Data scheduling and distribution method, device and equipment and computer readable storage medium

Abstract

The application provides a data scheduling and distribution method and device, equipment and a computer readable storage medium. The method comprises the following steps: acquiring multimedia data to be distributed, and acquiring a plurality of multimedia reference data obtained by crawling from a network; performing similarity matching on the multimedia data and the plurality of multimedia reference data to obtain a matching result; when it is determined that the multimedia data meets an accelerated distribution condition based on the matching result, acquiring a current processing state of the multimedia data; and based on the current processing state, adjusting a scheduling strategy of the multimedia data to accelerate the scheduling and distribution of the multimedia data. Through the application, the scheduling and content distribution speed of high-quality multimedia data can be improved.

Description

Technical Field

[0001] This application relates to Internet technology, and more particularly to a data scheduling and distribution method, apparatus, device, and computer-readable storage medium. Background Technology

[0002] In the era of rapid internet development, with the lowering of barriers to content production, video uploads have grown exponentially. These videos include various content creation organizations, such as self-media and institutional PGC and UGC content. The surge in video uploads necessitates rapid content review to ensure the security of distributed content. Currently, this is primarily achieved through a combination of extensive human resources and machine algorithms. As relevant departments place increasing emphasis on regulating social media platforms, coupled with the alarming damage that harmful content can inflict on these platforms, major social media platforms are now investing heavily in content security review. Content security review has become a top priority for platforms such as short videos, news, and live streaming, regardless of whether it's done manually or through systematic machine review, prioritizing the safest and most product-appropriate review results.

[0003] Because all content requires manual review, this process is both costly and inefficient, failing to guarantee quick approval for popular and high-quality content from creators. With the rapid increase in content volume, both costs and efficiency are extremely high, easily leading to content backlog. This is especially true for UGC content; if it cannot be reviewed and processed quickly, it cannot be distributed rapidly, reducing distribution efficiency and significantly impacting user experience. Summary of the Invention

[0004] This application provides a method, apparatus, and computer-readable storage medium that can improve the scheduling and distribution speed of high-quality multimedia data.

[0005] The technical solution of this application embodiment is implemented as follows:

[0006] This application provides a data scheduling and distribution method, including:

[0007] Acquire the multimedia data to be distributed, and obtain multiple multimedia reference data crawled from the network;

[0008] The multimedia data and the plurality of multimedia reference data are matched for similarity to obtain the matching results;

[0009] When it is determined from the matching results that the multimedia data meets the accelerated distribution conditions, the current processing status of the multimedia data is obtained;

[0010] Based on the current processing state, the scheduling strategy for the multimedia data is adjusted to accelerate the scheduling and distribution of the multimedia data.

[0011] This application provides a data scheduling and distribution device, including:

[0012] The first acquisition module is used to acquire the multimedia data to be distributed and to acquire multiple multimedia reference data crawled from the network.

[0013] The similarity matching module is used to perform similarity matching between the multimedia data and the plurality of multimedia reference data to obtain a matching result;

[0014] The second acquisition module is used to acquire the current processing status of the multimedia data when it is determined from the matching result that the multimedia data meets the accelerated distribution conditions.

[0015] The strategy adjustment module is used to adjust the scheduling strategy of the multimedia data based on the current processing state, so as to accelerate the scheduling and distribution of the multimedia data.

[0016] In some embodiments, the similarity matching module is further configured to:

[0017] Obtain the attribute information of the multimedia data and the attribute information of the plurality of multimedia reference data, wherein the attribute information includes title information;

[0018] The attribute information and multimedia data of the multimedia data are vectorized to obtain the first title vector and the first multimedia vector of the multimedia data.

[0019] The attribute information of the plurality of multimedia reference data and the multimedia reference data are vectorized to obtain the second title vector and the second multimedia vector of the plurality of multimedia reference data.

[0020] Determine the title similarity between the first title vector and each of the second title vectors, and the multimedia similarity between the first multimedia vector and each of the second multimedia vectors, respectively;

[0021] The matching results are determined based on the similarity of each title and the similarity of each multimedia element.

[0022] In some embodiments, the similarity matching module is further configured to:

[0023] Based on the title similarity and multimedia similarity, determine whether the target multimedia reference data exists among the multiple multimedia reference data;

[0024] The title similarity between the target multimedia reference data and the multimedia data is less than a first similarity threshold, and / or the multimedia similarity between the target multimedia reference data and the multimedia data is less than a second similarity threshold;

[0025] When the target multimedia reference data exists among the plurality of multimedia reference data, the matching result is determined to be a successful match.

[0026] In some embodiments, the device further includes:

[0027] The first determining module is used to input the first title vector and the first multimedia vector of the multimedia data and the second title vector and the second multimedia vector of the target multimedia reference data into a trained neural network model when the matching result is a successful match, in order to determine the target similarity between the multimedia data and the target multimedia reference data.

[0028] The third acquisition module is used to acquire the first audio data of the multimedia data and the second audio data of the target multimedia reference data when the target similarity is greater than the third similarity threshold.

[0029] The second determining module is used to determine the audio similarity between the first audio data and the second audio data;

[0030] The third determining module is used to determine that the multimedia data meets the preset accelerated distribution conditions when the audio similarity is greater than the fourth similarity threshold.

[0031] In some embodiments, the similarity matching module is further configured to:

[0032] Obtain the first publishing account identifier of the multimedia data and the second publishing account identifier of the plurality of multimedia reference data;

[0033] Determine whether a second publishing account identifier exists that is identical to the first publishing account identifier;

[0034] If a second publishing account with the same identifier as the first publishing account exists, the matching result is determined to be a successful match;

[0035] Correspondingly, the device also includes:

[0036] The fourth determining module is used to determine that the multimedia data meets the preset accelerated distribution conditions when the matching result is a successful match.

[0037] In some embodiments, the current processing state includes a manual review state, a machine review state, and a disabled state. Correspondingly, the policy adjustment module is further configured to:

[0038] When the current processing status is manual review, the distribution strategy of the multimedia data will be adjusted to a first-release-later-review strategy.

[0039] When the current processing status is machine review status, the processing priority of the multimedia data is increased;

[0040] When the current processing state is disabled, the processing state of the multimedia data is adjusted to the enabled state.

[0041] In some embodiments, the device further includes:

[0042] A tagging module is used to add first tagging information to the multimedia data when the multimedia data meets the accelerated distribution conditions;

[0043] The fourth acquisition module is used to acquire the initial distribution weight of multimedia data with the first tag information during the content distribution stage;

[0044] The weighting module is used to increase the initial distribution weight according to a preset weighting adjustment rule to obtain the target distribution weight;

[0045] The content distribution module is used to distribute the multimedia data with the first tag information based on the target distribution weight.

[0046] In some embodiments, the device further includes:

[0047] The fifth acquisition module is used to acquire the preset target website and crawling strategy;

[0048] The data crawling module is used to crawl multiple candidate multimedia data of a preset duration from the target website using the crawling strategy.

[0049] The sixth acquisition module is used to acquire multiple interactive information and multiple publishing account identifiers of the multiple candidate multimedia data, wherein the interactive information includes: number of views, number of likes, and number of shares;

[0050] The fifth determining module is used to determine multimedia reference data from the multiple candidate multimedia data based on the multiple interactive information and the multiple publishing account identifiers.

[0051] In some embodiments, the device further includes:

[0052] The seventh acquisition module is used to acquire negative feedback information of multimedia data with the first tagging information, wherein the negative feedback information includes the number of reports;

[0053] The review module is used to determine that the multimedia data with the first tag information needs to be reviewed again when the number of reports reaches a preset threshold.

[0054] In some embodiments, the device further includes:

[0055] The sixth determining module is used to determine the first total number of multimedia reference data crawled within a preset time period;

[0056] The seventh determining module is used to determine the second total number of multimedia data that meet the accelerated distribution conditions within the preset time period;

[0057] The eighth determining module is used to determine the coverage of the multimedia reference data based on the first total and the second total.

[0058] The ninth determining module is used to determine the target publishing account identifier based on the multimedia reference data when the coverage rate is lower than a preset coverage rate threshold.

[0059] The sending module is used to send invitation information to the terminal corresponding to the target publishing account identifier, so as to invite the terminal to publish multimedia data.

[0060] In some embodiments, the device further includes:

[0061] The eighth acquisition module is used to acquire the multimedia data to be published uploaded by the content generation terminal, and to acquire the publishing account identifier of the multimedia data to be published.

[0062] The tenth determining module is used to determine the review level of the publishing account identifier based on the historical multimedia data when the historical multimedia data corresponding to the publishing account identifier can be obtained;

[0063] The data update module is used to add the publishing account identifier to the multimedia reference data when the review level is greater than a preset level threshold.

[0064] In some embodiments, the tenth determining module is further configured to:

[0065] Obtain interactive information from the historical multimedia data, including the number of views, likes, and shares.

[0066] The review level of the publishing account identifier is determined based on the number of views, likes, and shares.

[0067] This application provides a data scheduling and distribution device, including:

[0068] Memory, used to store executable instructions;

[0069] A processor, when executing executable instructions stored in the memory, implements the method provided in the embodiments of this application.

[0070] This application provides a computer-readable storage medium storing executable instructions for inducing a processor to execute and implement the method provided in this application.

[0071] This application provides a computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the data scheduling and distribution method described above.

[0072] The embodiments of this application have the following beneficial effects:

[0073] After acquiring the multimedia data to be distributed, multiple multimedia reference data are obtained from the web. These reference data can be the latest popular multimedia data or include identifiers of high-quality publishing accounts. Then, the multimedia data and the multiple multimedia reference data are matched for similarity to obtain matching results. When the multimedia data meets the accelerated distribution conditions based on the matching results, the current processing status of the multimedia data is acquired. Based on the current processing status, the scheduling strategy of the multimedia data is adjusted to accelerate the scheduling and distribution of the multimedia data. Thus, when the multimedia data to be published matches the content of the multimedia reference data or is multimedia data published by a high-quality publishing account, the scheduling strategy of the multimedia data will be adjusted so that the multimedia data can be distributed first. This not only enables the rapid location of high-quality content, but also enables high-quality content or content created by high-quality creators to be activated and distributed faster with a shorter time delay, effectively reducing the content processing time. Attached Figure Description

[0074] Figure 1 This is a schematic diagram of the structure of a content moderation and processing system in related technologies;

[0075] Figure 2 This is a schematic diagram of the network architecture of the data scheduling and distribution system provided in the embodiments of this application;

[0076] Figure 3 This is a schematic diagram of the server structure provided in an embodiment of this application;

[0077] Figure 4 This is a schematic diagram illustrating an implementation process of the data scheduling and distribution method provided in an embodiment of this application.

[0078] Figure 5A schematic diagram illustrating another implementation of the data scheduling and distribution method provided in this application embodiment;

[0079] Figure 6 A schematic diagram illustrating another implementation process of data scheduling and distribution provided in the embodiments of this application;

[0080] Figure 7 This is a schematic diagram of the structure of a data scheduling and distribution system based on content vector matching provided in an embodiment of this application;

[0081] Figure 8 A schematic diagram illustrating the capabilities of the web crawler and parsing services provided in the embodiments of this application;

[0082] Figure 9 This is a schematic diagram illustrating the implementation of the similarity matching service provided in the embodiments of this application;

[0083] Figure 10 This is a schematic diagram illustrating the generation of vectors using a Siamese network, as provided in an embodiment of this application. Detailed Implementation

[0084] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0085] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0086] In the following description, the terms "first, second, third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0087] In the implementation of this application, the collection and processing of relevant data should strictly comply with the requirements of relevant laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of laws and regulations and the authorization of the personal information subject.

[0088] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0089] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0090] 1) Articles can refer to articles that a self-media platform actively edits and publishes after opening a public account, and recommends to users for reading. These articles may include videos or pictures.

[0091] 2) Professionally Generated Content (PGC), an internet term, generally refers to personalized content, diverse perspectives, and virtualized social relationships. Also known as Professionally-Produced Content (PPC).

[0092] 3) Multi-Channel Network (MCN) unites PGC content and, with strong capital support, ensures the continuous output of content, thereby ultimately achieving stable commercial monetization.

[0093] 4) User-generated content (UGC) emerged alongside the Web 2.0 concept, which emphasizes personalization. It is not a specific business service, but rather a new way users engage with the internet, shifting from primarily downloading to a balance between downloading and uploading.

[0094] 5) Professional User Generated Content (PUGC) is professional audio content produced in UGC format that is relatively close to PGC.

[0095] 6) Terminal programs, which are applications that run on the terminal, such as instant messaging and social platforms, and can receive messages and feeds.

[0096] 7) Server-side: A server program deployed on multiple servers, specifically designed to provide remote network services for terminal programs.

[0097] 8) Feeds, also translated as sources, deliveries, information providers, summaries, sources, news subscriptions, or web sources, are a data format through which websites disseminate the latest information to users. They are typically arranged in a timeline format; the timeline is the most original, intuitive, and basic form of feed display. A prerequisite for users to subscribe to a website is that the website provides feeds. The aggregation of feeds in one place is called aggregation, and the software used for aggregation is called an aggregator. For end users, an aggregator is software specifically designed for subscribing to websites; it is also commonly referred to as an RSS reader, feed reader, or news reader.

[0098] 9) Machine Learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, and many other disciplines. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance.

[0099] 10) Deep learning: The concept of deep learning originated from the research on artificial neural networks. A multilayer perceptron with multiple hidden layers is a type of deep learning structure. Deep learning discovers distributed feature representations of data by combining low-level features to form more abstract high-level representations of attribute categories or features.

[0100] 11) Short videos, also known as short films, are a form of internet content dissemination. They are generally video content with a duration of less than 5 minutes that is disseminated on new internet media. With the popularization of mobile terminals and the acceleration of network speed, short, fast, and high-traffic content has gradually gained favor from major platforms, fans, and capital.

[0101] 12) Perceptual Hash Algorithms, including aHash, pHash, and dHash. Perceptual hashing does not calculate hash values ​​strictly, but rather in a more relative way, because "similarity" is a relative judgment. aHash: Average hash. Relatively fast, but often inaccurate; pHash: Perceptual hash. Relatively accurate, but slower; dHash: Difference hash. Highly accurate and very fast. Therefore, dHash was chosen as a fast algorithm for single-image deduplication and is suitable for large-scale applications.

[0102] To better understand the multimedia data processing method provided in the embodiments of this application, the implementation methods of processing multimedia data to achieve multimedia data publishing in related technologies and their shortcomings will be explained first.

[0103] Social networks originated from online social networking, which began with email. The internet is essentially a network of computers. Early email solved the problem of long-distance email transmission and remains the most widespread application on the internet, marking the beginning of online social networking. BBS (Baidu, Blog, and Forum) went a step further, normalizing "mass messaging" and "forwarding," theoretically enabling the dissemination of information and discussion of topics to everyone (the limit being the number of BBS visitors). It became an early platform for spontaneous content generation on the internet. Recently, due to the widespread adoption of smartphones, the ubiquitous availability of Wi-Fi, the general reduction in 4G tariffs, and the imminent arrival of the 5G era, in the current context of the mobile internet, users' information consumption needs are transitioning from a text-based era to a video-based era. Therefore, short videos will gradually become one of the dominant content formats on the mobile internet, to some extent replacing text-based content consumption, and gradually gaining dominance in text-based media such as news and social platforms. This content is typically displayed as a feed stream for users to quickly refresh. Facebook's Newsfeed on the homepage can be seen as a new type of aggregator, where users subscribe to feeds from their friends or followed public figures, and the content consists of their publicly posted updates. When you have a large number of active friends, you can receive constantly updated content; this is the most common feed format. Weibo and Zhihu are similar. Time is the ultimate dimension that a feed follows, because content updates are the result of continuous requests to the server. The Timeline is the most primitive, intuitive, and basic form of feed display; if there are better ones, they are designs built upon the Timeline.

[0104] Short videos refer to frequently pushed video content, ranging from a few seconds to a few minutes, played on various new media platforms, suitable for viewing on mobile devices and during short leisure periods. The content integrates themes such as skill sharing, humor, fashion trends, social hot topics, street interviews, public welfare education, advertising creativity, and commercial customization. Due to their short length, they can be standalone videos or part of a series. Unlike micro-films and live streams, short video production does not have the specific expressive forms and team configuration requirements of micro-films. They are characterized by simple production processes, low production barriers, and high participation, and have greater dissemination value than live streams. The ultra-short production cycle and entertaining content pose a challenge to the copywriting and planning skills of short video production teams. Excellent short video production teams usually rely on well-established self-media or IPs, possessing not only high-frequency and stable content output but also strong fan channels. The emergence of short videos has enriched the forms of native advertising on new media. Currently, short videos have evolved from initial UGC, PGC, and user uploads to specialized short video production agencies, MCNs, and the rise of numerous leading traffic platforms such as professional short video apps. Short videos have become a crucial dissemination method for content creation and social media platforms. While sparking a frenzy among content creators and impacting video media platforms, short videos have further amplified their influence, leading to a fierce competition among major news platforms. Consequently, the variety and richness of short video content have increased dramatically. Both producers and consumers of short video content have become a massive group.

[0105] Traditional databases consist of structured tables containing symbolic information. For example, an image set might be represented by a list of indexed photos, each row containing one. Each row contains information such as image identifiers and descriptive phrases. Each row can also be associated with entries from other tables, such as photos being associated with lists of names. Many AI tools generate high-dimensional vectors, such as text embedding tools like word2vec and CNN descriptors trained with deep learning. These representations are more powerful and flexible than fixed symbolic representations. However, traditional databases retrieved using SQL are not adapted to these new vector representations, resulting in very low efficiency. First, the massive amounts of new multimedia streams create billions of vectors. Second, and more importantly, finding similar entries means finding closely spaced high-dimensional vectors. For current standard search languages ​​like Lexal, this is extremely inefficient, or even impossible. For similarity search and classification, the following operations are required: given a search vector, return a list of database objects that are closest to this vector in Euclidean distance; given a search vector, return a list of database objects with the highest vector dot product. Traditional SQL database systems are not very usable because they are optimized for hash-based searches or 1D interval searches.

[0106] Faiss is an open-source clustering and similarity search library from Facebook AI. It provides efficient similarity search and clustering for dense vectors, supporting searches on the order of billions of vectors, and is currently the most mature approximate nearest neighbor search library. It includes various algorithms for searching vector sets of arbitrary sizes, as well as support code for algorithm evaluation and parameter tuning. The Faiss library includes multiple methods for similarity search, with core modules including high-performance clustering, Principal Component Analysis (PCA), and Product Quantization (PQ). It assumes that instances are represented as vectors and identified by integers, and that vectors can be compared with L2 distance or dot product. Vectors similar to the query vector are those with the lowest L2 distance or the highest dot product with the query vector. It also supports cosine similarity, as this is the dot product over normalized vectors. In other words, Faiss primarily uses two similarity calculation methods: Euclidean distance and dot product. To clarify the dot product, after vector normalization, the cosine similarity between two vectors is consistent with their dot product. The key technologies used in Faiss include: OpenMP, heap sort, PQ algorithm, inverted index, Kmeans clustering algorithm, and principal component analysis (PCA).

[0107] Faiss is specifically optimized for memory usage and speed, providing state-of-the-art GPU execution for related indexing methods. Once these vectors are extracted by the learning machine (from images, videos, text files, or other sources), they can be input into the similarity search library. Sacrificing some accuracy can result in orders of magnitude speed improvements in similarity search. For example, swapping the first and second results of an image similarity search might have little difference, as they are likely both correct answers for a given retrieval. Accelerating search involves some preprocessing of the dataset, which Facebook calls indexing. Faiss is typically used as a foundational component for underlying vector retrieval.

[0108] From content production to distribution, a necessary content review system is required. Because the security and quality of video content directly impact a platform's survival, user experience, and related social responsibilities, relevant companies place great importance on this. Currently, mainstream technical review processes typically involve two systems: a sensitive content review system and a separate system for reviewing other content. After content is created but before distribution, it is necessary to utilize systems such as… Figure 1 The content review and processing system 111 shown performs content review and processing.

[0109] Companies may adjust the leniency of other review conditions based on their operational needs, even skirting the rules. However, no company is willing to take the risk of reviewing sensitive content. Content review systems are basically based on the laws and regulations of different countries, as well as the policies and regulations of local internet supervision departments. For other review content, relying solely on manual review in the video field is both inefficient and extremely costly. Therefore, the mainstream technology currently uses relevant machine learning algorithms to preprocess and prompt content for review. In the initial stage, machine processing filters out content that obviously violates the law, and then a second filtering review is conducted in conjunction with manual review. For video content, depending on the content, such as live broadcasts, short videos, and personally uploaded videos, videos are composed of frames of images and audio. Audio often has problems such as audio-visual asynchrony. In video processing, frame-by-frame data comparison with the server is usually used for identification. Its review mode is the same as that of image review, judging the scene (outdoor or indoor), the faces (whether the people in the picture are celebrities or sensitive content), and whether it is pornographic (based on the nudity of the picture, it can be classified as normal, sexy, pornographic, etc.). For video streaming apps, content review often relies heavily on machine-based screening. Content containing graphic violence, pornography, or sensitive news is prioritized for filtering, while content not explicitly shown visually is difficult for machines to detect. The current mainstream review process can be understood as a combination of algorithms and human assistance. The main assistance involves algorithms providing prompts to reviewers, such as video clarity, whether it's a trending topic, the clickbait nature of the title, and whether it contains borderline content. Additionally, audio is extracted from the original video to identify any illegal information. For content that is uninformative, clickbait-prone, subjectively descriptive, or logically flawed, human reviewers rely primarily on their experience and established standards, making the actual effectiveness difficult to guarantee.

[0110] The disadvantages of multimedia moderation schemes in related technologies include the following:

[0111] First, manual review increases costs and is inefficient, making it difficult to guarantee quick approval for content from top-rated and high-quality creators online. The entire process takes too long, especially for trending content.

[0112] Secondly, it is impossible to keep abreast of the high-quality content produced by top self-media authors across the entire network, and to provide timely coverage and targeted support within the information flow content ecosystem, such as accelerated review and weighted distribution.

[0113] Third, in the content recommendation stage, there is an urgent need to build user reputation and improve key indicators such as user stickiness and usage time with high-quality content. Due to the uneven distribution of recommendations, the role of high-quality content will be more prominent. It is impossible to quantify and detect the proportion and coverage of high-quality content on the platform and across the entire network, and it is impossible to accelerate the processing of high-quality content that has already been introduced.

[0114] Based on the above problems, embodiments of this application provide a multimedia data processing method, apparatus, device, and computer-readable storage medium, which can perform targeted mining, processing, and acceleration of high-quality top content on network platforms during the content introduction, content processing, and content distribution stages.

[0115] The following describes exemplary applications of the multimedia data processing device provided in the embodiments of this application. The device provided in the embodiments of this application can be implemented as various types of user terminals such as laptops, tablets, desktop computers, set-top boxes, and mobile devices (e.g., mobile phones, portable music players, personal digital assistants, dedicated messaging devices, portable gaming devices), or as a server. The following will describe exemplary applications when the device is implemented as a server.

[0116] See Figure 2 , Figure 2 This is a schematic diagram of the network architecture of the data scheduling and distribution system 100 provided in the embodiments of this application, as shown below. Figure 2 As shown, the network architecture includes a content production terminal 200, a network 300, a server 400, and a content consumption terminal 500. The content production terminal 200 and the content consumption terminal 500 are connected to the server 200 through the network 300, which can be a wide area network, a local area network, or a combination of both.

[0117] The user determines the multimedia data to be published through the content generation terminal 200. This multimedia data can be a short video or micro-video recorded by the content generation terminal 200, an article edited by the content generation terminal, or locally stored text and image content or video. Then, in response to the data publishing instruction, the content generation terminal 200 sends the multimedia data to be published to the server 400. After receiving the multimedia data to be published, the server 400 stores it in the multimedia database and obtains high-quality multimedia reference data that has been crawled from the network in advance. Then, it matches the multimedia data to be published with the high-quality multimedia reference data, and if the multimedia data to be published meets the conditions for accelerated distribution, it prioritizes the distribution of the multimedia data and sends it to the content consumption terminal 500.

[0118] Server 400 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. Content production terminal 200 and content consumption terminal 500 can be smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, wearable devices, in-vehicle computers, etc., but are not limited to these. Terminals and servers can be directly or indirectly connected via wired or wireless communication, which is not limited in this embodiment.

[0119] See Figure 3 , Figure 3 This is a schematic diagram of the structure of the server 400 provided in an embodiment of this application. Figure 3 The server 400 shown includes at least one processor 410, memory 450, at least one network interface 420, and a user interface 430. The various components in server 400 are coupled together via a bus system 440. It is understood that the bus system 440 is used to implement communication between these components. In addition to a data bus, the bus system 440 also includes a power bus, a control bus, and a status signal bus. However, for clarity, ... Figure 2 The general labeled all buses as Bus System 440.

[0120] Processor 410 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0121] User interface 430 includes one or more output devices 431 that enable the presentation of media content, including one or more speakers and / or one or more visual displays. User interface 430 also includes one or more input devices 432, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0122] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state storage, hard disk drives, optical disk drives, etc. The memory 450 may optionally include one or more storage devices physically located away from the processor 410.

[0123] The memory 450 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), and the volatile memory may be random access memory (RAM). The memory 450 described in this application embodiment is intended to include any suitable type of memory.

[0124] In some embodiments, memory 450 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, as illustrated below.

[0125] Operating system 451 includes system programs for handling various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, driver layer, etc., for implementing various basic business functions and handling hardware-based tasks;

[0126] The network communication module 452 is used to reach other computing devices via one or more (wired or wireless) network interfaces 420, exemplary network interfaces 420 including: Bluetooth, WiFi, and Universal Serial Bus (USB), etc.

[0127] Presentation module 453 is configured to enable the presentation of information (e.g., a user interface for operating peripheral devices and displaying content and information) via one or more output devices 431 associated with user interface 430 (e.g., a display screen, a speaker, etc.).

[0128] The input processing module 454 is used to detect and translate one or more user inputs or interactions from one or more input devices 432.

[0129] In some embodiments, the apparatus provided in this application can be implemented in software. Figure 2 A data scheduling and distribution device 455 stored in memory 450 is shown. It can be software in the form of programs and plug-ins, including the following software modules: a first acquisition module 4551, a similarity matching module 4552, a second acquisition module 4553, and a strategy adjustment module 4554. These modules are logically related and can therefore be arbitrarily combined or further split according to the functions they implement.

[0130] The functions of each module will be explained below.

[0131] In other embodiments, the apparatus provided in this application can be implemented in hardware. For example, the apparatus provided in this application can be a processor in the form of a hardware decoding processor, which is programmed to execute the data scheduling and distribution method provided in this application. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0132] To better understand the methods provided in the embodiments of this application, we will first explain artificial intelligence, its various branches, and the application fields involved in the methods provided in the embodiments of this application.

[0133] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0134] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning. The solutions provided in this application mainly relate to machine learning techniques in artificial intelligence, which will be described below.

[0135] Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, and inductive learning.

[0136] Artificial intelligence cloud services, also commonly known as AI as a Service (AIaaS), are a mainstream service model for artificial intelligence platforms. Specifically, AIaaS platforms break down several common AI services and provide them independently or as bundled services in the cloud. This service model is similar to opening an AI-themed marketplace: all developers can access and use one or more AI services provided by the platform through API interfaces. Some experienced developers can also use the AI ​​framework and AI infrastructure provided by the platform to deploy and maintain their own dedicated cloud AI services. In the information recommendation method provided in this embodiment of the invention, the data scheduling and distribution system can be deployed and maintained through the AI ​​framework and AI infrastructure provided by the artificial intelligence cloud service.

[0137] The data scheduling and distribution method provided in this application will be described in conjunction with exemplary applications and implementations of the server provided in the embodiments of this application.

[0138] See Figure 3 , Figure 3 This is a schematic diagram illustrating an implementation flow of the data scheduling and distribution method provided in this application embodiment. This data scheduling and distribution method is applied to a server, and will be discussed below in conjunction with... Figure 3 The steps shown are explained.

[0139] Step S101: Obtain the multimedia data to be distributed and obtain multiple multimedia reference data crawled from the network.

[0140] Here, the multimedia data to be distributed can be video data, image and text data, or even data consisting only of text. In some embodiments, when a user records a video or edits an article using their own terminal (i.e., the content generation terminal in other embodiments) to obtain multimedia data to be published, and uploads the multimedia data to the server to request publication, the server, upon receiving the multimedia data, stores the multimedia data in the content database and adds it to the multimedia data queue to be distributed. The server's scheduling center service then schedules the multimedia data to be distributed from the multimedia data queue according to preset scheduling rules.

[0141] Multimedia reference data can be popular or high-quality multimedia data crawled from the Internet. In some embodiments, multimedia reference data can also include account information of high-quality publishers.

[0142] Step S102: Perform similarity matching on the multimedia data and the plurality of multimedia reference data to obtain matching results.

[0143] Here, step S102 can be implemented by obtaining the attribute information of the multimedia data and the vector of the multimedia data itself, and obtaining the attribute information of the multimedia reference data and the vector of the multimedia reference data itself. Then, the similarity between the two vectors is calculated to determine whether there is multimedia reference data that meets the matching conditions with the multimedia data, thereby obtaining the matching result. The attribute information of the multimedia data may at least include title information, and in some embodiments, the attribute information may also include a cover image.

[0144] In some embodiments, step S102 may involve comparing the publishing account identifier of the multimedia data with the publishing account identifier in the multimedia reference data. When the publishing account identifier of the multimedia data is the same as a publishing account identifier in the multimedia reference data, a successful matching result is obtained.

[0145] Step S103: When it is determined from the matching result that the multimedia data meets the accelerated distribution conditions, the current processing status of the multimedia data is obtained.

[0146] When vector matching is performed between multimedia data and multimedia reference data, and a successful match is obtained, it indicates that the multimedia data is similar to one or more multimedia reference data. In this case, further precise matching is required. When the multimedia data is video data, it is also necessary to verify the audio of the multimedia data and one or more multimedia reference data that match it to ensure that the multimedia data meets the conditions for accelerated distribution. When the matching result is obtained by matching the publishing account identifier of the multimedia data, a successful match means that the multimedia data meets the conditions for accelerated distribution.

[0147] The current processing status of multimedia data can include manual review status, machine review status, and disabled status.

[0148] Step S104: Based on the current processing state, adjust the scheduling strategy of the multimedia data to accelerate the scheduling and distribution of the multimedia data.

[0149] Here, when the current processing status is manual review, it means that the multimedia data is in the manual review queue. Since manual review is slow and requires a long waiting time, the scheduling strategy for multimedia data can be adjusted to "send first, then review" to speed up scheduling and distribution. If the current processing device is machine review, it means that the multimedia data is in the machine review queue. In this case, the processing priority of the multimedia data can be increased, and the multimedia data can be reviewed first to shorten the machine review time, which can also speed up scheduling and distribution. If the current processing status of the multimedia data is disabled, the multimedia data should be updated to enabled to proceed with the subsequent review and distribution process.

[0150] In the data scheduling and distribution method provided in this application embodiment, after obtaining the multimedia data to be distributed, multiple multimedia reference data crawled from the network are obtained. These multiple multimedia reference data can be the latest popular multimedia data, and can also include high-quality publishing account identifiers. Then, the multimedia data and the multiple multimedia reference data are matched for similarity to obtain a matching result. When it is determined based on the matching result that the multimedia data meets the accelerated distribution conditions, the current processing state of the multimedia data is obtained. Based on the current processing state, the scheduling strategy of the multimedia data is adjusted to accelerate the scheduling and distribution of the multimedia data. Thus, when the multimedia data to be published matches the content of the multimedia reference data or is multimedia data published by a high-quality publishing account, the scheduling strategy of the multimedia data will be adjusted so that the multimedia data can be distributed first. This not only enables the rapid location of high-quality content, but also enables high-quality content or content created by high-quality creators to be activated and distributed faster within a shorter time delay, effectively reducing the content processing time.

[0151] In some embodiments, Figure 3 The step S102 shown can be implemented in two ways: one is to match the similarity between the attribute information of the multimedia data and the multimedia data itself and the attribute information of the multimedia reference data; the other is to match the publishing account identifier of the multimedia data with the publishing account identifier in the multimedia reference data. The following will explain these two methods respectively.

[0152] When implemented in the first manner, step S102, "performing similarity matching between the multimedia data and the plurality of multimedia reference data to obtain a matching result," can be achieved through the following steps:

[0153] Step S1021A: Obtain the attribute information of the multimedia data and the attribute information of the plurality of multimedia reference data.

[0154] Here, the attribute information includes title information. In some embodiments, the attribute information also includes a cover image. When the multimedia data has descriptive information, the attribute information also includes descriptive information. The descriptive information can be a summary of the video content or the article content. The descriptive information and title information are generally text information, and the cover image is image information.

[0155] Step S1022A: The attribute information of the multimedia data and the multimedia data are vectorized to obtain the first title vector and the first multimedia vector of the multimedia data.

[0156] In implementing step S1022A, the attribute information and the multimedia data itself can be vectorized based on the attribute information and the type of multimedia data. For example, when the multimedia data is an article including images and text, the attribute information includes title information, description information, and cover image, and the multimedia data includes images and body text. For the title and description information, the simhash algorithm can be used to generate corresponding title vectors and description vectors. For the images and body text, a dual-tower neural network model can be used to generate vectors. Through this dual-tower neural network model, images and body text can be mapped to the same vector space, and image vectors and body text vectors can be generated in this vector space. If the multimedia data is an article including images and body text, then the image vectors and body text vectors constitute the first multimedia vector of the multimedia data. When the multimedia data is video, frames can be extracted from the video file first, and then the semantic fingerprint vector of the video can be constructed through the video frames to obtain the first multimedia vector.

[0157] Step S1023A: The attribute information of the plurality of multimedia reference data and the multimedia reference data are vectorized to obtain the second title vector and the second multimedia vector of the plurality of multimedia reference data.

[0158] Step S1023A is implemented in a similar manner to step S1022A. It involves vectorizing the attribute information and the multimedia reference data itself to obtain the corresponding second title vector and second multimedia vector. If the attribute information of the multimedia reference data includes a cover image and description information, a second cover image vector and a second description vector will also be obtained.

[0159] Step S1024A: Determine the title similarity between the first title vector and each of the second title vectors, and the multimedia similarity between the first multimedia vector and each of the second multimedia vectors.

[0160] Since we have obtained the first title vector and the first multimedia vector of the multimedia data, as well as the second title vector and the second multimedia vector of each multimedia reference data, we can determine the similarity of each title by calculating the distance between the first title vector and each second title vector, and thus determine the similarity of each multimedia by calculating the distance between the first multimedia vector and each second multimedia vector.

[0161] The closer the distance between two vectors, the higher the similarity.

[0162] Step S1025A: Determine the matching results based on the similarity of each title and the similarity of each multimedia.

[0163] This step S1025A can be achieved through the following steps:

[0164] Step SA1: Based on the title similarity and multimedia similarity, determine whether the target multimedia reference data exists among the multiple multimedia reference data;

[0165] Here, the title similarity between the target multimedia reference data and the multimedia data is less than a first similarity threshold, and / or the multimedia similarity between the target multimedia reference data and the multimedia data is less than a second similarity threshold. In other words, the title or multimedia data of the target multimedia reference data is similar to the multimedia data to be distributed.

[0166] Step SA2: When the target multimedia reference data exists among the plurality of multimedia reference data, the matching result is determined to be a successful match.

[0167] In some embodiments, when the target multimedia reference data is not found among the multiple multimedia reference data, it means that there is no multimedia reference data similar to the multimedia data to be distributed. In this case, the multimedia data is reviewed by machine and by human review according to the normal scheduling strategy before being recommended for distribution.

[0168] In steps S1021A to S1025A above, by performing vectorized matching on the multimedia data and the multimedia reference data, it can be determined whether the multimedia data to be distributed is similar to popular and high-quality multimedia reference data crawled from the Internet, thereby further determining whether to accelerate distribution.

[0169] In some embodiments, after completing the content similarity matching of multimedia data and multimedia reference data through steps S1021A to S1025A as described above, it is also necessary to determine whether the multimedia data meets the accelerated scheduling conditions through the following steps:

[0170] Step S201A: When the matching result is a successful match, the first title vector and the first multimedia vector of the multimedia data and the second title vector and the second multimedia vector of the target multimedia reference data are input into the trained neural network model to determine the target similarity between the multimedia data and the target multimedia reference data.

[0171] Here, through the steps S1021A to S1025A described above, it can be determined whether the multimedia data to be distributed is similar to the multimedia reference data, which can be considered as completing the recall process. In order to perform more accurate matching, the multimedia data and the target multimedia reference data need to be input into a well-trained neural network model to determine the target similarity between the multimedia data and the target multimedia reference data.

[0172] In practical implementation, the trained neural network model can be a dual-tower convolutional neural network model.

[0173] Step S202A: When the target similarity is greater than the third similarity threshold, the first audio data of the multimedia data and the second audio data of the target multimedia reference data are obtained.

[0174] The third similarity threshold is higher than the aforementioned first and second similarity thresholds to achieve a more accurate match. When the multimedia data is video data, since the above process involves similarity comparisons of titles, images, and video frames, in order to obtain more accurate results, in step S202A, a further verification process can be performed using the first audio data of the multimedia data and the second audio data of the target multimedia reference data.

[0175] Step S203A: Determine the audio similarity between the first audio data and the second audio data.

[0176] In step S203A, the first audio fingerprint of the first audio data and the second audio fingerprint of the second audio data are first determined. The audio fingerprint can be constructed by performing a Fast Fourier Transform (FFT) on the audio data to create a time-spectrum graph corresponding to the audio data, and the amplitude peak of the time-spectrum graph is used as the audio fingerprint. The audio fingerprint is a one-dimensional vector; generally, the length of the audio fingerprint corresponding to audio data of different durations is also different. Then, the first and second audio fingerprints are used to determine the audio similarity between the first and second audio data.

[0177] Step S204A: When the audio similarity is greater than the fourth similarity threshold, it is determined that the multimedia data meets the preset accelerated distribution conditions.

[0178] When the audio similarity is greater than the fourth similarity threshold, it is considered that the audio of the multimedia data and the target multimedia reference data also match. At this point, the post-recall verification process is completed, confirming that the multimedia data and the target multimedia reference data are precisely matched not only in the title, the data itself, and the audio, thus determining that the multimedia data meets the preset accelerated distribution conditions, and then proceeding to step S103. In some embodiments, if the title and video content of training or lecture videos are highly similar, resulting in a target similarity greater than the third similarity threshold, but the audio data of different lecturers differs greatly, the accuracy of the final result can be ensured by determining whether the audio data is similar.

[0179] When implemented in the second manner, step S102, "performing similarity matching between the multimedia data and the plurality of multimedia reference data to obtain a matching result," can be achieved through the following steps:

[0180] Step S1021B: Obtain the first publishing account identifier of the multimedia data and the second publishing account identifier from the plurality of multimedia reference data.

[0181] When a user publishes multimedia data through a content generation terminal, the publishing request includes the user's publishing account identifier. This identifier, unlike nicknames or other identifiers, is unique. Multiple multimedia reference data sets may contain multiple publishing account identifiers, which can be publishing accounts corresponding to high-quality multimedia content.

[0182] Step S1022B: Determine whether there is a second publishing account identifier that is the same as the first publishing account identifier.

[0183] Here, when a second publishing account identifier exists that is identical to the first publishing account identifier, it indicates that the multimedia data to be distributed was published by a pre-selected high-quality publisher, and the process proceeds to step S1023B; when it is determined that no second publishing account identifier exists that is identical to the first publishing account identifier, it indicates that the multimedia data to be distributed was not published by a pre-selected high-quality publisher. In some embodiments, the multimedia data to be distributed can be scheduled and distributed according to the normal process. In some embodiments, steps S1021A to S1025A can be used to determine whether the content of the multimedia data is high-quality content, thereby determining whether to accelerate the scheduling and distribution.

[0184] Step S1023B: When there is a second publishing account identifier that is the same as the first publishing account identifier, the matching result is determined to be a successful match.

[0185] In some embodiments, after the content similarity matching of multimedia data and multimedia reference data is completed through the above steps S1021B to S1023B, when the matching result is a successful match, it indicates that the multimedia data was published by a high-quality publisher. At this time, it is directly determined that the multimedia data meets the preset accelerated distribution conditions, and then step S103 is executed, which can further improve the scheduling and distribution rate.

[0186] In some embodiments, the current processing state includes a manual review state, a machine review state, and a disabled state, correspondingly, Figure 3 Step S104, "Adjusting the distribution strategy of the multimedia data based on the current processing state," can be achieved through the following steps:

[0187] Step S1041: When the current processing state is manual review, the distribution strategy of the multimedia data is adjusted to a first-release-later-review strategy.

[0188] If the current processing status is manual review, it means the multimedia data is in the manual review queue. Since manual review is slow and requires a long waiting time, the scheduling strategy for the multimedia data can be adjusted to "release first, review later." Adjusting the scheduling strategy to "release first, review later" means not subjecting the multimedia data to manual review, but directly adding it to the content distribution queue and prioritizing its distribution. This reduces manual review time and thus speeds up scheduling and distribution.

[0189] Step S1042: When the current processing state is machine review state, increase the processing priority of the multimedia data.

[0190] If the current processing device is in machine review mode, it means that the multimedia data is in the machine review queue. At this time, the processing priority of the multimedia data can be increased, and the multimedia data can be reviewed first to shorten the machine review time and accelerate the scheduling and distribution.

[0191] Step S1043: When the current processing state is disabled, adjust the processing state of the multimedia data to the enabled state.

[0192] If the current processing status of multimedia data is disabled, update the multimedia data to enabled status for subsequent review and distribution processes, so that high-quality content will not be blocked or filtered out.

[0193] In some embodiments, for multimedia data in a disabled state, it is necessary to obtain the reason for the disablement. When it is determined based on the reason for the disablement that the data can be adjusted to an enabled state, the current processing state of the multimedia data is adjusted to an enabled state, and subsequent review and distribution processes are carried out. When it is determined based on the reason for the disablement that the data cannot be adjusted to an enabled state, the current processing state of the multimedia data is kept in a disabled state.

[0194] In some embodiments, in addition to adjusting the scheduling strategy through steps S1041 to S1043 described above to accelerate scheduling, such as Figure 5 As shown, after step S104, the content distribution speed can be accelerated in the content distribution stage through the following steps:

[0195] Step S301: When the multimedia data meets the accelerated distribution conditions, add first tag information to the multimedia data.

[0196] Here, the first tag information is used to characterize the multimedia data as high-quality multimedia data, and the first tag information is used to increase the distribution weight during the content distribution stage.

[0197] Step S302: Obtain the initial distribution weight of the multimedia data with the first tag information in the content distribution stage.

[0198] Step S303: Increase the initial distribution weight according to the preset weight adjustment rules to obtain the target distribution weight.

[0199] Here, the weight adjustment rule can be to multiply the weight by a preset real number greater than 1 to obtain the target distribution weight, or to add a preset positive number to the initial distribution weight to obtain the target distribution weight.

[0200] Step S304: Based on the target distribution weight, the multimedia data with the first tag information is distributed.

[0201] When implementing step S304,

[0202] In some embodiments, multimedia reference data can be crawled from the network through the following steps:

[0203] Step S401: Obtain the preset target website and crawling strategy.

[0204] Here, the preset target website can be determined based on the behavioral data of the content consumption terminal. In some embodiments, a preset application (App, Application) can also be obtained. The crawling strategy can include breadth-first traversal strategy, depth-first traversal strategy, partial PageRank strategy, online page importance calculation (OPIC), large site priority strategy, etc.

[0205] Step S402: Use the crawling strategy to crawl multiple candidate multimedia data of a preset duration from the target website.

[0206] Steps S401 to S404 can be executed once at a preset time interval, which can be a day, a week, etc. When implementing step S402, obtaining multiple candidate multimedia data for a preset time interval means obtaining multiple candidate multimedia data uploaded by the content generation terminal during the period from the current time to a historical time interval of the preset time interval.

[0207] Step S403: Obtain multiple interactive information and multiple publishing account identifiers of the multiple candidate multimedia data.

[0208] Here, interactive information includes: number of views, number of likes, number of shares, and in some embodiments, the number of comments may also be included.

[0209] Step S404: Based on the multiple interactive information and the multiple publishing account identifiers, determine the multimedia reference data from the multiple candidate multimedia data.

[0210] In implementing step S404, it can be as follows: candidate multimedia data whose number of views is greater than the first threshold, and / or whose number of likes is greater than the second threshold, and / or whose number of shares is greater than the third threshold can be determined as multimedia reference data. Alternatively, it can be as follows: obtain historical multimedia data published by each publishing account, calculate the historical average number of views, historical average number of likes, and historical average number of shares corresponding to each publishing account, and add the publishing account IDs whose historical average number of views is greater than the first threshold, and / or whose historical average number of likes is greater than the second threshold, and / or whose historical average number of shares is greater than the third threshold to the multimedia reference data.

[0211] In some embodiments, the determined multimedia reference data can be stored in a reference database.

[0212] Through the above steps S401 to S404, popular and high-quality multimedia data with high number of views, likes, and shares can be crawled from the network, thereby providing matching criteria for filtering high-quality data from the multimedia data to be distributed.

[0213] In some embodiments, the coverage of multimedia reference data can be determined through the following steps, and when the coverage is low, high-quality publishing accounts can be attracted through business development to publish the multimedia data:

[0214] Step S001: Determine the first total number of multimedia reference data crawled within the preset time period.

[0215] Step S002: Determine the second total number of multimedia data that meet the accelerated distribution conditions within the preset time period.

[0216] Step S003: Determine the coverage of the multimedia reference data based on the first total and the second total.

[0217] Here, the ratio of the second total to the first total is determined as the coverage rate of the multimedia reference data, thereby realizing the specific quantification and detection of the coverage rate of the multimedia reference data.

[0218] Step S004: When the coverage rate is lower than a preset coverage rate threshold, determine the target publishing account identifier based on the multimedia reference data.

[0219] When the coverage rate is lower than the preset coverage rate threshold, it means that there are fewer multimedia reference data and multimedia data to be distributed in the current multimedia database. In order to increase the page views (PV) of the website or app, it is necessary to identify the top N most popular publishing accounts from the multimedia reference data and identify these top N publishing accounts as the target publishing accounts for business expansion.

[0220] Step S005: Send an invitation message to the terminal corresponding to the target publishing account identifier to invite the terminal to publish multimedia data.

[0221] Sending invitation information to the terminal corresponding to the target publishing account identifier can be achieved by first obtaining the communication information corresponding to the target publishing account identifier, and then sending invitation information to that communication information, which can be an email address or the target publishing account itself.

[0222] In steps S001 to S005 above, the coverage rate of multimedia reference data is first calculated to detect the coverage rate. If the coverage rate is low, the target publishing account identifier is selected from the popular or high-quality multimedia reference data, and an invitation message is sent to the terminal corresponding to the target publishing account identifier to attract the self-media author corresponding to the high-quality publishing account identifier to open an account and publish content, thereby increasing the platform's click-through rate and traffic.

[0223] In some embodiments, when a content generation terminal uploads multimedia data to be published, the historical performance of the publishing account identifier of the content generation terminal can be obtained through the following steps to determine the review level of the publishing account, thereby marking a portion of high-quality accounts. High-quality accounts will have a higher priority for manual review scheduling:

[0224] Step S501: Obtain the multimedia data to be published uploaded by the content generation terminal, and obtain the publishing account identifier of the multimedia data to be published.

[0225] Step S502: Determine whether the historical multimedia data corresponding to the publishing account identifier can be obtained.

[0226] Here, when the historical multimedia data corresponding to the publishing account identifier can be obtained, proceed to step S503 to determine the review level of the publishing account identifier based on the historical multimedia data; when the historical multimedia data corresponding to the publishing account identifier cannot be obtained, proceed to step S505.

[0227] Step S503: Determine the review level of the publishing account identifier based on the historical multimedia data.

[0228] Here, when implementing step S503, it may first obtain the interaction information of the historical multimedia data, including the number of views, the number of likes, and the number of shares; and then determine the review level of the publishing account identifier based on the number of views, the number of likes, and the number of shares.

[0229] In practical applications, a pre-defined correspondence between views, likes, shares, and review levels can be established. Therefore, after obtaining the interaction information from the historical multimedia data, the historical average views, likes, and shares corresponding to the publishing account identifier are determined. Based on this correspondence, the review level corresponding to the publishing account identifier is then determined. Review levels can include Level 1, Level 2, Level 3, etc. Higher views, likes, and shares indicate higher popularity for the publishing account identifier, resulting in a higher review level.

[0230] Step S504: When the review level is greater than the preset level threshold, the publishing account identifier is added to the multimedia reference data.

[0231] Here, when the review level is greater than the threshold, it indicates that the publishing account is identified as a high-quality publishing account, and the publishing account identifier is added to the multimedia reference data. In some embodiments, since the publishing account identifier is added to the multimedia reference data, the multimedia data to be published can be considered to meet the accelerated distribution conditions, thereby adjusting the scheduling strategy of the multimedia data to be published to achieve accelerated scheduling and distribution.

[0232] Step S505: Determine the initial review level as the review level of the publishing account identifier.

[0233] The initial review level is preset and can be set to level one. After step S505, the multimedia data to be published is scheduled and distributed according to the normal scheduling and distribution process.

[0234] Based on the foregoing embodiments, this application provides a data scheduling and distribution method, applied to... Figure 2 The network architecture shown is as follows. Figure 6 This is a schematic diagram illustrating another implementation flow of the data scheduling and distribution method provided in the embodiments of this application, as follows: Figure 6 As shown, the process includes:

[0235] Step S601: The content generation terminal acquires the multimedia data to be published.

[0236] Here, the multimedia data to be published can be videos recorded by users through content generation terminals, edited articles from public accounts, or locally stored videos, etc.

[0237] In step S602, the content generation terminal responds to the operation command to publish data and sends the multimedia data to the server.

[0238] In implementation, the content generating terminal may respond to the operation instruction of publishing data by sending a publishing request to the server. This publishing request may carry multimedia data and may also carry the publishing account identifier corresponding to the content generating terminal.

[0239] Step S603: After receiving the multimedia data, the server obtains the metadata of the multimedia data.

[0240] Here, metadata refers to information about information. Metadata can also be considered as the attribute information of multimedia data, such as the size of the multimedia data, the cover image link, the title, the publication time, the account author, the source channel, etc.

[0241] In step S604, the server stores the metadata and multimedia data in the content database and adds the multimedia data to the scheduling center queue.

[0242] In step S605, the server retrieves the multimedia data to be distributed from the scheduling center queue.

[0243] Step S606: The server obtains multiple multimedia reference data crawled from the network.

[0244] In implementation, the server may obtain multiple multimedia reference data from a reference database. In some embodiments, the reference database may also store vectorized identifiers of the multimedia reference data. In this case, the server may obtain multiple multimedia reference data and obtain the vectorized identifiers of the multiple multimedia reference data.

[0245] In step S607, the server performs similarity matching between the multimedia data and the plurality of multimedia reference data to obtain a matching result.

[0246] Step S608: The server determines whether the multimedia data meets the accelerated distribution conditions based on the matching result.

[0247] Here, when it is determined that the multimedia data meets the accelerated distribution conditions, proceed to step S609; when it is determined that the multimedia data does not meet the accelerated distribution conditions, proceed to step S617.

[0248] Step S609: The server obtains the current processing status of the multimedia data.

[0249] The current processing status includes, but is not limited to, manual review status, machine review status, and disabled status.

[0250] Step S610: Based on the current processing state, the server adjusts the scheduling strategy of the multimedia data to accelerate the scheduling and distribution of the multimedia data.

[0251] When the current processing status is manual review, it means that the multimedia data is in the manual review queue. Since manual review is slow and requires a long waiting time, the scheduling strategy for multimedia data can be adjusted to "send first, then review" to speed up scheduling and distribution. If the current processing device is machine review, it means that the multimedia data is in the machine review queue. In this case, the processing priority of the multimedia data can be increased, and the multimedia data can be reviewed first to shorten the machine review time, which can also speed up scheduling and distribution. If the current processing status of the multimedia data is disabled, the multimedia data should be updated to enabled to proceed with the subsequent review and distribution process.

[0252] Step S611: The server adds first tag information to the multimedia data.

[0253] Here, the first tag information is used to characterize whether the multimedia data is high-quality or popular multimedia data.

[0254] Step S612: The server obtains the initial distribution weight of the multimedia data with the first tag information in the content distribution stage.

[0255] Step S613: The server increases the initial distribution weight according to the preset weight adjustment rules to obtain the target distribution weight;

[0256] Step S614: The server distributes the multimedia data with the first tagging information based on the target distribution weight.

[0257] Here, in the implementation of step S614, the distribution priority of multimedia data is determined by using the target distribution weight, and the multimedia data is distributed preferentially based on the distribution priority.

[0258] Step S615: The server obtains negative feedback information of multimedia data containing the first tagging information.

[0259] Here, negative feedback information includes the number of reports. Since the multimedia adjustment strategy was adjusted in step S610, and a "post-release, review-later" scheduling strategy was implemented under manual review, it is necessary to detect negative feedback and report information to control the overall risk of content distribution.

[0260] Step S616: When the number of reports of the multimedia data with the first tag information reaches a preset threshold, the server determines that the multimedia data needs to be reviewed again.

[0261] Here, when the number of reports reaches a threshold, it is determined that the multimedia data needs to be reviewed again. In practice, this can be done by directly sending it to a human reviewer for secondary confirmation, thereby reducing and controlling the risk of distributing content that is published first and then reviewed later.

[0262] Step S617: The server performs machine review and manual review on the multimedia data to obtain the review results.

[0263] Step S618: When the review result is "approved", add it to the content distribution queue.

[0264] In step S619, the server obtains the multimedia data to be distributed from the content distribution queue, determines the content consumption terminal, and sends the multimedia data in the content distribution queue to the content consumption terminal.

[0265] In the data scheduling and distribution method provided in this application embodiment, after the content generation terminal generates multimedia data to be published and sends the multimedia data to the server, the server stores the multimedia data in a multimedia database and matches the multimedia data with multiple multimedia reference data. These multiple multimedia reference data can be the latest popular multimedia data, and can also include high-quality publishing account identifiers. When it is determined based on the matching results that the multimedia data meets the accelerated distribution conditions, the current processing state of the multimedia data is obtained. Based on the current processing state, the scheduling strategy of the multimedia data is adjusted to accelerate the scheduling and distribution of the multimedia data. Thus, when the multimedia data to be published is matched with multimedia... When matching content from reference data or multimedia data published by high-quality accounts, the scheduling strategy for that multimedia data will be adjusted to prioritize its distribution. This not only enables rapid location of high-quality content but also allows high-quality content or content created by high-quality creators to be activated and distributed more quickly with shorter latency, effectively reducing content processing time. After prioritizing the distribution of multimedia data, the server will detect negative feedback information. If the number of reports in the negative feedback information reaches a threshold, the multimedia data that has reached the threshold will undergo a second review to reduce and control the risk of distributing content that is published first and then reviewed, ensuring the compliance and legality of network data.

[0266] The following will describe an exemplary application of the embodiments of this application in a real-world application scenario.

[0267] This application provides a data scheduling method based on content vector matching for accelerating the distribution of high-quality content from the network header. This method is applied to... Figure 7 The system structure shown is as follows: Figure 7 As shown, the system includes: a content generation terminal 701, an uplink / downlink content interface server 702, a content consumption terminal 703, a content database 704, a scheduling center service 705, a manual review module 706, a machine review module 707, a statistical reporting interface and analysis service 708, a vectorized matching service 709, a network header content library 710, a header content acceleration scheduling service 711, a network crawling and parsing service 712, and a recommendation distribution and content distribution exit service 713. The most core parts are the vectorized matching service 709 and the header content acceleration scheduling service 711.

[0268] The vectorized matching service 709 is primarily responsible for vectorizing the content in the information flow distribution and processing chain, including the content title, cover image, content body, and video content itself, and then performing vectorized retrieval and matching. After a match is found, based on the overall performance of this content on the network (such as distribution volume, number of comments, number of likes, number of favorites, and other interaction metrics), if a certain content in the content processing chain matches the crawled header content, the scheduling of that content in the content processing chain is accelerated.

[0269] The process of vectorizing content in the content processing chain and performing vectorized retrieval and matching includes:

[0270] Step S801: The web crawling and parsing service 712 defines crawling rules based on the content consumed by the information distribution terminal.

[0271] The crawling rules can include which websites to crawl and which apps to crawl. Apps include news apps like Toutiao, Weibo, Douyin, and Kuaishou, while websites include news portals like Sina, Sohu, Tencent, and Phoenix.com.

[0272] Step S802: The web crawling and parsing service 712 crawls data from the Internet according to the crawling rules.

[0273] Figure 8 This is a schematic diagram illustrating the capabilities of the web crawler and parsing service provided in the embodiments of this application, as shown below. Figure 8 As shown, the capabilities provided by this web crawling and parsing service include: a visual configuration platform 811, a scheduling service 812, a crawler engine 813, intelligent algorithms 814, and componentization 815, among which:

[0274] The 811 visual configuration platform allows users to configure crawling rules, crawling types, crawling strategies, and view task information.

[0275] To improve crawling efficiency, multiple crawling tasks are usually created using multi-threading and multi-service methods. Therefore, it is necessary to use the scheduling service 812 for automatic task management, which mainly includes distributed management, task scheduling, machine detection, and task keep-alive.

[0276] The 813 crawler engine supports crawling any complex app across the entire web. The prototype for the crawler engine within the crawler system can be implemented using Puppeteer. Puppeteer's Application Programming Interface (API) allows for easy browser control, enabling crawler applications, website screenshots, and website PDF generation. It primarily offers two modes: Headless and FullHead (with a user interface). The main difference between the two lies in the request headers and rendering methods (e.g., whether the website detects header information and the rendering environment). The crawler engine needs to possess certain anti-crawling capabilities (e.g., using IP pools and controlling crawling frequency) to simulate real user logins.

[0277] The intelligent algorithm 814 is used to automatically identify article titles and content, mainly including list recognition, article title recognition, article content recognition, and video link extraction.

[0278] Componentized 815, that is, using modular cracking units, supports one-click access.

[0279] When crawling content, we also acquire post-hoc data such as the number of reads, likes, and shares of these content distributions. Using this post-hoc data and account information, we can filter out top content based on criteria such as (new, trending, popular, local).

[0280] Step S803: The header content scheduling service reads the header content.

[0281] In this embodiment of the application, the web crawling and parsing service 712 stores the crawled header content in the web header content library, and the header content scheduling service reads the header content from the web header content library.

[0282] In step S804, the vectorized matching service 709 obtains each piece of content in the content processing chain and performs vectorized matching.

[0283] here, Figure 9 This is a schematic diagram illustrating the implementation function of the similarity matching service provided in the embodiments of this application, such as... Figure 9 As shown, the similarity matching service provides functions including data prevention and control (901), capability matrix (902), recall retrieval (903), and accurate decision-making (904), among which:

[0284] Data Control 901 primarily implements data verification and authentication frequency control.

[0285] Capability matrix 902 primarily enables the vectorization of content in the content processing chain, including vectorization of multiple modalities such as text content (title, body text), cover image, and video content itself. In this application embodiment, the following vector generation methods can be used for multimodal vectorization:

[0286] I. Locality Sensitive Hashing (simhash) Vector Generation: Used for deduplication of massive amounts of text. The simhash algorithm can calculate a hash value (64-bit integer). Two articles are considered similar if the distance between their simhash values ​​is less than or equal to 3. This distance is calculated using Hamming distance, which is obtained by XORing two simhash values ​​and counting the number of bits with a value of 1 (N bits). It is used for deduplication of short title texts. The simhash vector of the body text is used for initial recall, and then the article's Bidirectional Encoder Representation from Transformers (BERT) vector is used for fine-grained recall.

[0287] II. Image Vector and Text Semantic Vector Generation: This is achieved through Siamese networks, also known as conjoined networks. The conjoined structures in a Siamese network are implemented by sharing weights, making it a special type of neural network architecture and a form of supervised learning used to measure learning. This differs from a model that learns to classify its inputs, such as... Figure 10 As shown, this neural network has two inputs (input 1001 and input 1002), which are fed into two neural networks (Network1011 and Network1012). These two neural networks map the input content to new vector spaces, forming vector representations of the input content in these new spaces. The similarity between the two inputs is evaluated by calculating a loss function. The loss function is a contrastive loss function, where the text is encoded into vectors using a Siamese network. This is illustrated using the generation of semantic vectors for the main text and images. If the input is an image, then... Figure 10 The `doc` keyword is replaced with an image, and the text convolutional neural network (textCNN) is replaced with an image feature extraction network such as InceptionV3 or RestNet50. However, this is computationally slow. In engineering implementations, to ensure efficiency, the results are usually subjected to vector dimensionality reduction to obtain a 0-1 vector, and dhash is calculated simultaneously as a separate recall path.

[0288] At this point, the margin value is used to determine whether the text is repeated or similar. The input can be either body text or a title, and both long and short texts are supported.

[0289] III. Text BERT Vector Generation: BERT is essentially a two-stage Natural Language Processing (NLP) model. The first stage is called Pre-training, similar to word embedding, which trains a language model using existing unlabeled corpora. The second stage is called Fine-tuning, which uses the pre-trained language model to complete downstream NLP tasks. In this embodiment, it is mainly used to vectorize the title and body text content during the preprocessing stage, providing vector input for subsequent tasks such as deduplication and similarity calculation. This is primarily to increase the recall of deduplication, especially for semantically similar content.

[0290] IV. Video Content Vector Generation: This process involves extracting frames from video files and then constructing semantic fingerprint vectors from these frames to represent the video itself. Each video frame is a separate image. This allows the video itself to be vectorized, facilitating subsequent deduplication and calculation by the user. A frame is the smallest unit of image in animation, equivalent to a single frame on film. On the animation software's timeline, a frame is represented by a single frame or marker. A keyframe is equivalent to a key drawing in 2D animation. It refers to the frame containing the crucial action or change of a character or object. Animation between keyframes can be created by software, called transition frames or intermediate frames. Image fingerprint extraction involves obtaining keyframe images and breaking them down into small segments with overlapping content, such as 5 seconds, 1 second, or 2 seconds as the overlap unit, depending on the video length. Greater overlap increases computational load but improves the final result. There are two main methods: uniform frame extraction and variable-length frame extraction (frame extraction is mainly based on keyframes, with an uncertain interval). Each frame is equivalent to a separate image, and then it is processed according to the image vector. The vector of a video is equivalent to the concatenation of multiple image vectors. The simplest way to concatenate is to directly connect them, but this will result in a large amount of data. It is better to merge frames that are adjacent in time series, such as 5 to 10 frames.

[0291] After vector generation, vectorized matching is performed based on the Entity Framework (EF), with Fasis as the underlying implementation, building a general similarity matching service. In its implementation, the vectorized matching service 709 performs vectorized content matching between crawled external content and content already stored in the information flow platform's database. For example... Figure 9As shown, after matching and recall, a twin network of text and video is used for more accurate similarity judgment. For video content, the audio fingerprint is extracted as an auxiliary feature for secondary verification to improve accuracy, thus ensuring the accuracy of the final result. This is especially effective for lecture and training videos, where the video visuals and tasks are very similar, but the audio content differs significantly, allowing for good matching and differentiation.

[0292] If the content cannot be located, it means that this content has not yet been introduced. In this case, it can be introduced at the source through content, that is, through business development (BD), to attract self-media authors to open accounts and publish content. For content that can be matched in the content library, the head content acceleration and scheduling service is used to accelerate the scheduling, so as to ensure that this content is not backed up or filtered out in the content processing chain as much as possible.

[0293] In some embodiments, the coverage of header content on the information flow distribution platform can be measured by the daily average amount of header content crawled and the content matched in the content library, while the coverage of platform content can be quantified by detecting the link processing efficiency of this batch of content.

[0294] Step S805: The header content acceleration scheduling service performs acceleration scheduling.

[0295] The header content acceleration scheduling service adjusts the processing strategy based on the current processing status of the matched content (machine processing, manual queuing, disabled). If it's in a manual queuing state, human review is accelerated; in implementation, this can be done by publishing first and then reviewing, directly accelerating content distribution. If it's in machine processing state, machine processing is accelerated; in implementation, it can be inserted into a high-priority queue for machine processing. If it's disabled, it can be re-enabled. In this way, the header content matched by vectorization is given higher priority for review, increasing the activation rate and the amount of content activated.

[0296] In order to control the risks of content distribution as a whole, this part of the content is distributed on the device through data monitoring, especially negative feedback and reports. If the risk reaches a certain threshold, it will be sent directly to human review for secondary confirmation and processing.

[0297] To address the funnel-like issues in machine-based and human-based filtering, utilizing the header content acceleration scheduling service can improve the accuracy of header matching content processing, reduce false positives in the link, optimize similarity deduplication capabilities, improve the accuracy of similarity calculation, and reduce the proportion of header matching content affected by "link deduplication" errors.

[0298] In addition, for top-tier content on the network, accounts that publish first and then review can be verified, and the ability to publish first and then review can be adopted during scheduling to reduce the processing time in the chain; otherwise, human review can be used to enhance the authority and allocate a higher review scheduling priority to reduce the processing time in the chain. The activation of these top-tier content also adds corresponding content tags, and the recommendation side performs cold start weighted exposure to improve the overall effect and priority of top-tier content distribution, so that the content of top-quality creators can be activated and distributed faster in a shorter time delay.

[0299] In this embodiment of the application, the page views (PV) and video views (VV) and consumption time percentage of the vectorized matching header content during the overall distribution process can also be obtained to measure the effect of the header content in accelerating the distribution.

[0300] The following combination Figure 7 The functions of each service module in the multimedia processing system and the multimedia processing process are explained.

[0301] Content producers, whether PGC, UGC, MCN, or PUGC, provide text and image content or upload video content, including short videos and mini-videos, through the content generation end 701 (mobile terminal or backend interface API system). These are the main sources of content for distribution. The content generation end 701 obtains the upload server interface address through communication with the upstream and downstream content interface server 702, and then publishes the content.

[0302] The upstream and downstream content interface server 702 communicates directly with the content generation end 701. After receiving the content submitted by the content generation end 701, it obtains the content's title, publisher, summary, cover image, publication time, etc., and stores the content in the content database 705. Based on the publisher's account source, the upstream and downstream content interface server 702 sets the initial review account level through operational configuration, and can mark a portion of high-quality accounts. This is mainly closely related to operational strategies, and high-quality accounts will have a higher priority in manual review scheduling. Simultaneously, it reports the posting flow information of each account to the statistics reporting interface and analysis service 708, including posting time, content type, and stores the content tagging information provided by the self-media, such as category, tags, selected cover image, and title, as extended information in the content database.

[0303] The content consumer 703 communicates directly with the upstream and downstream content interface server 702 to obtain the index information of the accessed content. Then, it communicates with the upstream and downstream content interface server 702 and the content export service to directly consume the content. The prerequisite for consumption is to obtain the index of the content through Feeds recommendation distribution.

[0304] In some embodiments, the Feeds and user click behavior and environment reporting module collects the user's current network environment, the user's click behavior on intermediate information in the Feeds, and the exposure data of the Feeds content, and reports them to the statistics reporting interface server. If it is video content, it will also report the actual playback duration of the video, the caching time, and various interactive behaviors of the content, such as comments, forwarding, sharing, collection, liking, etc., as well as negative behaviors such as reporting and negative feedback behaviors.

[0305] Content database 704 is the core database for content. All metadata about content published by all content producers is stored in this business database. The focus is on the metadata of the content itself, such as size, cover image link, title, publication time, account author, source channel, and entry time. It also includes the content classification during manual review (including first, second, and third-level categories and tag information; for example, an article explaining Huawei phones might have a first-level category of "technology," a second-level category of "smartphones," a third-level category of "domestic phones," and the tag information "Huawei" and "Mate 30"). During the manual review process, information is read from content database 705, and the results and status of the manual review are also sent back to content database 705 for storage.

[0306] In the multimedia data processing flow provided in this application embodiment, the processing of each content in the content database 704 mainly includes machine processing and manual review processing. For example, the image and text deduplication server will load content that has been entered into the database and enabled in the past period of time (such as one week, and the validity period of video content is longer, such as 3 months) according to business needs. For content that is entered into the database repeatedly, a filter mark will be added and it will no longer be provided to the content recommendation service output to the user.

[0307] The dispatch center 705 is responsible for the entire scheduling process of content flow. It receives the content into the database through the upstream and downstream content interface servers, and then retrieves the content's metadata from the content database. The dispatch machine review module includes intercepting and filtering content that violates the law, such as pornography, gambling, and drugs, as well as handling duplicate content. For content that does not meet the criteria for posting before review, such as content that requires manual review due to security issues, the manual review system is invoked for manual review, which is the posting before review mechanism.

[0308] The manual review module 706 reads the original information of the video content itself from the content database. It is usually a complex web database-based system. Its main function is to ensure that the pushed content complies with local laws and policies, such as whether it involves pornography or gambling, and to perform an initial filtering. It receives content that needs to be manually reviewed from statistical reporting interfaces and analysis services, including content that needs to be reviewed from negative feedback and reports, to reduce and control the risk of distributing content that is released first and then reviewed. Finally, the results of the manual review are written into the content database through the scheduling center.

[0309] The machine review module 707 is used to automatically filter content including pornography, gambling, drugs, and content that violates legal boundaries, as well as to process duplicate content.

[0310] The statistical reporting interface and analysis service 708 communicates with the content consumer 702 to receive reports from the content consumer 703 regarding the current network environment, user click behavior on Feeds intermediate information, and Feeds article exposure data. For example, it receives reports of UGC short texts commenting on content, likes, reposts, favorites, and other interactive information. It also receives reports and negative feedback from the content consumer 703, and then performs real-time statistics on negative feedback and reports according to content. If the negative feedback and reports exceed a certain threshold and number of times, they are pushed to the manual review system for verification.

[0311] Vectorized Matching Service 709 is used to learn vectorized identifiers for content based on the steps and processes described above, including vectorizing text content (title, text body), cover image, and video content itself, while building a vectorized index library; it matches the crawled and parsed web header content with the currently processed content in the vectorized index library, thereby accelerating the link processing of these header contents and improving efficiency and activation rate.

[0312] The network header content library 710 is used to store the network header content crawled by the network crawling and parsing service 712, and can communicate with the header content acceleration scheduling service to provide the original network header content library as the basis for link header content matching. It is equivalent to using the information board measurement system outside the network to measure the processing and coverage of header content.

[0313] The header content acceleration scheduling service 711 communicates with the network header content library 710 and the vectorized matching service 709, and operates independently based on the processing flow and strategy described above. It adjusts the link processing scheduling strategy according to the status of the matched content, including human review acceleration and distribution priority enhancement, thereby achieving accelerated scheduling.

[0314] The web crawling and parsing service 712, based on the capabilities of the web crawler and parsing service described above, supports different terminals according to the target website from which the content to be crawled is to be obtained, and writes the crawled and parsed content into the header content library 711 according to the configured rules.

[0315] The recommendation distribution and content distribution exit service 713 is usually a group of access services deployed near the user in a geographically close area. It communicates with the recommendation distribution system to obtain the recommendation distribution results and sends the distribution results to the content consumer terminal 703, and displays them in the user's Feeds list.

[0316] This application provides a method and system for accelerating the distribution of high-quality web header content based on content vector matching. In the content acquisition stage, header content from various information platforms is crawled. Then, the vectorized header content is matched with the currently stored content in the information flow system to find header content already stored and being processed. Simultaneously, the coverage rate of this header content is detected and calculated. Then, for header content, machine and human review are prioritized and accelerated during content processing, and the accuracy of machine review is improved. This includes addressing the reasons for not enabling machine and human review filters, improving the accuracy of content matching and accelerating the review process, while reducing false positives and processing time. Finally, in the content distribution stage, high-quality header content is weighted and matched to accelerate its cold start. Through the embodiments of this application, it is possible to significantly increase the activation volume of high-quality online top content in the recommendation pool while reducing the investment of human review; at the same time, it is possible to effectively detect and quantify the coverage and content processing efficiency of online top information content on the platform, including coverage and processing timeliness; it is also possible to enable the content of top high-quality creators to be activated and distributed faster with shorter latency, effectively reducing content processing time and playing a huge role in promoting the ecosystem of information flow content creation and distribution.

[0317] The following description continues to illustrate the exemplary structure of the data scheduling and distribution device 455 provided in the embodiments of this application as a software module. In some embodiments, such as Figure 3 As shown, the software modules stored in the data scheduling and distribution device 455 in the memory 440 may include:

[0318] The first acquisition module 4551 is used to acquire multimedia data to be distributed and to acquire multiple multimedia reference data crawled from the network.

[0319] The similarity matching module 4552 is used to perform similarity matching between the multimedia data and the plurality of multimedia reference data to obtain a matching result;

[0320] The second acquisition module 4553 is used to acquire the current processing status of the multimedia data when it is determined from the matching result that the multimedia data meets the accelerated distribution conditions.

[0321] The strategy adjustment module 4554 is used to adjust the scheduling strategy of the multimedia data based on the current processing state in order to accelerate the scheduling and distribution of the multimedia data.

[0322] In some embodiments, the similarity matching module is further configured to:

[0323] Obtain the attribute information of the multimedia data and the attribute information of the plurality of multimedia reference data, wherein the attribute information includes title information;

[0324] The attribute information and multimedia data of the multimedia data are vectorized to obtain the first title vector and the first multimedia vector of the multimedia data.

[0325] The attribute information of the plurality of multimedia reference data and the multimedia reference data are vectorized to obtain the second title vector and the second multimedia vector of the plurality of multimedia reference data.

[0326] Determine the title similarity between the first title vector and each of the second title vectors, and the multimedia similarity between the first multimedia vector and each of the second multimedia vectors, respectively;

[0327] The matching results are determined based on the similarity of each title and the similarity of each multimedia element.

[0328] In some embodiments, the similarity matching module is further configured to:

[0329] Based on the title similarity and multimedia similarity, determine whether the target multimedia reference data exists among the multiple multimedia reference data;

[0330] The title similarity between the target multimedia reference data and the multimedia data is less than a first similarity threshold, and / or the multimedia similarity between the target multimedia reference data and the multimedia data is less than a second similarity threshold;

[0331] When the target multimedia reference data exists among the plurality of multimedia reference data, the matching result is determined to be a successful match.

[0332] In some embodiments, the device further includes:

[0333] The first determining module is used to input the first title vector and the first multimedia vector of the multimedia data and the second title vector and the second multimedia vector of the target multimedia reference data into a trained neural network model when the matching result is a successful match, in order to determine the target similarity between the multimedia data and the target multimedia reference data.

[0334] The third acquisition module is used to acquire the first audio data of the multimedia data and the second audio data of the target multimedia reference data when the target similarity is greater than the third similarity threshold.

[0335] The second determining module is used to determine the audio similarity between the first audio data and the second audio data;

[0336] The third determining module is used to determine that the multimedia data meets the preset accelerated distribution conditions when the audio similarity is greater than the fourth similarity threshold.

[0337] In some embodiments, the similarity matching module is further configured to:

[0338] Obtain the first publishing account identifier of the multimedia data and the second publishing account identifier of the plurality of multimedia reference data;

[0339] Determine whether a second publishing account identifier exists that is identical to the first publishing account identifier;

[0340] If a second publishing account with the same identifier as the first publishing account exists, the matching result is determined to be a successful match;

[0341] Correspondingly, the device also includes:

[0342] The fourth determining module is used to determine that the multimedia data meets the preset accelerated distribution conditions when the matching result is a successful match.

[0343] In some embodiments, the current processing state includes a manual review state, a machine review state, and a disabled state. Correspondingly, the policy adjustment module is further configured to:

[0344] When the current processing status is manual review, the distribution strategy of the multimedia data will be adjusted to a first-release-later-review strategy.

[0345] When the current processing status is machine review status, the processing priority of the multimedia data is increased;

[0346] When the current processing state is disabled, the processing state of the multimedia data is adjusted to the enabled state.

[0347] In some embodiments, the device further includes:

[0348] A tagging module is used to add first tagging information to the multimedia data when the multimedia data meets the accelerated distribution conditions;

[0349] The fourth acquisition module is used to acquire the initial distribution weight of multimedia data with the first tag information during the content distribution stage;

[0350] The weighting module is used to increase the initial distribution weight according to a preset weighting adjustment rule to obtain the target distribution weight;

[0351] The content distribution module is used to distribute the multimedia data with the first tag information based on the target distribution weight.

[0352] In some embodiments, the device further includes:

[0353] The fifth acquisition module is used to acquire the preset target website and crawling strategy;

[0354] The data crawling module is used to crawl multiple candidate multimedia data of a preset duration from the target website using the crawling strategy.

[0355] The sixth acquisition module is used to acquire multiple interactive information and multiple publishing account identifiers of the multiple candidate multimedia data, wherein the interactive information includes: number of views, number of likes, and number of shares;

[0356] The fifth determining module is used to determine multimedia reference data from the multiple candidate multimedia data based on the multiple interactive information and the multiple publishing account identifiers.

[0357] In some embodiments, the device further includes:

[0358] The seventh acquisition module is used to acquire negative feedback information of multimedia data with the first tagging information, wherein the negative feedback information includes the number of reports;

[0359] The review module is used to determine that the multimedia data with the first tag information needs to be reviewed again when the number of reports reaches a preset threshold.

[0360] In some embodiments, the device further includes:

[0361] The sixth determining module is used to determine the first total number of multimedia reference data crawled within a preset time period;

[0362] The seventh determining module is used to determine the second total number of multimedia data that meet the accelerated distribution conditions within the preset time period;

[0363] The eighth determining module is used to determine the coverage of the multimedia reference data based on the first total and the second total.

[0364] The ninth determining module is used to determine the target publishing account identifier based on the multimedia reference data when the coverage rate is lower than a preset coverage rate threshold.

[0365] The sending module is used to send invitation information to the terminal corresponding to the target publishing account identifier, so as to invite the terminal to publish multimedia data.

[0366] In some embodiments, the device further includes:

[0367] The eighth acquisition module is used to acquire the multimedia data to be published uploaded by the content generation terminal, and to acquire the publishing account identifier of the multimedia data to be published.

[0368] The tenth determining module is used to determine the review level of the publishing account identifier based on the historical multimedia data when the historical multimedia data corresponding to the publishing account identifier can be obtained;

[0369] The data update module is used to add the publishing account identifier to the multimedia reference data when the review level is greater than a preset level threshold.

[0370] In some embodiments, the tenth determining module is further configured to:

[0371] Obtain interactive information from the historical multimedia data, including the number of views, likes, and shares.

[0372] The review level of the publishing account identifier is determined based on the number of views, likes, and shares.

[0373] It should be noted that the descriptions of the above data scheduling and distribution device embodiments are similar to the descriptions of the methods described above, and have the same beneficial effects as the method embodiments. For technical details not disclosed in the data scheduling and distribution device embodiments of this application, those skilled in the art should refer to the descriptions of the method embodiments of this application for understanding.

[0374] This application provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the data scheduling and distribution method described in this application.

[0375] This application provides a computer-readable storage medium storing executable instructions. When these executable instructions are executed by a processor, they cause the processor to perform the method provided in this application, for example... Figure 4 , Figure 5 and Figure 6 The method shown.

[0376] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0377] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0378] As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborating files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0379] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0380] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. A method for scheduling and distributing multimedia data, characterized in that, include: Acquire the multimedia data to be distributed, and obtain multiple multimedia reference data crawled from the network; The multimedia data and the attribute information in the multimedia data including title information, as well as the multiple multimedia reference data, are vectorized to obtain a first title vector and a first multimedia vector, and a second title vector and a second multimedia vector. When it is determined that target multimedia reference data exists in the plurality of multimedia reference data based on the title similarity between the first title vector and each of the second title vectors, and the multimedia similarity between the first multimedia vector and each of the second multimedia vectors, the matching result is determined to be a successful match, and the first title vector, the first multimedia vector, and the second title vector and the second multimedia vector of the target multimedia reference data are input into the trained neural network model to determine the target similarity between the multimedia data and the target multimedia reference data; When the target similarity is greater than the third similarity threshold, the first audio data of the multimedia data and the second audio data of the target multimedia reference data are obtained; Determine the audio similarity between the first audio data and the second audio data; when the audio similarity is greater than a fourth similarity threshold, determine that the multimedia data meets the preset accelerated distribution conditions; When it is determined from the matching results that the multimedia data meets the accelerated distribution conditions, the current processing status of the multimedia data is obtained; Based on the current processing state, the scheduling strategy for the multimedia data is adjusted to accelerate the scheduling and distribution of the multimedia data.

2. The method according to claim 1, characterized in that, Before performing vectorization processing on the multimedia data and the attribute information in the multimedia data including title information, as well as the plurality of multimedia reference data, to obtain a first title vector and a first multimedia vector, and a second title vector and a second multimedia vector, the method further includes: Obtain the attribute information of the multimedia data, and obtain the attribute information of the multiple multimedia reference data.

3. The method according to claim 2, characterized in that, The title similarity between the target multimedia reference data and the multimedia data is less than a first similarity threshold, and / or the multimedia similarity between the target multimedia reference data and the multimedia data is less than a second similarity threshold.

4. The method according to claim 1, characterized in that, The method further includes: Obtain the first publishing account identifier of the multimedia data and the second publishing account identifier of the plurality of multimedia reference data; Determine whether a second publishing account identifier exists that is identical to the first publishing account identifier; If a second publishing account with the same identifier as the first publishing account exists, the matching result is determined to be a successful match; When the matching result is successful, it is determined that the multimedia data meets the preset accelerated distribution conditions.

5. The method according to claim 1, characterized in that, The current processing state includes manual review, machine review, and disabled state. Correspondingly, adjusting the distribution strategy of the multimedia data based on the current processing state includes: When the current processing status is manual review, the distribution strategy of the multimedia data will be adjusted to a first-release-later-review strategy. When the current processing status is machine review status, the processing priority of the multimedia data is increased; When the current processing state is disabled, the processing state of the multimedia data is adjusted to the enabled state.

6. The method according to claim 5, characterized in that, The method further includes: When the multimedia data meets the accelerated distribution conditions, a first tag information is added to the multimedia data; Obtain the initial distribution weight of multimedia data with the first tagging information during the content distribution stage; The initial distribution weight is increased according to a preset weight adjustment rule to obtain the target distribution weight; Based on the target distribution weight, the multimedia data with the first tag information is distributed.

7. The method according to claim 1, characterized in that, The method further includes: Obtain the preset target website and crawling strategy; The crawling strategy is used to crawl multiple candidate multimedia data of a preset duration from the target website; Acquire multiple interactive information and multiple publishing account identifiers of the multiple candidate multimedia data, wherein the interactive information includes: number of views, number of likes, and number of shares; Based on the multiple interactive information and the multiple publishing account identifiers, multimedia reference data is determined from the multiple candidate multimedia data.

8. The method according to claim 6, characterized in that, The method further includes: Obtain negative feedback information of multimedia data with the first tagging information, the negative feedback information including the number of reports; When the number of reports of the multimedia data with the first tag information reaches a preset threshold, it is determined that the multimedia data needs to be reviewed again.

9. The method according to any one of claims 1 to 6, characterized in that, The method further includes: Determine the first total number of multimedia reference data crawled within the preset time period; Determine a second total number of multimedia data that meet the accelerated distribution conditions within the preset time period; The coverage of the multimedia reference data is determined based on the first total and the second total. When the coverage rate is lower than a preset coverage rate threshold, the target publishing account identifier is determined based on the multimedia reference data; An invitation message is sent to the terminal corresponding to the target publishing account identifier to invite the terminal to publish multimedia data.

10. The method according to any one of claims 1 to 6, characterized in that, The method further includes: Obtain the multimedia data to be published uploaded by the content generation terminal, and obtain the publishing account identifier of the multimedia data to be published; When the historical multimedia data corresponding to the publishing account identifier can be obtained, the review level of the publishing account identifier is determined based on the historical multimedia data; When the review level is greater than the preset level threshold, the publishing account identifier is added to the multimedia reference data.

11. The method according to claim 10, characterized in that, The process of determining the review level of the publishing account identifier based on the historical multimedia data includes: Obtain interactive information from the historical multimedia data, including the number of views, likes, and shares. The review level of the publishing account identifier is determined based on the number of views, likes, and shares.

12. A data scheduling and distribution device, characterized in that, include: The first acquisition module is used to acquire the multimedia data to be distributed and to acquire multiple multimedia reference data crawled from the network. The similarity matching module is used to perform vectorization processing on the multimedia data and the attribute information in the multimedia data including title information, as well as the multiple multimedia reference data, to obtain a first title vector and a first multimedia vector, and a second title vector and a second multimedia vector; when it is determined that there is target multimedia reference data in the multiple multimedia reference data based on the title similarity between the first title vector and each of the second title vectors, and the multimedia similarity between the first multimedia vector and each of the second multimedia vectors, the matching result is determined to be a successful match; The first determining module is used to input the first title vector, the first multimedia vector, and the second title vector and the second multimedia vector of the target multimedia reference data into a trained neural network model when the matching result is a successful match, in order to determine the target similarity between the multimedia data and the target multimedia reference data; The third acquisition module is used to acquire the first audio data of the multimedia data and the second audio data of the target multimedia reference data when the target similarity is greater than the third similarity threshold. The second determining module is used to determine the audio similarity between the first audio data and the second audio data; The third determining module is used to determine that the multimedia data meets the preset accelerated distribution conditions when the audio similarity is greater than the fourth similarity threshold. The second acquisition module is used to acquire the current processing status of the multimedia data when it is determined from the matching result that the multimedia data meets the accelerated distribution conditions. The strategy adjustment module is used to adjust the scheduling strategy of the multimedia data based on the current processing state, so as to accelerate the scheduling and distribution of the multimedia data.

13. The apparatus according to claim 12, characterized in that, The similarity matching module is also used to obtain attribute information of the multimedia data and attribute information of the multiple multimedia reference data.

14. The apparatus according to claim 13, characterized in that, The title similarity between the target multimedia reference data and the multimedia data is less than a first similarity threshold, and / or the multimedia similarity between the target multimedia reference data and the multimedia data is less than a second similarity threshold.

15. The apparatus according to claim 12, characterized in that, The similarity matching module is further configured to obtain the first publishing account identifier of the multimedia data and the second publishing account identifier of the plurality of multimedia reference data; and determine whether there is a second publishing account identifier that is the same as the first publishing account identifier; If a second publishing account with the same identifier as the first publishing account exists, the matching result is determined to be a successful match; Correspondingly, the device also includes: The fourth determining module is used to determine that the multimedia data meets the preset accelerated distribution conditions when the matching result is a successful match.

16. The apparatus according to claim 12, characterized in that, The current processing status includes manual review status, machine review status, and disabled status; Correspondingly, the strategy adjustment module is also used to adjust the distribution strategy of the multimedia data to a first-release-then-review strategy when the current processing state is manual review; to increase the processing priority of the multimedia data when the current processing state is machine review; and to adjust the processing state of the multimedia data to an enabled state when the current processing state is disabled.

17. The apparatus according to claim 16, characterized in that, The device further includes: A tagging module is used to add first tagging information to the multimedia data when the multimedia data meets the accelerated distribution conditions; The fourth acquisition module is used to acquire the initial distribution weight of multimedia data with the first tag information during the content distribution stage; The weighting module is used to increase the initial distribution weight according to a preset weighting adjustment rule to obtain the target distribution weight; The content distribution module is used to distribute the multimedia data with the first tag information based on the target distribution weight.

18. The apparatus according to claim 12, characterized in that, The device further includes: The fifth acquisition module is used to acquire the preset target website and crawling strategy; The data crawling module is used to crawl multiple candidate multimedia data of a preset duration from the target website using the crawling strategy. The sixth acquisition module is used to acquire multiple interactive information and multiple publishing account identifiers of the multiple candidate multimedia data, wherein the interactive information includes: number of views, number of likes, and number of shares; The fifth determining module is used to determine multimedia reference data from the multiple candidate multimedia data based on the multiple interactive information and the multiple publishing account identifiers.

19. The apparatus according to claim 17, characterized in that, The device further includes: The seventh acquisition module is used to acquire negative feedback information of multimedia data with the first tagging information, wherein the negative feedback information includes the number of reports; The review module is used to determine that the multimedia data with the first tag information needs to be reviewed again when the number of reports reaches a preset threshold.

20. The apparatus according to any one of claims 12 to 17, characterized in that, The device further includes: The sixth determining module is used to determine the first total number of multimedia reference data crawled within a preset time period; The seventh determining module is used to determine the second total number of multimedia data that meet the accelerated distribution conditions within the preset time period; The eighth determining module is used to determine the coverage of the multimedia reference data based on the first total and the second total. The ninth determining module is used to determine the target publishing account identifier based on the multimedia reference data when the coverage rate is lower than a preset coverage rate threshold. The sending module is used to send invitation information to the terminal corresponding to the target publishing account identifier, so as to invite the terminal to publish multimedia data.

21. The apparatus according to any one of claims 12 to 17, characterized in that, The device further includes: The eighth acquisition module is used to acquire the multimedia data to be published uploaded by the content generation terminal, and to acquire the publishing account identifier of the multimedia data to be published. The tenth determining module is used to determine the review level of the publishing account identifier based on the historical multimedia data when the historical multimedia data corresponding to the publishing account identifier can be obtained; The data update module is used to add the publishing account identifier to the multimedia reference data when the review level is greater than a preset level threshold.

22. The apparatus according to claim 21, characterized in that, The tenth determining module is further configured to acquire interactive information of the historical multimedia data, including the number of views, the number of likes, and the number of shares; and to determine the review level of the publishing account identifier based on the number of views, the number of likes, and the number of shares.

23. A data scheduling and distribution device, characterized in that, include: Memory, used to store executable instructions; A processor, when executing the executable instructions stored in the memory, implements the method according to any one of claims 1 to 11.

24. A computer-readable storage medium, characterized in that, It stores executable instructions for implementing the method according to any one of claims 1 to 11 when executed by a processor.

25. A computer program, characterized in that, The computer program includes computer instructions stored in a computer-readable storage medium, the processor reads the computer instructions from the computer-readable storage medium and executes the computer instructions to implement the method according to any one of claims 1 to 11.

Images