened out from the current distribution aging. There can be various ways to screen out the target distribution aging of the target content from the current distribution aging based on the comparison result. For example, the current distribution aging corresponding to the shortest aging length can be screened out from the current distribution aging to obtain the target distribution aging of the target content.

[0137] Optionally, in some embodiments, when it is not detected that the target event has changed, the target distribution aging of the target content can be screened out from the first candidate distribution aging and the second distribution aging. The method of screening out the target distribution aging of the target content from the first candidate distribution aging and the second distribution aging can be similar to the method of screening out the target distribution aging of the target content from the event distribution aging, the first candidate distribution aging, and the second candidate distribution aging, as described in detail above, and will not be elaborated here one by one.

[0138] Optionally, in some embodiments, when the current aging type includes at least one high-recall aging, the target content is sent to the audit server so that the audit server can audit the target content. When the current aging type includes at least one high-recall aging, it can indicate that the recall rate of the target content is relatively high. At this time, the target content needs to be audited again. The audit method can be through the audit server or manually on the audit server side. After the audit passes, content distribution can continue. When the audit fails, the target content will not be distributed. At this time, it can be returned to the producer of the target content for modification or adjustment.

[0139] After determining the target distribution aging of the target content, the target content can be distributed based on the target distribution aging. There can be various ways to distribute the target content. For example, when the target content is already distributed content, the target content is continuously distributed. When the target content is to-be-distributed content, the start distribution time of the target content is obtained in the target distribution aging, and the target content is distributed based on the start distribution time. Then, the stop distribution time is extracted from the target distribution aging. When the stop distribution time is reached, the content distribution of the target content is stopped. At this time, the target content can exit.

[0140] Among them, with the continuous development of the Internet industry, the Internet has become the main channel for people to obtain information and for media dissemination. Subsequently, more and more articles in the form of pictures, texts and videos have emerged. However, how to ensure the timeliness and freshness of the articles recommended within the platform has become increasingly crucial. Different types of articles have different timeliness sensitivities, and it is necessary to judge according to the content and classification of the articles. In this solution, taking the target content as an article as an example, the overall distribution process for the target content can be as Figure 6 shown. The distribution timeliness of the article can be determined through four different modules, and the target distribution timeliness can be screened out among different distribution timeliness for content distribution. Specifically, it can be as follows:

[0141] Given an article, we input the article into the items module (the timeliness time words determine the first candidate distribution timeliness), the classification module (the timeliness classification determines the second candidate distribution timeliness), the rule engine module (the preset event words determine the first candidate distribution timeliness), and the event module (the target event determines the event distribution timeliness) respectively. Then the items module produces the items high-precision timeliness T1, the classification module produces the high-precision fine-grained timeliness T2, T3 and the high-recall fine-grained timeliness T4, T5, the rule engine module produces the keyword high-precision timeliness T6, and the keyword high-recall timeliness T7. The event module produces the event high-precision timeliness T8. If at least one high-precision timeliness is generated by the above modules, then the minimum value strategy is adopted to take the short value of the above timeliness. The process of taking the short value can be as shown in formula (1). Specifically, it can be as follows:

[0142] T final = min(T1, T2, T3, T6, T8) (1)

[0143] Among them, T final is the target distribution timeliness, T1 is the items high-precision timeliness, T2 and T3 are the high-precision fine-grained timeliness determined by the classification module respectively, T6 is the keyword high-precision timeliness, and T8 is the event high-precision timeliness.

[0144] If no high-precision timeliness is generated, it is necessary to further determine whether at least one high-recall timeliness is generated. If a high-recall timeliness is generated, it will enter the review process for withdrawal processing. If neither high-precision timeliness nor high-recall timeliness is generated, the preset distribution timeliness (fallback timeliness) can be used as the target distribution timeliness of the article. Then, based on the target distribution timeliness, the article is distributed until the stop distribution time. Then, the distribution of the article is stopped, and the article is subject to withdrawal processing.

[0145] Among them, in this solution, multiple modules can work together to jointly cover all articles (target content) in various scenarios, provide a distributable time limit, exit at the specified time, and greatly save the review cost and review resources. In the news scenario, for the short-time effectiveness judgment and recognition of videos and graphics and texts, the performance: the accuracy rate can be about 90%, and the recall rate can be about 80%.

[0146] As can be seen from the above, in the embodiment of the present application, after obtaining at least one target content and extracting at least one time keyword from the target content, at least one first candidate distribution time limit corresponding to the target content is identified in the time keyword, and the target content is classified by time limit to obtain at least one second candidate distribution time limit. Then, a target event matching the target content is screened out from the preset event set, and event detection is performed on the target event. Based on the detection result, the first candidate distribution time limit and the second candidate distribution time limit, the target distribution time limit of the target content is determined, and the target content is distributed according to the target distribution time limit. Since this solution does not require a time stamp to judge the distribution time limit of the target content, the coverage of content distribution is relatively wide. In addition, for the distribution scenarios in different scenarios, the distribution time limit of the target content can be comprehensively judged through multiple dimensions such as time keywords, time limit classification, and event detection, so as to comprehensively judge and identify the timeliness of the target content. Therefore, the accuracy of content distribution can be improved.

[0147] According to the method described in the above embodiment, the following will give further detailed examples.

[0148] In this embodiment, it will be described by taking the content distribution device being specifically integrated in an electronic device, the electronic device being a server, and the target content being a target article as an example.

[0149] As Figure 7 shown, a content distribution method has the following specific process:

[0150] 201. The server obtains at least one target article.

[0151] For example, the server can receive at least one content to be distributed uploaded by a terminal or a client as the target article, or can obtain at least one content to be distributed or already distributed in a content platform or a content server as the target article, or can screen out at least one target article in a network or a content database, or can also receive a content distribution request, and the content distribution request carries the storage address of at least one target article. Based on this storage address, at least one target article is obtained, and so on.

[0152] 202. The server extracts at least one time keyword from the target article.

[0153] For example, when the content of the target article is text content, the server uses BERT to extract text features from the text content, and based on the text features, determines at least one text word in the text content, and filters out at least one time keyword from the text words. Or, the server can also use BERT to extract time keyword features from the text content, and based on the time keyword features, determine at least one time keyword, and so on.

[0154] When the content of the target article is other content than text content, the server converts the content of the target article into text content, and the server uses BERT to extract at least one time keyword from the text content.

[0155] 203. The server identifies at least one first candidate distribution timeliness corresponding to the target article from the time keywords.

[0156] For example, the server can query the timeliness time word and the preset time word in the time keywords. When there is at least one timeliness time word, identify the time information in the timeliness time word, and determine the time type of the time information. When the time information is the timeliness expiration time, identify the timeliness expiration time in the time information, and use the timeliness expiration time as the stop distribution time. When the target article is a distributed article, obtain the start distribution time of the target article, and based on the start distribution time and the stop distribution time, determine the first candidate distribution timeliness of the target article. When the target article is an article to be distributed, the current time or the preset distribution time can be used as the start distribution time, and based on the start distribution time and the stop distribution time, determine the first candidate distribution timeliness of the target article.

[0157] When the time information is the incident time, obtain the attribute information of the target article. Based on the attribute information and the time information, filter out the first candidate distribution timeliness corresponding to the target article from the preset timeliness set. Or, based on the attribute information and the time information, determine the content timeliness type of the target article, and filter out the timeliness corresponding to the content timeliness type from the preset timeliness set to obtain the first candidate distribution timeliness corresponding to the target article, and so on.

[0158] Optionally, in some embodiments, when the time information is the ordinary non-timeliness time, the server can determine that there is no first candidate distribution timeliness in the timeliness time word.

[0159] When there is at least one preset time word, the server queries for the preset timeliness corresponding to the preset time word in the first high-accuracy timeliness set to obtain a first query result, and queries for the preset timeliness corresponding to the preset time word in the first high-citation timeliness set to obtain a second query result. When the first query result indicates that there is at least one preset timeliness corresponding to the preset time word in the first high-accuracy timeliness set, the preset timeliness corresponding to the preset time word can be used as the first candidate distribution timeliness. When the second query result indicates that there is at least one preset timeliness corresponding to the preset time word in the first high-citation timeliness set, the preset timeliness corresponding to the preset time word can be used as the first candidate distribution timeliness, and so on.

[0160] 204. The server performs timeliness classification on the target article to obtain at least one second candidate distribution timeliness.

[0161] For example, the server determines the content type of the target article. When the content of the target article is graphic content, a high-accuracy classification model is used to extract high-accuracy text features from the target article, and the high-accuracy text features are classified for timeliness through the FC layer, so as to obtain at least one granularity of the first timeliness type. A high-citation classification model is used to extract high-citation text features from the target article, and based on the high-citation text features, at least one granularity of the second timeliness type is determined. The first timeliness type and the second timeliness type are used as the timeliness types corresponding to the text features.

[0162] When the content of the target article is video content, the server can frame the target article to obtain at least one video frame, perform text recognition on the video frame to obtain video text content (video ASR), obtain the article title of the target article, and use the BERT network of the high-accuracy classification model to extract high-accuracy text features from the article title and the video text content. The Swin Transformer network of the high-citation classification model is used to extract high-accuracy visual features from the target article or the video frame. The server uses the encoder of the two-layer Transformer (Encode) to fuse the high-accuracy text features and the high-accuracy visual features, so as to obtain high-accuracy content features, and uses the fully connected layer (FC) to classify the high-accuracy content features, so as to obtain at least one granularity of the high-accuracy timeliness type of the target article. The server can also use the high-citation classification model to extract high-citation text features and high-citation visual features from the target article, fuse the high-citation text features and the high-citation visual features to obtain high-citation content features, and based on the high-citation content features, use the fully connected layer (FC) to classify the high-citation content features, so as to obtain at least one granularity of the high-citation timeliness type.

[0163] When the content of the target article is audio content, the server can also extract at least one audio frame from the audio content, perform text recognition on the audio frame, and obtain the text content corresponding to the target article. At least one dimension of text features is extracted from the text content, so as to obtain the timeliness type corresponding to the text features.

[0164] 205. The server filters out the target event that matches the target article from the preset event set.

[0165] For example, the server can perform text linking on the events in the preset event set and the target article through the separator [sep] to obtain at least one linked content, or other linking methods can also be used to perform text linking on the events in the preset event set and the target article to obtain at least one linked content, and so on.

[0166] The server uses BERT to classify the linked content. Based on the classification result, the matching type between the target article and the events in the preset event set is determined. The events with the matching type of matching are filtered out in the preset event set to obtain the target event.

[0167] 206. The server performs event detection on the target event.

[0168] For example, the server detects the event of the target event to obtain the detection result of the target event.

[0169] 207. The server determines the target distribution timeliness of the target article based on the detection result, the first candidate distribution timeliness, and the second candidate distribution timeliness.

[0170] For example, when it is detected that the event of the target event changes, the server determines that the target event has changed, obtains the target time when the target event changes, and uses the target time as the current stop distribution time of the target article. The start distribution time of the target article is obtained, and the timeliness composed of the start distribution time and the stop distribution time is used as the event distribution timeliness.

[0171] When the current timeliness type includes at least one high-accuracy timeliness type, the server filters out at least one current distribution timeliness corresponding to the high-accuracy timeliness from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness, determines the timeliness length of each current distribution timeliness, compares the timeliness lengths, and filters out the current distribution timeliness corresponding to the shortest timeliness length among the current distribution timeliness to obtain the target distribution timeliness of the target article.

[0172] When no change in the target event is detected, the server can select the target distribution timeliness of the target article from the first candidate distribution timeliness and the second distribution timeliness. The method of selecting the target distribution timeliness of the target article from the first candidate distribution timeliness and the second distribution timeliness can be similar to the method of selecting the target distribution timeliness of the target article from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness. For details, please refer to the above description and will not be elaborated here one by one.

[0173] Optionally, in some embodiments, when the current timeliness type includes at least one high-call timeliness, the server sends the target article to the audit server for the audit server to audit the target article.

[0174] 208. The server distributes the target article according to the target distribution timeliness.

[0175] For example, when the target article is a distributed article, the server continues to distribute the target article. When the target article is a to-be-distributed article, the server obtains the start distribution time of the target article in the target distribution timeliness and distributes the target article based on the start distribution time. Then, the stop distribution time is extracted from the target distribution timeliness. When the stop distribution time is reached, the distribution of the target article is stopped. At this time, the target article can exit.

[0176] As can be seen from the above, in this embodiment, after the server obtains at least one target article and extracts at least one time keyword from the target article, it identifies at least one first candidate distribution timeliness corresponding to the target article in the time keywords, classifies the timeliness of the target article to obtain at least one second candidate distribution timeliness, filters out the target event matching the target article from the preset event set, and performs event detection on the target event. Based on the detection result, the first candidate distribution timeliness, and the second candidate distribution timeliness, it determines the target distribution timeliness of the target article and distributes the target article according to the target distribution timeliness. Since this solution does not require a timestamp to judge the distribution timeliness of the target article, the coverage of content distribution is relatively wide. In addition, for different distribution scenarios, the distribution timeliness of the target article can be comprehensively judged through multiple dimensions such as time keywords, timeliness classification, and event detection, so as to perform an all-round judgment and identification of the timeliness of the target article. Therefore, the accuracy of content distribution can be improved.

[0177] To better implement the above method, an embodiment of the present application further provides a content distribution device, which can be integrated in an electronic device, such as a server or a terminal, etc. The terminal can include a tablet computer, a notebook computer, and / or a personal computer, etc.

[0178] For example, as Figure 8As shown in the figure, the content distribution device may include an acquisition unit 301, an identification unit 302, a classification unit 303, a matching unit 304, and a distribution unit 305, as follows:

[0179] (1) Acquisition unit 301;

[0180] The acquisition unit 301 is configured to acquire at least one target content and extract at least one time keyword from the target content.

[0181] For example, the acquisition unit 301 may specifically be configured to acquire at least one target content, determine the content type of the target content, and when the target content is text content, use a text recognition model to extract at least one time keyword from the text content. When the target content is other content than text content, convert the target content into text content and use a text recognition model to extract at least one time keyword from the text content.

[0182] (2) Identification unit 302;

[0183] The identification unit 302 is configured to identify at least one first candidate distribution timeliness corresponding to the target content from the time keywords.

[0184] For example, the identification unit 302 may specifically be configured to query the timeliness time words and preset time words in the time keywords. When there is at least one timeliness time word, identify the first candidate distribution timeliness corresponding to the target content from the timeliness time words. When there is at least one preset time word, screen out at least one type of first candidate distribution timeliness from the preset first timeliness set.

[0185] (3) Classification unit 303;

[0186] The classification unit 303 is configured to perform timeliness classification on the target content to obtain at least one second candidate distribution timeliness.

[0187] For example, the classification unit 303 may specifically be configured to determine the content type of the target content and, based on the content type, perform timeliness classification on the target content to obtain at least one dimension of timeliness types, and screen out the preset timeliness corresponding to the timeliness types from the preset second timeliness set to obtain the second candidate distribution timeliness.

[0188] (4) Matching unit 304;

[0189] The matching unit 304 is configured to screen out a target event that matches the target content from the preset event set and perform event detection on the target event.

[0190] For example, the matching unit 304 can be specifically used to text-link the events in the preset event set with the target content to obtain at least one linked content, classify the linked content to obtain the matching type between the target content and the events in the preset event set, and based on the matching type, filter out the target events that match the target content from the preset event set, and detect the event of the target event, so as to obtain the detection result of the target event.

[0191] (5) Distribution unit 305;

[0192] The distribution unit 305 is used to determine the target distribution time limit of the target content based on the detection result, the first candidate distribution time limit, and the second candidate distribution time limit, and distribute the target content according to the target distribution time limit.

[0193] For example, the distribution unit 305 can be specifically used to determine the event distribution time limit of the target content when it is detected that the target event changes, respectively determine the current time limit types of the event distribution time limit, the first candidate distribution time limit, and the second candidate distribution time limit, and based on the current time limit types, filter out the target distribution time limit of the target content from the event distribution time limit, the first candidate distribution time limit, and the second candidate distribution time limit, and distribute the target content according to the target distribution time limit.

[0194] In specific implementation, each of the above units can be implemented as an independent entity, or can be combined arbitrarily to be implemented as the same or several entities. For the specific implementation of each of the above units, reference can be made to the foregoing method embodiments, which will not be elaborated herein.

[0195] As can be seen from the above, in this embodiment, after the acquisition unit 301 acquires at least one target content and extracts at least one time keyword from the target content, the recognition unit 302 recognizes at least one first candidate distribution time limit corresponding to the target content from the time keywords, the classification unit 303 performs time limit classification on the target content to obtain at least one second candidate distribution time limit, the matching unit 304 filters out the target events that match the target content from the preset event set and performs event detection on the target events, and the distribution unit 305 determines the target distribution time limit of the target content based on the detection result, the first candidate distribution time limit, and the second candidate distribution time limit, and distributes the target content according to the target distribution time limit; since this solution does not require a timestamp to judge the distribution time limit of the target content, the coverage of content distribution is relatively wide. In addition, for the distribution scenarios in different scenarios, the distribution time limit of the target content can be comprehensively judged through multiple dimensions such as time keywords, time limit classification, and event detection, so as to perform an all-round judgment and recognition of the timeliness of the target content. Therefore, the accuracy of content distribution can be improved.

[0196] The embodiment of the present application also provides an electronic device, such asFigure 9 As shown, it shows a schematic structural diagram of an electronic device involved in an embodiment of the present application. Specifically:

[0197] The electronic device may include a processor 401 with one or more processing cores, a memory 402 with one or more computer-readable storage media, a power supply 403, an input unit 404, and other components. Those skilled in the art can understand that Figure 9 the structural diagram of the electronic device shown in does not constitute a limitation on the electronic device, and it may include more or fewer components than shown, or combine certain components, or have different component arrangements. Among them:

[0198] The processor 401 is the control center of the electronic device, connecting various parts of the entire electronic device through various interfaces and lines. By running or executing software programs and / or modules stored in the memory 402, and by calling the data stored in the memory 402, it executes various functions of the electronic device and processes data. Optionally, the processor 401 may include one or more processing cores; preferably, the processor 401 may integrate an application processor and a modem processor. Among them, the application processor mainly processes the operating system, user interface, application programs, etc., and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor may not be integrated into the processor 401.

[0199] The memory 402 can be used to store software programs and modules. The processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area. Among them, the program storage area can store the operating system, application programs required for at least one function (such as the sound playback function, image playback function, etc.); the data storage area can store the data created according to the use of the electronic device. In addition, the memory 402 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices. Correspondingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.

[0200] The electronic device also includes a power supply 403 that powers each component. Preferably, the power supply 403 can be logically connected to the processor 401 through a power management system, so as to realize functions such as management of charging, discharging, and power consumption management through the power management system. The power supply 403 may also include any components such as one or more DC or AC power supplies, a recharge system, a power failure detection circuit, a power converter or inverter, and a power status indicator.

[0201] The electronic device may further include an input unit 404, which may be configured to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function controls.

[0202] Although not shown, the electronic device may further include a display unit and the like, which will not be elaborated here. Specifically, in this embodiment, the processor 401 in the electronic device will load the executable files corresponding to the processes of one or more application programs into the memory 402 according to the following instructions, and the processor 401 will run the application programs stored in the memory 402 to implement various functions as follows:

[0203] Obtain at least one target content, extract at least one time keyword from the target content, identify at least one first candidate distribution timeliness corresponding to the target content from the time keywords, perform timeliness classification on the target content to obtain at least one second candidate distribution timeliness, screen out the target events matching the target content from the preset event set, and perform event detection on the target events. Based on the detection results, the first candidate distribution timeliness and the second candidate distribution timeliness, determine the target distribution timeliness of the target content, and distribute the target content according to the target distribution timeliness.

[0204] For example, an electronic device may obtain at least one target content, determine the content type of the target content. When the target content is text content, a text recognition model is used to extract at least one time keyword from the text content. When the target content is other content than text content, the target content is converted into text content, and a text recognition model is used to extract at least one time keyword from the text content. An aging time word and a preset time word are queried in the time keywords. When there is at least one aging time word, the first candidate distribution aging corresponding to the target content is identified from the aging time words. When there is at least one preset time word, at least one type of first candidate distribution aging is screened out from the preset first aging set. Determine the content type of the target content, and based on the content type, perform aging classification on the target content to obtain at least one dimension of aging types, and screen out the preset aging corresponding to the aging type from the preset second aging set to obtain the second candidate distribution aging. The events in the preset event set are text-linked with the target content to obtain at least one linked content, the linked content is classified to obtain the matching type between the target content and the events in the preset event set, and based on the matching type, the target event matching the target content is screened out from the preset event set, and the event of the target event is detected to obtain the detection result of the target event. When it is detected that the target event has changed, determine the event distribution aging of the target content, respectively determine the current aging types of the event distribution aging, the first candidate distribution aging and the second candidate distribution aging, and based on the current aging types, screen out the target distribution aging of the target content from the event distribution aging, the first candidate distribution aging and the second candidate distribution aging, and distribute the target content according to the target distribution aging.

[0205] For the specific implementation of each of the above operations, reference may be made to the previous embodiments and will not be elaborated here.

[0206] As can be seen from the above, in the embodiment of the present application, after obtaining at least one target content and extracting at least one time keyword from the target content, at least one first candidate distribution aging corresponding to the target content is identified from the time keywords, the target content is subjected to aging classification to obtain at least one second candidate distribution aging, the target event matching the target content is screened out from the preset event set, and the target event is detected. Based on the detection result, the first candidate distribution aging and the second candidate distribution aging, the target distribution aging of the target content is determined, and the target content is distributed according to the target distribution aging. Since this solution does not require a time stamp to determine the distribution aging of the target content, the coverage of content distribution is relatively wide. In addition, for the distribution scenarios in different scenarios, the distribution aging of the target content can be comprehensively judged through multiple dimensions such as time keywords, aging classification, and event detection, so as to comprehensively judge and identify the timeliness of the target content. Therefore, the accuracy of content distribution can be improved.

[0207] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions or by controlling related hardware through instructions. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0208] For this purpose, an embodiment of the present application provides a computer-readable storage medium, which stores multiple instructions that can be loaded by a processor to execute the steps in any content distribution method provided by the embodiment of the present application. For example, the instructions can execute the following steps:

[0209] Obtain at least one target content, extract at least one time keyword from the target content, identify at least one first candidate distribution timeliness corresponding to the target content from the time keywords, perform timeliness classification on the target content to obtain at least one second candidate distribution timeliness, screen out target events that match the target content from a preset event set, and perform event detection on the target events. Based on the detection results, the first candidate distribution timeliness, and the second candidate distribution timeliness, determine the target distribution timeliness of the target content, and distribute the target content according to the target distribution timeliness.

[0210] For example, obtain at least one target content, determine the content type of the target content. When the target content is text content, use a text recognition model to extract at least one time keyword from the text content. When the target content is other content than text content, convert the target content into text content, and use the text recognition model to extract at least one time keyword from the text content. Query the time limit time words and preset time words in the time keywords. When there is at least one time limit time word, identify the first candidate distribution time limit corresponding to the target content from the time limit time words. When there is at least one preset time word, screen out at least one type of first candidate distribution time limit from the preset first time limit set. Determine the content type of the target content, and based on the content type, perform time limit classification on the target content to obtain at least one dimension of time limit types, and screen out the preset time limits corresponding to the time limit types from the preset second time limit set to obtain the second candidate distribution time limit. Link the events in the preset event set with the target content to obtain at least one linked content, classify the linked content to obtain the matching type between the target content and the events in the preset event set, and based on the matching type, screen out the target events matching the target content from the preset event set, and detect the event of the target event to obtain the detection result of the target event. When it is detected that the target event has changed, determine the event distribution time limit of the target content, respectively determine the current time limit types of the event distribution time limit, the first candidate distribution time limit, and the second candidate distribution time limit, and based on the current time limit types, screen out the target distribution time limit of the target content from the event distribution time limit, the first candidate distribution time limit, and the second candidate distribution time limit, and distribute the target content according to the target distribution time limit.

[0211] For the specific implementation of each of the above operations, reference may be made to the previous embodiments and will not be elaborated here.

[0212] Among them, the computer-readable storage medium may include: read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc.

[0213] Since the instructions stored in the computer-readable storage medium can execute the steps in any content distribution method provided by the embodiments of the present application, the beneficial effects that can be achieved by any content distribution method provided by the embodiments of the present application can be realized. For details, reference may be made to the previous embodiments and will not be elaborated here.

[0214] Among them, according to one aspect of the present application, a computer program product or a computer program is provided. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device executes the methods provided in the various alternative implementations of the above content distribution aspect or article distribution aspect.

[0215] The above has introduced in detail a content distribution method, device, electronic device and computer-readable storage medium provided by the embodiments of the present application. Specific examples are used in this article to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A content distribution method, characterized in that, including: obtaining at least one target content, and extracting at least one time keyword from the target content; identifying at least one first candidate distribution timeliness corresponding to the target content from the time keywords; classifying the timeliness of the target content to obtain at least one second candidate distribution timeliness; screening out a target event matching the target content from a preset event set, and performing event detection on the target event; determining the target distribution timeliness of the target content based on the detection result, the first candidate distribution timeliness, and the second candidate distribution timeliness, and distributing the target content according to the target distribution timeliness.

2. The content distribution method according to claim 1, wherein The identifying at least one first candidate distribution timeliness corresponding to the target content from the time keywords includes: querying a timeliness time word and a preset time word in the time keywords; determining at least one first candidate distribution timeliness corresponding to the target content based on the query result.

3. The content distribution method according to claim 2, characterized in that, The determining at least one first candidate distribution timeliness corresponding to the target content based on the query result includes: when there is at least one of the timeliness time words, identifying the first candidate distribution timeliness corresponding to the target content from the timeliness time words; when there is at least one of the preset time words, screening out at least one type of first candidate distribution timeliness from a preset first timeliness set.

4. The content distribution method according to claim 3, wherein The identifying the first candidate distribution timeliness corresponding to the target content from the timeliness time words includes: identifying time information in the timeliness time words and determining the time type of the time information; when the time information is a timeliness expiration time, converting the time information into a distribution timeliness to obtain a first candidate distribution timeliness; when the time information is an incident time, obtaining the attribute information of the target content, and determining the first candidate distribution timeliness corresponding to the target content based on the attribute information and the time information.

5. The content distribution method according to claim 3, wherein The preset first timeliness set includes a first high-accuracy timeliness set and a first high-recall timeliness set. The first high-accuracy timeliness set includes at least one preset timeliness with an accuracy rate exceeding a first preset accuracy threshold, and the first high-recall timeliness set includes at least one preset timeliness with a recall rate exceeding a first preset recall threshold. The screening out at least one type of first candidate distribution timeliness from the preset timeliness set includes: querying the preset timeliness corresponding to the preset time word in the first high-accuracy timeliness set to obtain a first query result; querying the preset timeliness corresponding to the preset time word in the first high-recall timeliness set to obtain a second query result; determining the first candidate distribution timeliness corresponding to the target content based on the first query result and the second query result.

6. The content distribution method according to claim 1, characterized in that The classifying the timeliness of the target content to obtain at least one second candidate distribution timeliness includes: determining the content type of the target content, and classifying the timeliness of the target content based on the content type to obtain at least one dimension of timeliness types; screening out the preset timeliness corresponding to the timeliness type from a preset second timeliness set to obtain a second candidate distribution timeliness.

7. The content distribution method according to claim 6, characterized in that Performing timeliness classification on the target content based on the content type to obtain timeliness types of at least one dimension, including: When the target content is graphic content, extracting text features of at least one dimension from the target content to obtain the timeliness type corresponding to the text features; When the target content is video content, extracting text features and visual features from the target content to obtain timeliness types of at least one dimension of the target content; When the target content is audio content, converting the target content into text content and extracting text features of at least one dimension from the text content to obtain the timeliness type corresponding to the text features.

8. The content distribution method according to claim 7, wherein The extracting text features of at least one dimension from the target content to obtain the timeliness type corresponding to the text features includes: Using a high-accuracy classification model to extract high-accuracy text features from the target content, and determining the first timeliness type of at least one granularity based on the high-accuracy text features, where the high-accuracy classification model includes a classification model with a classification accuracy exceeding a first preset threshold; Using a high-recall classification model to extract high-recall text features from the target content, and determining the second timeliness type of at least one granularity based on the high-recall text features, where the high-recall classification model includes a classification model with a classification recall rate exceeding a second preset threshold; Taking the first timeliness type and the second timeliness type as the timeliness type corresponding to the text features.

9. The content distribution method according to claim 7, characterized in that The extracting text features and visual features from the target content to obtain timeliness types of at least one dimension of the target content includes: Using a high-accuracy classification model to extract high-accuracy text features and high-accuracy visual features from the target content to obtain high-accuracy timeliness types of at least one granularity of the target content; Using a high-recall classification model to extract high-recall text features and high-recall visual features from the target content to obtain high-recall timeliness types of at least one granularity of the target content; Taking the high-accuracy timeliness type and the high-recall timeliness type as the timeliness type of the target content.

10. The content distribution method according to claim 9, characterized in that, The using a high-accuracy classification model to extract high-accuracy text features and high-accuracy visual features from the target content to obtain high-accuracy timeliness types of at least one granularity of the target content includes: Using a high-accuracy classification model to extract high-accuracy text features and high-accuracy visual features from the target content; Fusing the high-accuracy text features and the high-accuracy visual features to obtain high-accuracy content features; Based on the high-accuracy content features, determining high-accuracy timeliness types of at least one granularity of the target content.

11. The content distribution method according to claim 1, characterized in that The screening out target events matching the target content from a preset event set includes: Linking the events in the preset event set with the target content to obtain at least one linked content; Classifying the linked content to obtain the matching type between the target content and the events in the preset event set; Based on the matching type, screening out target events matching the target content from the preset event set.

12. The content distribution method according to claim 1, wherein Determining the target distribution timeliness of the target content based on the detection result, the first candidate distribution timeliness, and the second candidate distribution timeliness includes: When it is detected that the target event changes, determining the event distribution timeliness of the target content; Respectively determining the current timeliness types of the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness; Based on the current timeliness type, screening out the target distribution timeliness of the target content from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness.

13. The content distribution method according to claim 12, characterized in that, Determining the event distribution timeliness of the target content includes: Obtaining the target time when the target event changes, and using the target time as the current stop distribution time of the target content; Based on the current stop distribution time, generating the event distribution timeliness of the target content.

14. The content distribution method according to claim 12, wherein Based on the current timeliness type, screening out the target distribution timeliness of the target content from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness includes: When the current timeliness type includes at least one high-accuracy timeliness, screening out the target distribution timeliness of the target content from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness; When the high-accuracy timeliness and the high-citation timeliness are not included in the current timeliness type, using the preset distribution timeliness as the target distribution timeliness of the target content.

15. The content distribution method according to claim 14, characterized in that, Screening out the target distribution timeliness of the target content from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness includes: Screening out at least one current distribution timeliness corresponding to the high-accuracy timeliness from the event distribution timeliness, the first candidate distribution timeliness, and the second candidate distribution timeliness; Determining the timeliness length of each current distribution timeliness, and comparing the timeliness lengths; Based on the comparison result, screening out the target distribution timeliness of the target content from the current distribution timeliness.

16. The content distribution method according to claim 14, characterized in that, Further includes: When the current timeliness type includes at least one of the high-citation timeliness, sending the target content to an audit server so that the audit server audits the target content.

17. A content distribution device, characterized in that, Includes: An acquisition unit, configured to acquire at least one target content, and extract at least one time keyword from the target content; An identification unit, configured to identify at least one first candidate distribution timeliness corresponding to the target content from the time keywords; A classification unit, configured to perform timeliness classification on the target content to obtain at least one second candidate distribution timeliness; A matching unit, configured to screen out a target event matching the target content from a preset event set, and perform event detection on the target event; A distribution unit, configured to determine the target distribution timeliness of the target content based on the detection result, the first candidate distribution timeliness, and the second candidate distribution timeliness, and distribute the target content according to the target distribution timeliness.

18. An electronic device, characterized in that, Includes a processor and a memory, the memory stores an application program, and the processor is configured to run the application program in the memory to execute the steps in the content distribution method according to any one of claims 1 to 16.

19. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by a processor, the steps in the content distribution method according to any one of claims 1 to 16 are implemented.

20. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions, and the instructions are adapted to be loaded by a processor to execute the steps in the content distribution method according to any one of claims 1 to 16.

Images