Data loading method and data loading system

Abstract

The embodiment of the invention provides a data loading method and a data loading system. The data loading method comprises the steps of obtaining user data for a target multimedia application; based on the user data, determining an association relationship between multimedia data units in the target multimedia application; predicting and generating target multimedia data based on the association relationship; and preloading the target multimedia data to a memory. The association relationship between multimedia data units is modeled based on user data, the possibility of user selection is quantified, target multimedia data is generated, and target voice resources of a specific language version are preloaded, so that the memory occupation is remarkably reduced, immediate calling can be realized during switching, the waiting delay of immediate loading is thoroughly eliminated, and the user experience is improved. The system operation stability is guaranteed, meanwhile, real-time response of language switching is achieved, the user experience in a multi-language interaction scene is practically improved, and therefore the method can be widely applied to the field of digital culture product manufacturing software.

Description

Technical Field

[0001] This specification relates to the field of digital cultural product technology, and in particular to a data loading method and a data loading system. Background Technology

[0002] Currently, multilingual interactive multimedia applications are developing rapidly, and voice resources, as the core interactive medium, directly affect the user's audiovisual experience. Therefore, existing multilingual interactive multimedia applications require voice resources to support multiple language versions and enable flexible switching during user operation, which places higher demands on resource loading mechanisms, memory usage control, and real-time scheduling.

[0003] Currently, multilingual voice resource switching mainly employs two modes: full loading mode, which loads all voice resources for a single language when the application starts, requiring an application restart to reload the target language resources when switching languages; and instant loading mode, which dynamically downloads or reads the target language voice resources when switching languages, but requires waiting for the resources to be fully loaded before use. Full loading mode keeps all language voice resources resident, leading to a surge in memory usage, easily causing application lag, slow response, or even memory overflow, reducing operational stability, and each load can only be for a single language. Instant loading, on the other hand, is highly dependent on device performance and has a long loading time delay, making it difficult to meet users' real-time access needs.

[0004] Therefore, existing multilingual voice resource management solutions have significant shortcomings in terms of switching efficiency, memory optimization, and user experience, and there is an urgent need for a dynamic and efficient resource management solution. Summary of the Invention

[0005] In view of this, embodiments of this specification provide a data loading method. One or more embodiments of this specification also relate to a data loading system, a computing device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0006] According to a first aspect of the embodiments of this specification, a data loading method is provided, comprising: Acquire user data for the target multimedia application; Based on user data, determine the relationships between multimedia data units in the target multimedia application; Based on the correlation, predict and generate target multimedia data; Preload the target multimedia data into memory.

[0007] According to a second aspect of the embodiments of this specification, a data loading system is provided, including a data prediction module and a streaming loading module; The data prediction module is used to acquire user data for the target multimedia application; based on the user data, determine the correlation between multimedia data units in the target multimedia application; and based on the correlation, predict and generate target multimedia data. The streaming module is used to preload target multimedia data into memory.

[0008] According to a third aspect of the embodiments of this specification, a computing device is provided, comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the data loading method described above.

[0009] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores a computer program / instructions that, when executed by a processor, implement the steps of the data loading method described above.

[0010] According to a fifth aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the data loading method described above.

[0011] The data loading method provided in one or more embodiments of this specification acquires user data for a target multimedia application; determines the association relationships between multimedia data units in the target multimedia application based on the user data; predicts and generates target multimedia data based on the association relationships; and preloads the target multimedia data into memory. By acquiring user data for the target multimedia application, the object of multimedia data generation is identified, and the association relationships between multimedia data units in the target multimedia application are determined based on the user data. This allows for modeling the association relationships between multimedia data units based on user data, quantifying the probability of user selection, generating target multimedia data, and preloading target speech resources in a specific language version. This significantly reduces memory usage and avoids stuttering and overflow caused by full loading. The target data is loaded before the user triggers a switch, so it can be called immediately during the switch, completely eliminating the waiting delay of instant loading. While ensuring the stability of system operation, it achieves real-time response to language switching and effectively improves the user experience in multilingual interaction scenarios. Attached Figure Description

[0012] Figure 1 An architecture diagram of a data loading system provided in one embodiment of this specification; Figure 2 A flowchart illustrating the processing steps of a data loading method provided in one embodiment of this specification; Figure 3A flowchart illustrating a data loading method provided in one embodiment of this specification; Figure 4 This is a schematic diagram of a data storage structure provided in one embodiment of this specification; Figure 5 This is a schematic diagram of a data switching process provided in one embodiment of this specification; Figure 6 A flowchart illustrating a loading process provided for one embodiment of this specification; Figure 7 A schematic diagram of the system structure of a data loading system provided in one embodiment of this specification; Figure 8 This is a structural block diagram of a computing device provided for one embodiment of this specification. Detailed Implementation

[0013] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0014] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items. The term “at least one” in one or more embodiments of this application means “one or more,” and “a plurality of” means “two or more.” The term “comprising” is an open-ended description and should be understood as “including but not limiting,” and may include other content in addition to what has been described.

[0015] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0016] Furthermore, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0017] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0018] Decision trees are intuitive machine learning models that classify or predict data using a tree-like structure, simulating the human decision-making process. Their core construction is recursive partitioning: starting from the root node, the algorithm selects the best features based on criteria such as information gain to ask questions, dividing the data into cleaner subsets and forming branches. This process is repeated at each child node until the data cannot be further distinguished or the stopping condition is met, and finally, the leaf node at the end provides a clear conclusion. However, a single tree is prone to overfitting due to excessive focus on training details; therefore, pruning or ensemble methods are often used to improve its generalization ability.

[0019] It should be noted that the data loading method provided in this manual can be applied to multiple industries or scenarios, such as virtual reality processing software, digital cultural product production software, digital cultural creative software, digital cultural creative design, education, news, cultural content industry software, digital publishing software, digital music development and production, and digital mobile multimedia development and production.

[0020] This specification provides a data loading method, and also relates to a data loading system, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0021] Considering the large number of parameters in the machine learning model used by the data loading method and the limited computing resources on the client side, the data loading method proposed in the embodiments of this specification can be applied to, for example... Figure 1 The data loading system shown is not limited to this. See also Figure 1 , Figure 1 This specification provides an architecture diagram of a data loading system according to one embodiment. The data loading system may include a client 100 and a server 120. Client 100 is used to obtain user data for the target multimedia application and send it to server 120; Server 120 is used to determine the relationship between multimedia data units in the target multimedia application based on user data; predict and generate target multimedia data based on the relationship; and send the target multimedia data to client 100.

[0022] Client 100 is also used to preload target multimedia data into memory.

[0023] like Figure 1 As shown, the machine learning model is deployed on server 120. Server 120 can connect to one or more clients 100 via a local area network (LAN), wide area network (WAN), internet connection, or other types of data network. Clients 100 may include, but are not limited to, smartphones, tablets, laptops, PDAs, personal computers, smart home devices, and in-vehicle devices. Clients 100 can also interact with users through a graphical user interface to invoke the machine learning model, thereby implementing the data loading method provided in the embodiments of this specification.

[0024] Client 100 can be a browser, application (APP), or web application such as HyperText Markup Language 5 (H5) application, or a lightweight application (also known as a mini-program), or cloud application, etc. Client 100 can be developed based on the software development kit (SDK) of the corresponding service provided by server 120, such as based on the Real Time Communication (RTC) SDK. Client 100 can be deployed in electronic devices and depends on the device to run or on certain APPs on the device. Electronic devices may have a display screen and support information browsing, such as personal mobile terminals such as mobile phones, tablets, and personal computers. Various other types of applications can also be configured in electronic devices, such as human-computer interaction applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0025] Server 120 may include servers providing various services, such as servers providing communication services for multiple clients, servers supporting backend training of machine learning models used on clients, and servers processing data sent by clients. It should be noted that server 120 can be implemented as a distributed server cluster composed of multiple servers, or as a single server. The server can also be a server in a distributed system, or a server integrated with blockchain. The server can also be 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, content delivery networks (CDNs), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0026] It is worth noting that the data loading method provided in the embodiments of this specification is generally executed by the server. However, in other embodiments of this specification, the client may also have similar functionality to the server, thereby executing the data loading method provided in the embodiments of this specification. In other embodiments, the data loading method provided in the embodiments of this specification may also be executed jointly by the client and the server.

[0027] The following combination Figure 2 The data loading method provided in this manual will be used for explanation. Figure 2 A flowchart illustrating the processing steps of a data loading method provided in one embodiment of this specification includes the following steps: Step 202: Obtain user data for the target multimedia application.

[0028] Target multimedia applications are software applications designed for end users, supporting multilingual content presentation and dynamic interaction. Their function is to organize, schedule, and present multimedia data units, such as voice prompts, character dialogue, operational feedback sounds, subtitles, and animations, on demand, based on user behavior and the context of the content, to achieve a natural, coherent, and personalized audiovisual interactive experience. Target multimedia applications are architecturally scalable and possess multilingual resource management capabilities, such as multilingual children's educational apps in smart TVs, in-vehicle voice navigation systems, and interactive video games, especially story-driven video games.

[0029] User data is a collection of information or data that reflects a user's individual characteristics, operating habits, and real-time context. User data can represent various types of information that characterize a user's individual preferences, behavioral habits, and real-time interaction status. It can include current multimedia data, multimedia data accessed by the user in the past, records of the user's current access behavior / historical access behavior, access behavior records of other users with similar characteristics (access behavior records can include access scene data, user's language preferences, and user's interaction behavior during the current multimedia data access process), scene state data, multimedia application data, and resource metadata maintained by multimedia applications, such as one or more of the following: language identifier, version number, and call frequency.

[0030] The current multimedia data refers to the media content that is currently playing or being interacted with. This can be static or dynamic data such as audio, video, images, animations, subtitles, and synthesized speech streams. It could be micro-lesson videos being studied by students on online education platforms, vertical videos that users swipe and linger on in short video apps, real-time voice prompts and map animations broadcast in in-vehicle navigation systems, audio stories played by smart speakers in response to commands, and so on. In this application, its core function is to serve as a prediction starting point, triggering the prediction and preloading process for the next potentially requested target multimedia data, thereby achieving seamless playback and low-latency response.

[0031] Current access behavior records are log data reflecting user actions on multimedia content and their environmental parameters. These actions can be performed by the user on multimedia content and related scenes within the current session or short-term context, including but not limited to play / pause / fast forward / jump nodes, dwell time, interactive actions (such as clicks, favorites, and shares), scene operations (movement, scene transitions), and language selection behaviors. Current access behavior serves as the source of dynamic behavioral features for the decision tree model, working in conjunction with content association probabilities to form a predictive basis. It expresses user intent patterns, assists in accurately determining the target decision branch, and thus locks onto the target multimedia data, ensuring that pre-loaded behavior closely matches the user's real, immediate, and contextualized operational intent.

[0032] This step acquires user data to anchor the prediction starting point and characterize the user's immediate intent, eliminating redundant calculations of unrelated resources. It also transforms subjective and discrete human-computer interaction behaviors into machine-recognizable and learnable numerical signals, thereby providing real-time and scenario-specific prediction basis for subsequent prediction processes. This is the data foundation for achieving low-latency, high-hit-rate streaming preloading.

[0033] Step 204: Based on user data, determine the relationships between multimedia data units in the target multimedia application.

[0034] Association relationships are the co-occurrence, temporal, or semantic connections between different multimedia data units based on actual user behavior; in other words, they are the structured mapping of user interaction patterns at the resource invocation level. Association relationships can include preset rules or static configurations, such as "after a user clicks the language button on the Chinese settings page, there is an 87% probability of calling an English prompt," or dynamic predictive associations derived through dynamic mining of user data, such as implementation deductions based on decision tree models or collaborative filtering recommendation models, possessing strong personalization and scenario adaptability. Association relationships can transform abstract user intent into quantifiable content and logical connections, enabling the system to accurately deduce "the most likely multimedia data to be invoked next" from the currently used multimedia data, thus providing a deterministic basis for predicting and generating target multimedia data.

[0035] Content feature analysis can be used to determine associations. Content feature analysis is the process of calculating the similarity and / or association between current multimedia data and other multimedia data units, and / or the calculation can include multiple dimensions such as semantics, structure, temporal sequence, theme, sentiment, acoustic attributes, and contextual functions. The goal of content feature analysis is to identify whether the current multimedia data expresses the same or highly related informational intent with other multimedia data units. The analysis results are output as content association probabilities, characterizing the potential rationality of subsequent user access to candidate data, providing a semantic basis for subsequent behavior-based dynamic predictions; that is, associations can be content association probabilities.

[0036] Content association probability is a numerical indicator that quantifies the potential association strength between two or more multimedia data. It can be obtained based on the semantic, temporal, user behavior, or contextual associations between multimedia data, reflecting the likelihood that the target multimedia data will be accessed subsequently relative to the current multimedia data. Content association probability can be used as one of the features in the determination of the target decision branch, thereby driving the system to accurately select high-value target multimedia data that needs to be preloaded from the candidate set, making the preloading strategy adaptable to different scenarios.

[0037] There are several ways to perform content feature analysis on current multimedia data and candidate multimedia data to obtain the content association probability. One possible way is to perform content feature analysis on current multimedia data and candidate multimedia data based on a large language model to obtain the content association probability.

[0038] Another possible approach is to use a rule engine to perform content feature analysis on the current multimedia data and candidate multimedia data to obtain the content association probability.

[0039] This step analyzes the relationships between multimedia data units within user data, quantifying these relationships to effectively build the semantic foundation for preloading decisions. This allows for the accurate identification of potential targets that are highly relevant to the user's current content in terms of information intent. It ensures that the system prioritizes preloading resources that are most likely to be accessed by the user next, thereby optimizing memory utilization and laying the groundwork for a seamless, low-latency experience.

[0040] Step 206: Based on the association, predict and generate target multimedia data.

[0041] Predictive generation is the process of generating target multimedia data based on the established relationships between multimedia data units and current user data. Under the constraints of these relationships, predictive generation identifies multimedia resources that are strongly coupled with the current context, highly relevant to the user's intent, and have a high probability of being invoked. This provides reasonable targets for subsequent preloading operations, ensuring immediate response to interactive actions such as multilingual switching, instant switching, and delayed switching without loading redundant resources.

[0042] The target multimedia data refers to multimedia resources that are more likely to be called in the current or upcoming user interaction than a preset or average value. The target multimedia data can be used as the direct object of the preloading operation, so that the system can load the most likely multimedia data, such as voice prompts, interface feedback sounds, subtitle fragments or sound effects resources, into memory in advance before the user triggers a switch or enters a new scene. This avoids the memory pressure caused by full resident, and avoids the response delay caused by real-time loading.

[0043] There are several ways to predict the generation of target multimedia data based on the association relationship. One possible way is to determine the access prediction probability of multiple candidate multimedia data based on the association relationship, and then determine the target multimedia data based on the access prediction probability.

[0044] Another possible approach is to determine the access prediction probability of candidate multimedia data based on the correlation relationship, correct the access prediction probability based on the current access behavior record, and determine the target multimedia data based on the correction result.

[0045] It should be noted that there can be one or more target multimedia data. When there are multiple target multimedia data, a streaming loading method can be used to load the target multimedia data one by one.

[0046] The revised version dynamically incorporates access behavior records into a user-based personalized probability calibration process, targeting content association probabilities.

[0047] The correction process involves using user intent recorded in access behavior data to reveal the differences and commonalities between the user's actual intent and the relationship between the data content and the prediction. This correction process elevates prediction from a simple content matching to a comprehensive judgment of both content and behavior, truly achieving user-centric dynamic prediction. For example, if two videos are semantically highly similar, but the user repeatedly skips similar content during their access to current multimedia data, the behavioral data will proactively lower the prediction probability. Conversely, even if the content relevance is moderate, if the user has just searched for related keywords or related dialogues, the system will significantly increase the prediction probability of their access.

[0048] Access prediction probability is a quantitative value representing the likelihood of a user actively triggering access to a candidate multimedia data. It inherits the content logic relationships revealed by content analysis and incorporates the user's immediate interests, operating habits, contextual constraints, and potential preference shifts through user behavior correction. Access prediction probability directly drives preloading decisions: candidates with high prediction probabilities are prioritized for entry into the cache queue, thus completing resource preparation before user action occurs and achieving a low-latency experience with instant loading.

[0049] There are several ways to obtain the access prediction probability by correcting the content association probability based on the current access behavior record. One possible way is to use a machine learning model to correct the content association probability based on the current access behavior record.

[0050] Another possible approach is to construct behavioral record features based on current access behavior records, construct content association features based on content association probabilities, and model the current access behavior records, content association probabilities, current multimedia data, and candidate multimedia data through an attention mechanism to obtain access prediction probabilities.

[0051] There are several ways to generate target multimedia data. It can be generated based on multiple multimedia data in a multimedia database, or it can be generated based on text in a multimedia database. Alternatively, it can be generated based on user data, a large language model, or a text generation model. The specific method to be selected depends on the actual situation, and this manual does not impose any restrictions on it.

[0052] This step combines the correlation between the user's current operation and historical behavior to enable the system to filter out specific resources from all multimedia resources that have a significantly higher probability of being called than the average in the current context, thereby outputting the target multimedia data. It limits the scope of preloaded objects to ensure that only resources that are truly likely to be used are loaded. This avoids memory redundancy and runtime burden caused by full loading, as well as invalid occupation caused by indiscriminate caching or random prefetching. At the same time, it provides a precise premise for subsequent zero-latency calls. Ultimately, without increasing system overhead, it supports the real-time performance and stability of interactive actions such as multi-language switching, instant response, and scene adaptation.

[0053] Step 208: Preload the target multimedia data into memory.

[0054] Memory is a volatile runtime storage area in a terminal device that is dynamically allocated, temporarily stored, and accessed at high speed by applications. The core characteristics of memory are low-latency read / write, data loss upon power failure, limited capacity, and the need for active management. Memory, as the physical execution carrier of the preloading strategy, is the key physical layer support for achieving low-latency, high-response, and adaptive data loading in this application.

[0055] Preloading is a technology mechanism that proactively saves candidate multimedia data with a high probability of being accessed in advance to local memory or edge nodes. The objects to be preloaded are determined by the system based on intelligent prediction of the user's intent before the user actually triggers the access operation. The goal of preloading is to eliminate the latency caused by network requests, resource loading, parsing, and initial rendering, so that the user can get an instant response the moment they initiate an operation. For example, the picture is ready when you slide to the next frame, the video responds within seconds after you click the title, and the answer text and image are rendered after you issue a voice command. Preloading is a precise pre-preparation based on the access prediction probability, taking into account both improved user experience and efficient coordination of system resources such as bandwidth, storage, and power consumption.

[0056] The data loading method provided in this specification acquires user data for a target multimedia application, clarifies the object of multimedia data generation, and determines the relationship between multimedia data units in the target multimedia application based on the user data. This allows for the modeling of logically visible relationships between multimedia data units based on abstract user data, quantifies the likelihood of user selection, generates target multimedia data, and preloads target speech resources in a specific language version. This significantly reduces memory usage and avoids stuttering and overflow caused by full loading. Since the target data is loaded before the user triggers a switch, it can be immediately invoked during the switch, completely eliminating the waiting delay of instant loading. This ensures system stability while achieving real-time response to language switching, effectively improving the user experience in multilingual interaction scenarios.

[0057] An alternative implementation is as follows: Figure 3As shown, Figure 3 This is a flowchart illustrating a data loading method provided in one embodiment of this specification.

[0058] Step 1: Data Input: Input the current multimedia data ID (i.e., the unique identifier of the multimedia data that the user is accessing), the current scene ID, and the user's historical access records / user operation records.

[0059] Step 2: Rule Engine: Based on preset business rules, perform rapid matching and preliminary association analysis (basic prediction) on the current scenario ID. Calculate the content logical association degree between the current multimedia data and other multimedia data based on content logical association. Output: List of candidate multimedia data.

[0060] Step 3: Input User Behavior Characteristics: Perform feature processing on user's historical access records and operation records to extract quantifiable behavioral characteristics, such as: operations on current multimedia data, dwell time, click-through rate, and dwell time in the current scenario. Output: User behavior characteristics.

[0061] User behavior features are input into the LightGBM model to calculate the access probability. This is combined with a list of candidate multimedia data and the user behavior feature vector, then fed into a pre-trained LightGBM model. For each candidate multimedia data item in the list, the model outputs a probability value for that user's access. The output is a list of candidate multimedia data items sorted in descending order of probability.

[0062] Step 4: Associate language preferences and match the user's current / frequently used languages: Query the user's configuration center or local cache to obtain the user's target language identifier, with the following priority order: 1. Language explicitly selected in the current interface / session; 2. Language used most frequently in history. Output: Target language identifier.

[0063] Step 5: Generate preload requests: From the candidate multimedia data list, select several items or Top-N items whose probability exceeds a preset threshold; for each item, combine its multimedia data ID with the target language ID to generate standardized preload instructions. Output one or more structured preload requests.

[0064] Step 6: Send to the streaming module: Send the preload request generated in step 5 to the system's streaming module.

[0065] In one optional embodiment of this specification, the access behavior record includes at least one of the following: the current access scenario, the current operation record of the current multimedia data, and the language selection frequency.

[0066] The access scene serves as a comprehensive medium for constructing interactive experiences. Through environmental art, level design, and a symbolic system, it concretizes abstract game rules and worldviews into an explorable and interactive topological structure. The scene not only creates immersion through lighting, materials, and sound effects, but also acts as the physical carrier of game mechanics, guiding player behavior and defining challenge logic, thereby realizing the spatial narrative and content expression of multimedia data. Simultaneously, the scene fulfills the function of environmental narrative, conveying non-linear story information and emotional tension through spatial arrangement, object details, and dynamic events. Accordingly, the current access scene is a virtual environment modeled by the system's structured framework, used to present the current multimedia data.

[0067] The current operation log consists of user interaction data performed on multimedia data. It is captured and stored in real-time within the current session time window, reflecting the user's immediate intent and operational preferences. The operation log details specific user behaviors, such as "pausing at 3 minutes and 27 seconds," "clicking the subtitle switch twice," "voice input 'switch to English' during playback," and "sliding the progress bar back to the starting frame." This process parses fragmented user actions into structured events with temporal characteristics, participating in condition matching for decision nodes. This model models the user's intent characteristics at that moment, transforming ambiguous user intent into quantifiable and evaluable feature signals. This ensures that the generation of target decision branches possesses both behavioral authenticity and supports the directionality of evaluation decisions. Furthermore, the current operation log can be the current user's interaction with the current multimedia data, the interactions performed by other users with the current multimedia data, or interaction patterns extracted from the comprehensive analysis of multiple other users' interactions with the current multimedia data, or a combination of these. No specific limitation is made here. Other users can specifically be users with similar characteristics, user profiles, or behaviors to the current user.

[0068] Language selection frequency (LQF) is a statistical measure of how frequently users actively select different language versions of multimedia content. It quantifies the strength and stability of user language preferences. LQF is expressed as the proportion of each language option selected out of the total number of language actions. As a key dynamic feature in the decision tree for determining language-related branches, LQF not only identifies current user language preferences but also reveals underlying intent patterns. This drives evaluation decisions to precisely target different language versions of the multimedia data or to determine the target multimedia data based on the current language version.

[0069] There are several ways to predict the target decision branch from the decision tree based on the content association probability and the current access behavior record. One possible way is to predict the target decision branch from the decision tree by reasoning based on the nodes in the decision tree based on at least one of the following: content association probability, current access scenario, current operation record of current multimedia data, and language selection frequency.

[0070] This specification's embodiments incorporate user behavior signals such as the current access scenario, current operation records, and language selection frequency into decision tree modeling, thereby constructing a three-dimensional, hierarchical, and semantically strong user behavior feature recognition system within the decision tree. The current access scenario provides content and scenario preference features, the current operation records provide user behavior preference features, and language frequency models the characteristics of different language user groups based on user language preferences. This avoids the one-sidedness of single-dimensional modeling and achieves synergistic enhancement of multi-dimensional signals. As a result, the target decision tree can refine the modeling of user intent using multi-dimensional features, significantly improving the accuracy, robustness, and personalization level of the target decision branches.

[0071] In one possible embodiment of this specification, user data includes one or more of the following: current multimedia data, user behavior data, scene state data, and multimedia application data.

[0072] Based on the correlation, predict and generate target multimedia data, including: Based on relationships, the rule engine is used to predict candidate multimedia data corresponding to the current multimedia data. Candidate multimedia data is modified based on one or more of user behavior data, scene state data, and multimedia application data to generate target multimedia data.

[0073] The rule engine is a deterministic reasoning component based on predefined semantic logic and semantic relationships. It does not rely on data training but is generated through big data construction or automated semantic modeling. It is used to filter candidate multimedia data from the current multimedia data based on content association probabilities, identifying those with content associations that are present or whose content association probabilities exceed a preset threshold. The rule engine can provide a high-quality candidate data range for subsequent target decision tree predictions, reducing the prediction pressure on the target decision tree and improving the overall efficiency of target multimedia data prediction.

[0074] Candidate multimedia data refers to a pre-identified set of multimedia data that may be requested or invoked later, which has not yet been loaded but has a high potential access probability. It can be obtained by filtering multimedia data based on the content relevance between the multimedia data and the current multimedia data, or by calculating the probability based on a large number of multimedia data selected by the user after the current multimedia data.

[0075] The core role of candidate multimedia data is to serve as the target set for the output of the decision tree, thereby reducing the amount of data required for decision tree prediction and supporting the technical implementation of decision tree prediction.

[0076] Multimedia application data describes the configuration of a multimedia application's operation and may include one or more of the following: the application's running progress, the language identifier, version number, file size, encoding format, call priority, resource dependencies, update timestamp, associated functional module, and preloading markers for multimedia data units. The purpose of multimedia application data is to provide application-level data support for relationship building and predictive generation, assisting in the selection of suitable target objects for preloading. It should be noted that the multimedia application's running progress represents the functional stage or content position of the application in the current session, including one or more of the following: the current page path, playback timeline position, task flow node identifier, user completion percentage, interaction step sequence number, or the current processing index in the resource loading queue. The role of multimedia application data is to provide dynamic context anchors for relationship building, enabling the system to identify the temporal coupling between "current progress" and "potentially subsequent resource calls," thereby supporting the accurate location and priority ranking of target multimedia data in the predictive generation stage.

[0077] User behavior data refers to the interactive traces generated by users when performing operations in a target multimedia application, which can be recorded and identified by the system. It is an objective mapping of user intentions and habits at the operational level. User behavior data can include one or more of the following: clicking, long pressing, swiping, pausing, skipping, language switching, dragging playback progress, voice command input, and resource sharing. The role of user behavior data is to provide a basis based on dynamic user behavior for the mining and prediction of correlations, enabling the system to distinguish the usage patterns of different users, thereby supporting personalized filtering and probability correction.

[0078] Scene state data refers to objective parameters of the external environment and system operating conditions in which the target multimedia application is running, including one or more of the following: device current language settings, network connection type, geographical location information, system time, device battery level, foreground / background running status, screen orientation, and list of installed language packs. The role of scene state data is to provide contextual constraints for prediction generation, enabling the system to dynamically adjust resource selection strategies based on actual operating conditions.

[0079] There are several ways to use a rule engine to predict candidate multimedia data based on association relationships. One possible approach is to use a rule engine to perform content feature analysis on the current multimedia data and other multimedia data units to obtain the content association probability, and then determine the candidate multimedia data based on the content association probability.

[0080] There are several ways to generate target multimedia data by modifying candidate multimedia data based on one or more of user behavior data, scene state data, and multimedia application data. One possible way is to modify the content association probability corresponding to the candidate multimedia data based on one or more of user behavior data, scene state data, and multimedia application data to obtain the access prediction probability, and then generate the target multimedia data based on the access prediction probability.

[0081] This specification's embodiments acquire user data such as current multimedia data, user behavior data, scene state data, and multimedia application data. Using a rule engine as a pre-inference layer, it first quickly generates candidate multimedia data with strong content relevance based on correlation relationships. Then, it dynamically corrects the probability of the candidate set by combining one or more of the user behavior data, scene state data, and multimedia application data. Thus, without the need for machine learning model training or relying on historical annotation, it achieves interpretable, low-latency, and highly adaptable target multimedia data prediction. This hierarchical prediction mechanism not only utilizes the rule engine to ensure the certainty and real-time nature of the initial screening, but also introduces user-based personalized weights through behavior data correction, making the final generated target multimedia data more closely match the real user intent and immediate operating conditions. Therefore, in multilingual interaction scenarios, the system can avoid the memory waste caused by full loading and overcome the bias problem of pure behavioral statistical prediction under cold start or small sample conditions.

[0082] One possible embodiment of this specification involves modifying candidate multimedia data based on one or more of user behavior data, scene state data, and multimedia application data to generate target multimedia data, including: Based on the current multimedia data and candidate multimedia data, a target decision tree is determined. Each decision node in the target decision tree represents the evaluation decision of candidate multimedia data based on one or more of the user behavior data, scene state data and multimedia application data and their correlation. The target decision branch is determined from the decision tree based on one or more of the following: relationship and user behavior data, scene status data, and multimedia application data. Based on the target decision branch, target multimedia data is generated.

[0083] A decision tree is a hierarchical decision model used to predict a user's next high-probability request target. It is dynamically constructed through machine learning by pre-calculating one or more of the following: sample multimedia data, the content association probability between sample candidate multimedia data and sample multimedia data, sample user behavior data, sample scene state data, and sample multimedia application data. The internal nodes of the decision tree are interpretable conditional judgments, the branches represent different behavioral paths, and the leaf nodes output the ranking weights or feature values ​​of the corresponding candidate multimedia data.

[0084] A target decision tree is a decision tree trained with sample multimedia data identical to the current multimedia data, sample candidate multimedia data identical to the candidate multimedia data, and sample access behavior records including, but not limited to, content association probabilities, user behavior data, scene state data, and multimedia application data. The target decision tree aims to predict the candidate multimedia data that the user might choose when the current multimedia data is played. The target decision tree can transform multi-dimensional inputs into a decision structure, supporting accurate identification of the target multimedia data. There can be one or more target decision trees. When there are multiple target decision trees, the decision order and decision path for content association probabilities and sample user behavior data, sample scene state data, and sample multimedia application data may differ among the target decision trees. The feature results predicted by each target decision tree for selecting candidate multimedia data after the current multimedia data may also differ. The target multimedia data can be predicted jointly based on multiple target decision trees.

[0085] Decision nodes are internal logical units in the target decision tree used to perform single-condition judgments. They are interpretable judgment rule carriers that map multi-dimensional behavioral features and content features into binary or finite discrete branches. Each node encapsulates a judgment path with feature semantics and outputs a clear path identifier leading to child nodes or leaf nodes.

[0086] Decision nodes constitute the smallest decomposable, configurable, and verifiable reasoning unit of the target decision tree, thereby supporting the model's reasoning ability and interpretability, providing decision anchors for determining target multimedia data, and serving as structural nodes connecting data characteristics, business logic, and system execution.

[0087] Evaluation decision is the judgment result with clear execution direction output by the target decision tree after it performs layer-by-layer reasoning on the current input features. Evaluation decision is usually a decision on one or more decision nodes.

[0088] There are several ways to determine the target decision tree based on the current multimedia data and candidate multimedia data. One possible way is to determine the target decision tree based on the current multimedia data and candidate multimedia data, with the goal of predicting the probability of candidate multimedia data being accessed after the current multimedia data is accessed.

[0089] Another possible approach is to determine the target decision tree based on the current multimedia data identifier and candidate multimedia data identifiers.

[0090] This step uses one or more of the following as features: current multimedia data and candidate multimedia data, anchor points, content association probabilities, user behavior data, scene state data, and multimedia application data. It then calls a pre-trained target decision tree model that supports scene adaptation. This target decision tree model is a dedicated inference structure instantiated specifically for the current multimedia data context and user interaction behavior. Each judgment layer maps to the inference path that the current multimedia data might trigger in real-world use. This consistency ensures highly focused decision logic and non-redundant generalization, making the target decision tree truly serve the current reality. Its inference results can achieve a high degree of integration between content association probabilities and current access behavior records.

[0091] A decision branch is a clearly directional path sequence in the target decision tree, originating from the root node, passing through a series of consecutive decision nodes, and ultimately arriving at a leaf node that outputs a high-confidence feature or probability. The target decision branch is the only activated logical path during real-time inference, dynamically matched based on the current multimedia data, content association probability, and current access behavior records. Furthermore, the target decision branch is selected based on the current input content association probability and current access behavior records, enabling dynamic prediction of the target multimedia data based on the current scenario.

[0092] In this step, decision nodes are used as atomic units, target decision branches are used as reasoning trajectories, and evaluation decisions are used as the final output. This ensures that while guaranteeing prediction accuracy, each decision is based on the actual scenario at that time, and the decision process is traceable. This determines the accuracy, timeliness, and resource efficiency of subsequent preloading actions.

[0093] The target multimedia data is the multimedia data instance that the user is most likely to actively request or passively trigger in the next moment during the current multimedia data playback process. It is obtained through target decision tree reasoning and final selection by evaluation decision. The target multimedia data is the endpoint output of the decision branch and the direct execution object of the preloading action.

[0094] There are several ways to determine the target multimedia data based on the target decision branch. One possible way is to determine the target leaf node based on the target decision branch, and then determine the target multimedia data based on the leaf node.

[0095] Another possible approach is to determine the target leaf node based on the target decision branch, determine the selection probability of each candidate multimedia data based on the features of at least one target leaf node, and determine the target multimedia data based on the selection probability and the selection probability threshold.

[0096] This specification's embodiments construct a target decision tree model that uses current multimedia data and candidate multimedia data as anchors and one or more of the following as input features: content association probability and user behavior data, scene state data, and multimedia application data. This results in a well-structured, traceable hierarchical logical decision. Each decision node encapsulates lightweight rules with business semantics, and each target decision branch represents the intent reasoning path activated in the current context. Finally, the leaf nodes output the access prediction probability or directly point to the target multimedia data. This significantly improves the accuracy and real-time response capability of preloading decisions and achieves intelligent preloading with strong interpretability and controllability.

[0097] One optional embodiment of this specification determines a target decision branch from a decision tree based on one or more of content association probability and user behavior data, scene state data, and multimedia application data, including: Using machine learning models, based on one or more of the following: content association probability and user behavior data, scene state data, and multimedia application data, a target decision branch is predicted from a decision tree. The machine learning model includes either a classification model of the decision tree or a collaborative filtering recommendation model.

[0098] Machine learning models are trainable algorithm systems used to model users' multimedia access intentions and implement pre-loaded decision mappings from multi-dimensional behavioral and content features. They are used to encode statistical patterns and feature logic in sample data into deployable and interpretable mathematical functions.

[0099] Decision tree classification models are supervised learning models that use decision trees as their structural framework and aim at multi-class discrimination. The training objective of a decision tree classification model is to map input samples to predefined discrete decision processes. Decision tree classification models can be LightGBM, C4.5, hierarchical if-else logic after rule distillation, etc.

[0100] Collaborative filtering recommendation models are a typical recommendation algorithm paradigm that mines potential preferences based on the co-occurrence patterns of user and multimedia data interaction behaviors, without relying on content semantic features. Its core assumption is that "users with similar behaviors tend to like similar multimedia content." By constructing a user-multimedia data rating matrix and filling in missing values, it predicts the access probability of non-interactive items. This model does not directly output a recommendation list but focuses on producing a quantitative correlation between "current multimedia data and candidate multimedia data," which is then jointly input with user behavior records into the target decision tree. Therefore, collaborative filtering recommendation models are not independent recommendation systems but rather feature generators deeply embedded in the prediction chain, ensuring that content correlation judgments combine group statistical regularities with individual behavioral specificities.

[0101] There are multiple ways to predict the target decision branch from a decision tree, based on one or more of the following: content association probability and user behavior data, scene state data, and multimedia application data. One possible approach is to infer from the nodes in the decision tree based on one or more of the following: content association probability and user behavior data, scene state data, and multimedia application data, and predict the target decision branch from the decision tree.

[0102] The embodiments in this specification organically combine a decision tree classification model with a collaborative filtering recommendation model. The former performs specialized predictions based on the current context, using access behavior records and content association probabilities, while the latter relies on group co-occurrence patterns to mine user personality traits and access preferences. Together, they provide the decision tree with predictive inputs that are both discriminative and generalizable. This not only significantly improves the prediction accuracy and robustness of the target decision branch in complex scenarios, but also maintains path traceability, condition interpretability, and strategy configurability throughout the process.

[0103] In one optional embodiment of this specification, before determining the target decision tree based on the current multimedia data and candidate multimedia data, the method further includes: Acquire sample multimedia data, the probability of sample content association between sample candidate multimedia data and sample multimedia data, and one or more of the following: sample user behavior data, sample scene state data, and sample multimedia application data for sample multimedia data. Based on the probability of association between one or more of the sample user behavior data, sample scene state data, and sample multimedia application data and the sample content, predict the sample access prediction probability corresponding to the sample candidate multimedia data. Based on the predicted probability of sample access, an evaluation decision is constructed; The evaluation decision is identified as the decision node, and a decision tree is constructed corresponding to the sample multimedia data and the sample candidate multimedia data.

[0104] The sample multimedia data are multimedia data instances used as a starting point for training or validating the machine learning model in this application. They are structured data units with complete data annotation and behavioral context association. Each sample can contain identifiable content identifiers, basic attributes, and playback context.

[0105] Candidate multimedia data are examples of multimedia data that have an access sequence relationship with the sample multimedia data. They can form a co-prediction mechanism with the sample multimedia data to model the user's transition pattern from the current item to the next item in data interaction. Each candidate multimedia data is associated with a specific sample multimedia data. Candidate multimedia data can serve as a semantic reference system and evaluation benchmark for the output of the leaf nodes of the target decision tree, thus limiting the decision tree's decision objective to the positive sample pairs formed by the sample multimedia data and the corresponding candidate multimedia data.

[0106] The sample content association probability is a quantitative indicator that represents the strength of potential access association between each pair of "sample multimedia data and sample candidate multimedia data" during the sample data construction stage. In essence, it is confidence data after fusing multiple dimensions such as semantic similarity, behavioral co-occurrence frequency, and logical continuity.

[0107] The sample user behavior data is historical user behavior log data used to train the target decision tree, which fully reproduces the entire process of real users interacting with multimedia content in a specific context; the sample access behavior records can include: (1) the behavior event itself, such as pausing, jumping to 02:15, turning on English subtitles, voice command: replay the beginning. (2) feedback on subsequent behavior results, such as whether the animation loading is triggered after the pause, whether the playback is completed after jumping, and whether the dwell time increases by 35% after the subtitles are turned on. In this way, the user's behavioral intention under the sample multimedia data is modeled, so that the model can automatically summarize generalizable rules from massive historical sessions, thereby ensuring that the target decision tree's matching of the current access behavior record is a reliable inference based on the real behavior pattern during online inference.

[0108] The sample scenario state data is a structured label for the application context to which the sample multimedia data belongs. It can include one or more of the following: functional area, content scenario, etc. Its function is to anchor the sample multimedia data to a specific interactive scenario, so that the model can establish a mapping relationship between a specific functional area / content scenario and the corresponding multimedia data. This supports the decision tree in accurately activating the decision path that matches the calling logic of the sample multimedia data based on the user's actual scenario state data during online inference.

[0109] The sample multimedia application data is metadata reflecting the configuration of the target multimedia application, including one or more of the following: the running progress of the multimedia application corresponding to the sample multimedia data, the version number of the multimedia application, the identifier of the currently loaded functional module, the resource package version, the preloading strategy flag, the dependency declaration of the multimedia data unit, the installation status of the language resource package, and the context session ID of the user session to which the sample belongs; its function is to provide the sample with the application layer running context, so that the model can identify the differences in voice resources under different multimedia application data progress.

[0110] The sample access prediction probability is a quantitative estimate of the likelihood of the event "the user actually accesses the corresponding candidate multimedia data in this sample scenario" generated through supervised learning or rule fitting, based on historically collected sample multimedia data, its corresponding candidate multimedia data, the probability of the sample content association between the two, and the sample access behavior records strongly associated with the sample multimedia data.

[0111] In practical applications, there are multiple ways to obtain sample multimedia data, the probability of sample content association between sample candidate multimedia data and sample multimedia data, and the record of sample access behavior for sample multimedia data. One possible way is to obtain sample multimedia data, the probability of sample content association between sample candidate multimedia data and sample multimedia data, and the record of sample access behavior for sample multimedia data based on a pre-built sample set.

[0112] Another possible approach is to collect and obtain user access data in real time, and then obtain sample multimedia data, the probability of sample content association between sample candidate multimedia data and sample multimedia data, and the record of sample access behavior for sample multimedia data.

[0113] This specification's embodiments construct a structured, semantically rich, and behaviorally interpretable sample data system. This system includes sample multimedia data with complete context annotations, candidate sample multimedia data that form transfer relationships with the sample multimedia data, sample content association probabilities representing the content correlation between multimedia data, and user access behavior records for the sample multimedia data. This enables deep modeling of user interaction intent. Based on this, evaluation decisions are automatically generated and decision trees are constructed with the access probability as the optimization objective. This achieves traceable, verifiable, and interpretable rule-based reasoning learning, effectively modeling the predictive and generalization capabilities of the target decision tree for the language version access probability in the current scenario.

[0114] An optional embodiment of this specification includes acquiring sample multimedia data, the probability of sample content association between sample candidate multimedia data and sample multimedia data, and records of sample access behavior related to the sample multimedia data, including: According to a preset period, acquire sample multimedia data, the probability of sample content association between sample candidate multimedia data and sample multimedia data, and the sample access behavior records for sample multimedia data within the preset period.

[0115] The preset period is a time period pre-set by the system, used to periodically collect training sample data (including access behavior, content relationships, etc.). It is a fixed or configurable time window, such as 1 hour, 24 hours, 7 days, 1 month, etc., thereby providing high-quality training samples for building / iterating the target decision tree and driving the continuous updating and optimization of the decision tree model.

[0116] This specification embodiment constructs and continuously optimizes a target decision tree model by periodically collecting sample multimedia data, its content association probability with candidate multilingual resources, and user access behavior records within a preset period. This ensures the statistical stability and business timeliness of the training data, while also realizing dynamic modeling of user language preferences and content paths. The system can synchronize the user group status based on the latest samples, thereby accurately predicting target multimedia data at any time.

[0117] This specification provides an optional embodiment in which the candidate multimedia data includes multiple different language versions; Before preloading the target multimedia data into memory, the following steps are also included: Obtain user language preferences; If the user's language preference is obtained, the target language identifier is determined based on the user's language preference; If the user's language preference is not obtained, retrieve the historically frequently used languages ​​and determine the target language identifier based on the historically frequently used languages; Preload target multimedia data into memory, including: Based on the target language identifier, the target multimedia data in the target language version is retrieved from storage and preloaded into memory.

[0118] User language preference refers to the language a user chooses or prefers to understand, receive, and interact with, along with their preferred expression style. User language preference can manifest as a user's choice of interface language, content cultural adaptation, and other related preferences. The system identifies user language preferences through explicit settings, such as language options, or implicit behaviors, such as clicks, pauses, and search terms, and optimizes content display, recommendation results, and interactive experience accordingly to improve readability, user-friendliness, and user satisfaction.

[0119] Historically frequently used language refers to the language that users have used most frequently in the past when accessing data or services. It reflects the communication language that users are most familiar with and prefer, and is an important basis for the system to realize personalized language adaptation, content recommendation and interface localization.

[0120] A target language identifier is a standardized markup used to explicitly specify in which language content should be generated, displayed, or recommended. It can be a code snippet or a data identifier. It is typically determined by user preferences, device settings, geographic location, or business strategy, and is a core control signal for language adaptation, content localization, and cross-language recommendations. The system can use this identifier to filter, translate, or generate content in the corresponding language, ensuring a consistent and comprehensible user experience.

[0121] The embodiments in this specification intelligently determine the target language identifier by prioritizing the acquisition of user language preferences and alternative historically frequently used languages. Based on this, the corresponding language version of multimedia data is accurately preloaded from storage into memory, avoiding redundant loading of multilingual content. This not only improves the efficiency of system resource utilization but also ensures that users always receive content that best suits their language habits, significantly enhancing the personalized experience, system robustness, and scalability in multilingual scenarios.

[0122] This specification provides an optional embodiment in which storage includes local storage and cloud storage; Based on the target language identifier, the target multimedia data in the target language version is retrieved from storage, and the target multimedia data is preloaded into memory, including: Generate a preload request based on the target language identifier; The streaming module is invoked to determine the storage location of the target multimedia data for the target language version, and a preload request is sent to the storage location. If the storage location is local storage, a non-priority loading thread is invoked to preload the target multimedia data of the target language version into memory; If the storage location is cloud storage, a non-priority loading thread is invoked to preload the target multimedia data of the target language version into memory based on the streaming protocol.

[0123] Preloading is a browser optimization technique that reduces real-time loading time by pre-loading potentially needed critical resources earlier in the critical rendering path. The purpose of preloading is to eliminate the latency inherent in the traditional process of loading data from storage to memory based on instructions before rendering. Preloading can reduce user waiting time, improve playback smoothness and interactive responsiveness, and is especially suitable for scenarios with unstable network conditions or large content sizes. A preload request is a request used to instruct data to be preloaded.

[0124] The streaming loading module is a functional module used to acquire multimedia data on demand and in segments. It can determine whether the target content is located in local storage or cloud storage based on the target language identifier, and progressively preload the corresponding language version of the data into memory through non-priority loading threads using local file reading or network retrieval based on streaming protocols. This module supports streaming data loading, fast first frame rendering, frame priority setting, and can efficiently adapt to complex scenarios such as multi-language, large file size, and weak network, significantly improving loading speed and user experience.

[0125] Local storage serves as the persistent storage medium for user devices. Compared to memory, the main drawbacks of local storage are its significantly slower read and write speeds, failing to meet millisecond-level real-time response requirements. Furthermore, it suffers from high access latency and high I / O overhead, resulting in noticeable performance bottlenecks when frequently and randomly accessing small files. Additionally, while local storage offers larger capacities, it is limited by device space and involves complex logic such as file management and cache eviction. Memory, on the other hand, offers advantages such as high speed, low latency, and ease of management, making it more suitable for temporary preloading and high-frequency interaction scenarios. Therefore, local storage is typically used as a supplement to memory for long-term caching, rather than as the core carrier for real-time content scheduling.

[0126] Cloud storage is a virtualized, distributed storage service provided by a service provider based on a cloud computing architecture. Its underlying layer relies on storage server clusters in large-scale data centers, providing users with data storage and access capabilities through the network. In multilingual multimedia systems, cloud storage can serve as a unified content repository for centralized management of multimedia data in various language versions and supports efficient distribution based on target language identifiers. However, its access depends on the network, resulting in latency and bandwidth costs, and it cannot replace the role of memory in real-time performance.

[0127] There are several ways to call the streaming loading module to determine the storage location of the target multimedia data for the target language version and send the preload request to the storage location. One possible way is to call the streaming loading module, determine the storage location of the target multimedia data for the target language version based on the data management module, and then send the preload request to the data management module.

[0128] This specification's embodiments utilize a layered architecture that integrates local and cloud storage, combined with a target language identifier-driven preloading mechanism. After determining the user's target language, a preloading request is automatically generated. The streaming module dynamically determines the location of the target data. If it's in local storage, a non-priority loading thread is used to quickly read the cached content; if it's in cloud storage, a streaming protocol is used to fetch the data on demand, avoiding a full download. This fully leverages the low network dependency of local storage while utilizing cloud storage for centralized management and elastic scaling of massive amounts of multilingual content. Furthermore, by placing the preloading operation on a non-priority thread, main thread rendering is avoided, effectively reducing perceived latency for the user.

[0129] The data storage status can be found in [reference]. Figure 4 , Figure 4 This is a schematic diagram of a data storage structure provided in one embodiment of this specification.

[0130] Data Management Module: The data management module is the top-level management unit of the entire system, responsible for the unified coordination and monitoring of the storage, access, and status changes of all multimedia data. The data management module not only provides external interfaces but also maintains the status information of the global resource pool, ensuring that the system maintains consistency and high performance in multi-language, multi-scenario, and high-concurrency environments.

[0131] Multimedia data is first layered according to language dimension. Language dimension layering is the first-level substructure under the data management module, used to logically isolate and classify multimedia data according to language type. It includes multiple specific language branches, such as: Language 1, Chinese; Language 2, English; Language N, Japanese, etc.

[0132] Each language branch represents an independent storage namespace or data partition, which stores all the multimedia content for the corresponding language version. This hierarchical design supports parallel processing of multiple languages, avoids resource conflicts between different languages, and facilitates the implementation of language preference-driven precise loading strategies.

[0133] The data is then layered according to content dimensions. Content-level layering is a secondary structure under the language dimension, used to further organize multimedia data within the same language according to business scenarios or functional modules. It includes multiple scenario branches, such as: Scenario 1, Chapter 1, Scenario 2, Chapter 2, Scenario N, and Function Area.

[0134] Each scene represents a content set, containing several specific multimedia data items. This hierarchical approach supports context-aware recommendation and loading; for example, when Chapter 1 is playing, the system can prioritize loading subsequent content related to Chapter 1.

[0135] Each content dimension layer contains specific multimedia data items, which are the most basic concrete data entities representing a specific piece of multimedia content, such as: Multimedia Data 1-1, Multimedia Data 1-N, Multimedia Data 2-1, Multimedia Data 2-N. These data items belong to different scenarios and language versions, and their metadata includes storage format, file path, URL, data size, format, etc.

[0136] Metadata is supplementary information about multimedia data items, used to describe their basic characteristics, including but not limited to: data size, storage path, file format (e.g., MP4, PNG, MP3), creation time, and version number. This metadata is crucial for system resource scheduling, cache eviction, and streaming protocol selection, and forms the basis for intelligent loading.

[0137] The data management module may also include an associated index table, which is a key component of the data management module. It is used to establish the mapping relationship between language ID, data ID, and scene ID, and supports quick querying of storage location / status.

[0138] An optional embodiment of this specification further includes: If the target multimedia data has been loaded into memory, mark the callable state of the target multimedia data as callable. The callable state is used to indicate that the target multimedia data of the target language is called from memory.

[0139] The callable state is a key runtime identifier set for preloaded multimedia resources in this solution. It accurately indicates that the resource has not only been successfully loaded into memory, but has also completed all pre-preparations such as possible format parsing, decoder initialization, and playback buffering. It has the ability to respond to playback or switching commands immediately, enabling the system to prioritize identifying and directly call the target language voice resource in the callable state from memory when it receives a language switching command. This bypasses any loading or network transmission waiting links and is the core state contract that ensures the real-time performance, stability, and consistent user experience of multilingual interaction.

[0140] This specification's embodiments achieve deterministic availability management of preloaded voice resources through a refined resource readiness identifier called "callable state." When a user triggers a switch, the system can identify and directly schedule the corresponding language resource that is already in the callable state within milliseconds, completely avoiding stuttering, black screens, or retry delays caused by undecoded resources, insufficient buffering, or format compatibility issues. This significantly improves the determinism and real-time performance of the system's multimedia data switching path.

[0141] An optional embodiment of this specification further includes: Upon receiving a switching instruction for switching multimedia data, determine whether the multimedia data to be switched has been loaded into memory; If so, retrieve the multimedia data from memory and replace the current multimedia data; If not, the switching multimedia data will be loaded into memory, and the switching multimedia data will be retrieved from memory to replace the current multimedia data.

[0142] The switching command is a control signal that triggers the dynamic replacement of multilingual voice resources and is uniformly connected to the resource scheduling center.

[0143] The system switches to the multimedia data that needs to be replaced. Upon receiving a switching command, the system first retrieves the available multimedia data in memory for millisecond-level replacement, or retrieves the multimedia data in memory and performs the replacement based on memory. If no match is found, the system loads the multimedia data from storage into memory and replaces it, ensuring data availability. This significantly improves the naturalness, robustness, and immersiveness of multimedia data interaction while reducing user operation costs.

[0144] There are several ways to determine whether the switching multimedia data has been loaded into memory. One possible way is to determine whether the switching multimedia data is in a callable state; if not, determine whether the switching multimedia data has been loaded into memory.

[0145] There are several ways to call and switch multimedia data from memory to replace the current multimedia data. One possible way is to call and switch multimedia data from memory, gradually reduce the audio intensity of the current multimedia data to zero, and then increase the audio intensity of the switched multimedia data from zero to a preset intensity.

[0146] The embodiments in this specification first check whether the switching multimedia data has been loaded into memory; if not, the switching multimedia data is loaded into memory and then called from memory, forming a two-stage data determination and loading scheme. This ensures availability as a fallback, significantly improves the smoothness of interaction, reduces user-perceived latency, and optimizes the efficiency of the coordinated use of memory bandwidth and computing resources through precise resource scheduling paths.

[0147] In one optional embodiment of this specification, if not, switching multimedia data loaded into memory includes: If not, starting a priority loading thread will switch the multimedia data loading to memory.

[0148] Priority loading threads are threads that are given higher execution priority in system resource scheduling. They are used to handle immediate tasks that are critical to user experience or business logic, such as data retrieval that the user's current interaction depends on, to ensure low-latency response and a smooth experience.

[0149] There are several ways to start a priority loading thread to load multimedia data into memory. One possible way is to start a loading thread, raise the priority of the loading thread to the priority, and then load the multimedia data into memory based on the priority loading thread.

[0150] The embodiments in this specification initiate a priority loading thread when multimedia content needs to be switched instantly. This thread loads the switching data that the user's current interaction depends on into memory with high priority, effectively ensuring low-latency response and smooth playback for critical operations. Compared to the non-priority thread used in ordinary preloading, this mechanism can dynamically distinguish between immediate needs and predictive loading. When system resources are scarce, it prioritizes meeting the user's current behavior needs, avoiding lag and significantly improving the real-time performance of the interaction and the overall experience quality.

[0151] The specific replacement process is as follows: Figure 5 As shown, Figure 5 This is a schematic diagram of a data switching process provided in one embodiment of this specification.

[0152] When a user triggers a multimedia data switch through the front-end interface, the system first sends a request to the language preference module to obtain the correspondence between the target language and the current content; then it calls the streaming loader to query the resource status module to determine whether the target multimedia data has been loaded into memory; if the data is already in memory, the system immediately stops playing the current multimedia data and directly starts the playback process of the target multimedia data, without waiting for loading, achieving millisecond-level switching.

[0153] When a user triggers a switching request, if the system finds that the target multimedia data is not in memory and no other multimedia data is currently loading, the system will initiate the "target multimedia data priority loading" process. At this time, the system uses a high-priority thread to pull the corresponding language version data from local storage or cloud storage and gradually writes it into memory using a streaming protocol. During the loading process, the system continues to play the current multimedia data to avoid interrupting the experience. Once the target data is loaded, the system smoothly transitions and begins playing the new content, achieving a seamless transition between playback and switching.

[0154] In certain application scenarios, the system supports applications actively triggering operations such as subtitle display as part of the switching instruction. When a user triggers a switching request, the system detects that the target data is not in memory, but the priority loading process has been started. If the application supports this, some non-critical information, such as subtitles and cover images, can be displayed in advance to improve user perception. It also allows providing some feedback before the content is fully ready to alleviate user anxiety while waiting.

[0155] If it is determined that the multimedia data has been fully loaded and resides in memory, no additional loading operation is required. The currently playing multimedia data stream is immediately terminated, and the playback process of the target multimedia data is directly started. The entire process skips network requests, local reading or decoding waiting, etc., achieving millisecond-level response and zero-latency switching, providing users with a truly seamless and smooth multilingual or multi-content interactive experience.

[0156] In one optional embodiment of this specification, after the priority loading thread is started to switch the multimedia data to memory, the method further includes: Once loading is complete, stop the priority loading thread and switch back to the non-priority loading thread.

[0157] Non-priority loading threads are background threads assigned lower execution priority in the system task scheduler. They are mainly used to handle preloading or caching tasks that do not affect current user interaction and can be completed later. Non-priority loading threads do not preempt the main thread or high-priority task resources, avoiding blocking critical paths. They can gradually complete data loading when the system is idle or has sufficient bandwidth, improving resource utilization efficiency. Even if loading is not completed, it will not cause the current operation to lag, but will only affect the readiness speed of subsequent content.

[0158] This embodiment of the specification, after initiating a high-priority loading thread to complete the urgent switching of multimedia data loading, promptly stops the high-priority thread and switches back to a non-priority loading thread. This ensures both the low-latency response required for the user's current interaction and prevents high-priority tasks from occupying system resources for extended periods. By dynamically switching thread priorities, the system guarantees a smooth experience at critical moments and efficiently performs background preloading during idle periods, achieving intelligent resource scheduling and continuous optimization of the user experience. By assigning non-urgent preloading tasks to non-priority loading threads, the system can intelligently pre-fetch personalized content while ensuring smooth real-time interaction, achieving a balance between performance and user experience.

[0159] In one optional embodiment of this specification, the switching instruction includes a preset switching instruction and an immediate switching instruction; the immediate switching instruction is generated in response to a user's switching instruction; the preset switching instruction is generated in response to a user's access behavior triggering preset trigger conditions.

[0160] Instant switching commands are user-initiated signals that request an immediate interruption of the current multimedia content and the loading / playing of another target content, such as clicking on a recommended video, switching language versions, or jumping to a chapter. Instant switching commands are highly time-sensitive and require a rapid response from a priority loading thread to ensure that the target multimedia data is loaded into memory and rendered in the shortest possible time, thus achieving a seamless, low-latency interactive experience.

[0161] Preset switching instructions refer to switching intent signals that are passively generated and cached by the system based on contextual information, such as the current playback progress, recommendation strategy, or scene switching instructions, before the user actually operates them. Unlike real-time switching instructions, preset switching instructions do not originate from real-time user input, but are passive behaviors.

[0162] This specification's embodiments achieve a harmonious balance between fine-grained control, high responsiveness, and efficient resource utilization in multimedia content switching by distinguishing between real-time switching commands and preset switching commands. Whether a user actively triggers a real-time switching command or a preset switching command is automatically generated by the system based on triggering conditions, the system can load the target content quickly and with high priority, ensuring a low-latency, lag-free interactive experience and providing users with a more intelligent, smoother, and more personalized multilingual multimedia service experience.

[0163] For detailed procedures, please refer to [link / reference]. Figure 6 , Figure 6 This is a flowchart illustrating a loading process provided for one embodiment of this specification.

[0164] 1. Preload request processing path (not prioritized loading) When the system receives a preloading instruction from the data prediction module, the instruction is submitted to a "non-priority loading queue" and processed asynchronously by a background thread: The system calls the resource status query module to check if the target multimedia data is already in memory. If it exists, it is marked as "available" and the data management module status is updated. If it is not in memory, its storage location is further determined: if it is in local storage, the data is loaded into memory via streaming; if it is in cloud storage, the data is pulled from the remote server via streaming protocol and written into memory. After loading is complete, the system updates the data management module status and reports "preloading complete," achieving silent and imperceptible content preparation to support subsequent real-time switching.

[0165] 2. Instantly switch instruction processing paths (prioritize loading) When a user actively triggers an instant / preset switching command, such as clicking on a recommended video or switching language versions, the instant / preset switching command is submitted to a priority loading queue and responded to immediately by a high-priority thread. First, the content progress data module is called to obtain the multimedia data information of the current switching unit, and the required version is determined by combining it with the target language identifier. Next, the resource status query module checks whether the target data is already in memory. If it exists, it is used directly; if not, it is loaded according to its storage location: if stored locally, it is quickly read into memory; if stored in the cloud, it is retrieved in real-time via a streaming protocol. After loading is complete, a data switch is triggered, switching from the currently playing multimedia data to the switching multimedia data, ensuring that user operations can seamlessly switch to new content with low latency, guaranteeing smooth interaction.

[0166] In one optional embodiment of this specification, before acquiring the current multimedia data, candidate multimedia data, and the current access behavior record for the current multimedia data, the method further includes: In response to the user's launch command for the front-end multimedia interface, load the interface data, initial multimedia data, and interactive sound effects of the front-end multimedia interface into memory. The interface data, initial multimedia data, and interactive sound effects are retrieved from memory and displayed on the front end. The initial multimedia data is then displayed on the front end, allowing the user to access the initial multimedia data and play the interactive sound effects.

[0167] The front-end multimedia interface is a user-interactive, visual client-side interface used to display and control rich media content such as audio, video, images, and text. It can run on browsers, mobile apps, or smart terminals. The front-end multimedia interface is responsible for rendering multimedia content, such as players, galleries, and news feeds. It can also integrate interactive elements such as language switching, recommendation lists, preloading prompts, and playback controls, and work with back-end services to achieve dynamic content scheduling based on user language preferences, access behavior, and switching commands.

[0168] A startup command is a control signal used to trigger the loading, initialization, or playback of multimedia content. It is typically generated by user actions, system events, or preset logic. Upon receiving a startup command for the front-end multimedia interface, the system loads and generates the interface, prioritizing the loading of necessary interface data to ensure its effective operation. Simultaneously, it can load initial multimedia data and interactive sound effects, including multimedia data and sound effects used more frequently than a certain threshold.

[0169] This embodiment of the specification, upon receiving a user's command to launch the front-end multimedia interface, prioritizes preloading interface data, initial multimedia data, and frequently used interface interactive sound effects into memory. This allows the former to be rendered promptly upon use, improving the responsiveness and immersiveness of the initial interaction. Simultaneously, the loaded initial content is filtered based on user usage frequency, making it more aligned with user habits. Combined with a subsequent intelligent preloading mechanism based on language preferences and behavioral records, this lays the foundation for efficient and low-latency content rendering while ensuring smooth interface operation.

[0170] In one optional embodiment of this specification, the initial multimedia data is accessed more frequently than a preset frequency threshold.

[0171] The preset frequency threshold refers to a numerical standard set in advance by the system to measure the frequency of a user's use of a certain multimedia data or interface interactive sound effect. When the access frequency, playback frequency, or trigger frequency of a certain content in the historical behavior reaches or exceeds the threshold, it is determined to be high-frequency content, and the high-frequency content can be loaded into memory as initial multimedia data.

[0172] This embodiment loads multimedia data with access frequency higher than a preset frequency threshold into memory as initial content, ensuring that the most frequently used and interesting content is presented first when the user starts the front-end multimedia interface, significantly improving the relevance and interaction efficiency of the first screen multimedia data. This mechanism avoids resource waste caused by indiscriminate loading, effectively shortens startup waiting time, reduces memory usage, and enhances user stickiness and satisfaction while ensuring system performance.

[0173] One possible embodiment of this specification also includes: Detect the percentage of data usage in memory; If the data occupancy rate exceeds a preset occupancy rate threshold, release multimedia data from memory that has a call frequency lower than a preset call frequency threshold.

[0174] The data usage ratio is the ratio between the physical memory capacity occupied by cached multimedia data and the total available memory capacity. The data usage ratio can be expressed as a percentage, for example: used memory / total available memory × 100%. The data usage ratio quantifies the real-time pressure on memory resources, directly reflecting the saturation level of the cache pool. It is used to trigger key decisions such as data reclamation, preloading scheduling degradation, or quality adaptation, thereby achieving a fine-grained balance between user experience and system stability.

[0175] The preset memory usage threshold refers to a critical value set in advance by the system to trigger memory management actions, representing a percentage of memory data usage. The value of this preset threshold is determined by comprehensively considering device hardware capabilities, the current operating scenario, network status, and user habits. The preset memory usage threshold ensures that memory resources are not overloaded, preventing lag, while also avoiding excessive release that could impact preload hit rates and user experience. It is a key control parameter for balancing performance, power consumption, smoothness, and intelligence.

[0176] Access frequency refers to the number of times multimedia data is accessed within a specified time window. This can include the number of times it is successfully loaded and rendered, the number of times it is decoded and played by the player, the number of times it is read and displayed by UI components, or the number of times it is hit by pre-loaded modules and actually serves user operations. Access frequency is a measure of the real-time popularity and practical value of multimedia data in the current context. Access frequencies higher than a preset access threshold are preferentially retained in memory, while access frequencies lower than the preset access frequency threshold are preferentially released when memory is scarce. This achieves dynamic focus of cached resources on active content, balancing response speed and memory efficiency.

[0177] This specification's embodiments establish a memory adaptive management mechanism by real-time monitoring of memory data usage ratio and intelligently identifying and releasing multimedia data with a call frequency below the preset threshold when the usage ratio exceeds the preset threshold. This ensures that memory always prioritizes high-activity, high-value content, significantly improving cache loading efficiency and resource utilization efficiency, and providing a solid and reliable runtime resource guarantee for low-latency preloading and a long-term smooth experience.

[0178] It should be noted that the data loading method provided in this manual can be applied to multiple industries or scenarios, including virtual reality processing software, home entertainment product software, digital cultural product production software, digital cultural creative software, digital cultural creative design, education, news, cultural content industry software, digital publishing software, game and animation software; digital music development and production, digital mobile multimedia development and production, etc. In some cases, it can also be applied to animation and game production engine software and development systems, game and animation software, animation and game production engine software and development systems, game and animation software, animation and game production engine software and development systems, animation and game digital content services, digital film and television development and production, digital performance development and production, etc.

[0179] Corresponding to the above method embodiments, this specification also provides embodiments of a data loading system. Figure 7 This is a schematic diagram of the system architecture of a data loading system provided as an embodiment of this specification. Figure 7 As shown, the system includes: The data prediction module 702 is used to acquire user data for the target multimedia application; determine the correlation between multimedia data units in the target multimedia application based on the user data; and predict and generate target multimedia data based on the correlation.

[0180] The streaming module 704 is used to preload target multimedia data into memory.

[0181] The data prediction module 702 is also used to predict and generate target multimedia data based on the relationship, including: using the rule engine to predict candidate multimedia data corresponding to the current multimedia data based on the relationship; and correcting the candidate multimedia data based on the current access behavior record to generate target multimedia data.

[0182] The data prediction module 702 is also used to determine a target decision tree based on the current multimedia data and candidate multimedia data. Each decision node in the target decision tree represents the evaluation decision of candidate multimedia data based on one or more of user behavior data, scene state data, and multimedia application data and their correlations. Based on the correlations and one or more of user behavior data, scene state data, and multimedia application data, a target decision branch is determined from the decision tree. Based on the target decision branch, target multimedia data is generated.

[0183] The data prediction module 702 is also used to predict the target decision branch from the decision tree based on the content association probability and the current access behavior record using a machine learning model, wherein the machine learning model includes either the classification model of the decision tree or the collaborative filtering recommendation model.

[0184] The data prediction module 702 is also used to acquire sample multimedia data, the sample content association probability between sample candidate multimedia data and sample multimedia data, and the sample access behavior record for sample multimedia data; based on the sample access behavior record and the sample content association probability, predict the sample access prediction probability corresponding to the sample candidate multimedia data; construct an evaluation decision based on the sample access prediction probability; determine the evaluation decision as the decision node, and construct the decision tree corresponding to the sample multimedia data and the sample candidate multimedia data.

[0185] The data prediction module 702 is also used to acquire sample multimedia data, sample content association probability between sample candidate multimedia data and sample multimedia data, and sample access behavior records for sample multimedia data within a preset period, according to a preset period.

[0186] The data prediction module 702 is also used to perform content feature analysis on the current multimedia data and candidate multimedia data using a rule engine to obtain the content association probability.

[0187] The language preference setting module is used to obtain the user's language preferences. If the user's language preferences are obtained, the target language identifier is determined based on the user's language preferences. If the user's language preferences are not obtained, the historically frequently used languages ​​are obtained, and the target language identifier is determined based on the historically frequently used languages.

[0188] The streaming module 704 is also used to retrieve the target multimedia data in the target language version from storage based on the target language identifier, and preload the target multimedia data into memory.

[0189] The language preference settings module is also used to obtain and report the loading status of the streaming module to the client.

[0190] The streaming loading module 704 is also used to send a query command to the data management module to query the storage location of the target multimedia data in the target language version.

[0191] The data management module receives query commands, determines the storage location of the target multimedia data in the target language, and returns the storage location to the streaming loading module.

[0192] The language preference setting module is also used to obtain the user's language preferences; if the user's language preferences are obtained, the target language identifier is determined based on the user's language preferences; if the user's language preferences are not obtained, the historically frequently used languages ​​are obtained, and the target language identifier is determined based on the historically frequently used languages.

[0193] The data prediction module 702 is also used to retrieve target multimedia data in the target language version from storage based on the target language identifier, and preload the target multimedia data into memory.

[0194] The data prediction module 702 is also used to generate a preload request based on the target language identifier.

[0195] The streaming loading module 704 is also used to determine the storage location of the target multimedia data of the target language version based on the data management module based on the preloading request, and send the preloading request to the storage location; if the storage location is local storage, a non-priority loading thread is called to preload the target multimedia data of the target language version into memory; if the storage location is cloud storage, a non-priority loading thread is called to preload the target multimedia data of the target language version into memory based on the streaming protocol.

[0196] The streaming loading module 704 is also used to mark the callable state of the target multimedia data if the target multimedia data has been loaded into memory, wherein the callable state is used to indicate that the target multimedia data of the target language is called from memory.

[0197] The streaming loading module 704 is further configured to, upon receiving a switching instruction for switching multimedia data, determine whether the switching multimedia data has been loaded into memory; if so, retrieve the switching multimedia data from memory and replace the current multimedia data; if not, load the switching multimedia data into memory and retrieve the switching multimedia data from memory to replace the current multimedia data.

[0198] The streaming loading module 704 is also used to, if not, start a priority loading thread to switch multimedia data to be loaded into memory.

[0199] The streaming module 704 is also used to stop the priority loading thread and switch back to the non-priority loading thread when loading is complete.

[0200] The core module of the multimedia application is used to respond to the user's start command for the front-end multimedia interface, load the interface data, initial multimedia data and interface interactive sound effects of the front-end multimedia interface into memory; retrieve the interface data, initial multimedia data and interface interactive sound effects from memory, display the front-end multimedia interface on the front end, and display the initial multimedia data on the front-end multimedia interface so that the user can access the initial multimedia data and play the interface interactive sound effects.

[0201] The data management module is used to detect the data usage ratio in memory; when the data usage ratio exceeds a preset usage ratio threshold, it releases multimedia data whose call frequency is lower than a preset call frequency threshold from memory.

[0202] The data loading system provided in this specification acquires user data for a target multimedia application, clarifies the object of multimedia data generation, determines the relationship between multimedia data units in the target multimedia application based on the user data, and models the relationship between multimedia data units based on the user data. This quantifies the possibility of user selection, generates target multimedia data, and preloads target audio resources in a specific language version, significantly reducing memory usage and avoiding stuttering and overflow caused by full loading. The target data is loaded before the user triggers the switch, so it can be called immediately during the switch, completely eliminating the waiting delay of real-time loading. While ensuring the stability of system operation, it achieves real-time response to language switching and effectively improves the user experience in multilingual interaction scenarios.

[0203] The above is an illustrative scheme of a data loading system according to this embodiment. It should be noted that the technical solution of this data loading system and the technical solution of the data loading method described above belong to the same concept. For details not described in detail in the technical solution of the data loading system, please refer to the description of the technical solution of the data loading method described above.

[0204] Figure 8 This is a structural block diagram of a computing device according to one embodiment of this specification. The components of the computing device 800 include, but are not limited to, a memory 810 and a processor 820. The processor 820 is connected to the memory 810 via a bus 830, and a database 850 is used to store data.

[0205] The computing device 800 also includes an access device 840, which enables the computing device 800 to communicate via one or more networks 860. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. The access device 840 may include one or more of any type of wired or wireless network interface (e.g., Network Interface Card (NIC)), such as an IEEE 802.11 Wireless Local Area Networks (WLAN) interface, a Wi-MAX (World Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0206] In one embodiment of this specification, the above-described components of the computing device 800 and Figure 8 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 8 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0207] The computing device 800 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 800 can also be a mobile or stationary server.

[0208] The processor 820 is used to execute computer programs / instructions, which, when executed by the processor, implement the steps of the data loading method described above.

[0209] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the data loading method described above belong to the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the data loading method described above.

[0210] An embodiment of this specification also provides a computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the data loading method described above.

[0211] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solution of the data loading method described above belong to the same concept. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the data loading method described above.

[0212] An embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described data loading method.

[0213] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the data loading method described above belong to the same concept. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solution of the data loading method described above.

[0214] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0215] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0216] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0217] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0218] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A data loading method, characterized in that, include: Acquire user data for the target multimedia application; Based on the user data, the association relationships between multimedia data units in the target multimedia application are determined; Based on the aforementioned association, target multimedia data is predicted and generated. The target multimedia data is preloaded into memory.

2. The method according to claim 1, characterized in that, The user data includes one or more of the following: current multimedia data, user behavior data, scene status data, and multimedia application data; The step of predicting and generating target multimedia data based on the aforementioned association includes: Based on the relationship, the rule engine is used to predict the candidate multimedia data corresponding to the current multimedia data; The candidate multimedia data is modified based on one or more of the user behavior data, the scene state data, and the multimedia application data to generate the target multimedia data.

3. The method according to claim 2, characterized in that, The step of modifying the candidate multimedia data based on one or more of the user behavior data, the scene state data, and the multimedia application data to generate the target multimedia data includes: Based on the current multimedia data and the candidate multimedia data, a target decision tree is determined, wherein each decision node in the target decision tree represents the evaluation decision of the candidate multimedia data based on one or more of the user behavior data, scene state data and multimedia application data and their correlation. Based on the aforementioned relationships and one or more of the user behavior data, scene state data, and multimedia application data, a target decision branch is determined from the decision tree; Based on the target decision branch, target multimedia data is generated.

4. The method according to claim 1, characterized in that, The target multimedia data includes multiple versions in different languages; Before preloading the target multimedia data into memory, the method further includes: Obtain user language preferences; If the user's language preference is obtained, the target language identifier is determined based on the user's language preference; If the user's language preference is not obtained, retrieve the historically frequently used languages ​​and determine the target language identifier based on the historically frequently used languages; The step of preloading the target multimedia data into memory includes: Based on the target language identifier, the target multimedia data of the target language version is retrieved from storage and preloaded into memory.

5. The method according to claim 4, characterized in that, The storage includes local storage and cloud storage; The step of retrieving target multimedia data in the target language version from storage based on the target language identifier and preloading the target multimedia data into memory includes: Based on the target language identifier, a preload request is generated; The streaming loading module is invoked to determine the storage location of the target multimedia data for the target language version, and the preloading request is sent to the storage location. If the storage location is local storage, a non-priority loading thread is invoked to preload the target multimedia data of the target language version into the memory; If the storage location is cloud storage, a non-priority loading thread is invoked to preload the target multimedia data of the target language version into the memory based on a streaming protocol.

6. The method according to claim 1, characterized in that, Also includes: Upon receiving a switching instruction for switching multimedia data, determine whether the switching multimedia data has been loaded into the memory; If so, retrieve the switched multimedia data from the memory and replace the current multimedia data; If not, load the switched multimedia data into the memory, and retrieve the switched multimedia data from the memory to replace the current multimedia data.

7. The method according to claim 6, characterized in that, If not, loading the switched multimedia data into the memory includes: If not, start a priority loading thread to load the switched multimedia data into the memory.

8. The method according to claim 7, characterized in that, If not, after starting the priority loading thread to load the switched multimedia data into the memory, the method further includes: Once loading is complete, stop the priority loading thread and switch back to the non-priority loading thread.

9. The method according to any one of claims 6-8, characterized in that, The switching instructions include preset switching instructions and instant switching instructions; the preset switching instructions are generated in response to the user's access behavior triggering preset trigger conditions; the instant switching instructions are generated in response to the user's switching instructions.

10. The method according to claim 1, characterized in that, Before acquiring user data for the target multimedia application, the method further includes: In response to a user's launch command for the front-end multimedia interface, the interface data, initial multimedia data, and interface interactive sound effects of the front-end multimedia interface are loaded into memory, wherein the access frequency of the initial multimedia data is higher than a preset frequency threshold. The interface data, the initial multimedia data, and the interface interactive sound effects are retrieved from the memory, and the front-end multimedia interface is displayed on the front end. The initial multimedia data is also displayed on the front-end multimedia interface, so that the user can access the initial multimedia data and play the interface interactive sound effects.

11. The method according to claim 1, characterized in that, Also includes: Detect the data usage ratio of the memory; If the data occupancy ratio exceeds a preset occupancy ratio threshold, release multimedia data whose call frequency is lower than a preset call frequency threshold from the memory.

12. A data loading system, characterized in that, Includes a data prediction module and a streaming loading module; The data prediction module is used to acquire user data for the target multimedia application; and based on the user data, determine the correlation between multimedia data units in the target multimedia application. Based on the aforementioned association, target multimedia data is predicted and generated. The streaming loading module is used to preload the target multimedia data into memory.

13. A computing device, characterized in that, include: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, The device stores a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 11.

15. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 11.

Images