Target object quality index determination method and device, and electronic device

Abstract

The application provides a quality index determination method and device for a target object, and electronic equipment, and the method comprises the following steps: extracting a target object feature vector of different dimensions based on basic historical data; determining a first target object of a training sample according to a search click log of the target object; obtaining a training sample set corresponding to the first target object according to the target object feature vector of different dimensions; training a quality index determination model based on the training sample set, determining model parameters of the quality index determination model, predicting the quality index of all target objects through the quality index determination model, and obtaining a quality index prediction result of all target objects. The embodiments of the application can also be applied to various scenes such as cloud technology, artificial intelligence, intelligent transportation and auxiliary driving.

Description

Technical Field

[0001] This invention relates to information processing technology, and more particularly to a method, apparatus, electronic device, and storage medium for determining the quality indicators of a target object. Background Technology

[0002] Search quality assessment is one of the core foundational tasks in the field of search research. The quality of search engine results largely depends on users' overall satisfaction with the query results, and the perceived quality indicators of search results are a crucial aspect of this satisfaction. When users encounter search results, content published by authoritative and high-quality accounts significantly enhances their search experience.

[0003] In related technologies, conventional account quality indicator evaluation methods mostly rely on business experience to rank accounts based on factors such as the number of followers, the number of posts, and the number of reads using a weighted formula. This approach is rather coarse-dimensional, lacking clarity in differentiating account quality levels, resulting in less than ideal quality indicator evaluation. This affects the accuracy of recommendations, preventing users from promptly accessing information created by high-level accounts, and ultimately impacting the quality of search engine recommendations. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a method, apparatus, electronic device, and storage medium for determining the quality indicators of a target object. The technical solution of the embodiments of the present invention is implemented as follows: This invention provides a method for determining the quality indicators of a target object, comprising: Retrieve basic historical data from all target objects; Based on the aforementioned historical data, feature vectors of target objects in different dimensions are extracted. Based on the search click logs of the target object, determine the first target object of the training sample; Based on the target object feature vectors of different dimensions, obtain the training sample set corresponding to the first target object; Based on the training sample set, the quality index determination model is trained to determine the model parameters of the quality index determination model; The quality index determination model is used to predict the quality index of all target objects, and the prediction results of the quality index of all target objects are obtained.

[0005] This invention also provides a device for determining the quality indicators of a target object, comprising: The information transmission module is used to acquire basic historical data from all target objects; The information processing module is used to extract feature vectors of target objects in different dimensions based on the aforementioned basic historical data; The information processing module is used to determine the first target object of the training sample based on the search click log of the target object; The information processing module is used to obtain the training sample set corresponding to the first target object based on the target object feature vectors of different dimensions. The information processing module is used to train the quality index determination model based on the training sample set and determine the model parameters of the quality index determination model. The information processing module is used to predict the quality indicators of all target objects through the quality indicator determination model, and obtain the quality indicator prediction results of all target objects.

[0006] In the above scheme, The information processing module is used to determine different dimensions of the feature vector of the target object, wherein the different dimensions include at least one of the following: basic attribute feature dimension, article content feature dimension, interaction data feature dimension, service capability feature dimension, and search performance feature dimension. The information processing module is used to extract historical data of different dimensions from the basic historical data based on the different dimensions of information. The information processing module is used to perform feature processing on the historical data of different dimensions to obtain feature vectors of target objects of different dimensions.

[0007] In the above scheme, The information processing module is used to determine the business requirements corresponding to the model based on the quality indicators, and to determine the target object feature vectors of different dimensions that need to be crossed. The information processing module is used to perform feature crossing on the feature vectors of the target objects of different dimensions that need to be crossed, so as to obtain the cross feature vector.

[0008] In the above scheme, The information processing module is used to extract data on objects of interest, authentication data, and official entity number information from the basic historical data when the dimensional information is the basic attribute feature dimension. The information processing module is used to extract historical posting lists, original posting tags, posting category target tags, and posting quality tags from the basic historical data when the dimension information is a posting content feature dimension. The information processing module is used to extract the number of reads, forwards, comments, and favorites from the basic historical data when the dimension information is an interactive data feature dimension. The information processing module is used to extract the number of menus and the number of menu visits from the basic historical data when the dimension information is a service capability feature dimension. The information processing module is used to extract the click-through rate parameter and the average click dwell time parameter from the basic historical data when the dimension information is a search performance feature dimension.

[0009] In the above scheme, The information processing module is used to filter random search terms and low-frequency search terms based on the search click logs of the target object; The information processing module is used to determine the second target object corresponding to the random search term; The information processing module is used to determine the third target object corresponding to the low-frequency search term; The information processing module is used to combine the second target object and the third target object to obtain the first target object of the training sample.

[0010] In the above scheme, The information processing module is used to determine the target search engine that matches the quality indicator determination model; The information processing module is used to trigger the corresponding word segmentation library based on the search term parameter information carried in the search click log of the target search engine. The information processing module is used to perform word segmentation on the search click logs of the target search engine through the triggered word segmentation dictionary to form different word-level search terms and sentence-level search terms; The information processing module is used to determine the search results corresponding to the different word-level search terms and sentence-level search terms respectively; The information processing module is used to perform noise reduction processing on the different word-level search terms and sentence-level search terms to form search click data corresponding to the search click logs of the target search engine.

[0011] In the above scheme, The information processing module is used to determine the name of the word segmentation library used when performing word segmentation processing on the search term text; The information processing module is used to determine the parameters of a word segmentation library that matches the word-level feature vector corresponding to the search term text, based on the name of the word segmentation library. The parameters of the word segmentation library include: The types of word segmentation libraries, the names of word segmentation libraries, and the versions of word segmentation libraries.

[0012] In the above scheme, The information processing module is used to calculate the gear information of each target object based on the quality index prediction results of the target object; The information processing module is used to adjust the recall strategy for multimedia information based on the gear information of each target object and the prediction result of the quality index, and to recommend multimedia information through the adjusted recall strategy.

[0013] In the above scheme, The information processing module is used when the multimedia information is a video advertisement. The information processing module is used to send the exposure parameters of the video advertisement during playback to the detection server, so that the detection server can obtain the exposure parameters of the video advertisement. The information processing module is used to use the exposure parameter as an evaluation parameter for the playback effect of the multimedia information, and to find the target exposure parameter based on the adjustment result of the recall strategy.

[0014] In the above scheme, The information processing module is used to obtain the historical browsing information of the audience corresponding to the target industry; The information processing module is used to determine the multimedia information exposure history corresponding to the historical browsing information based on the historical browsing information of the audience corresponding to the target industry. The information processing module is used to dynamically adjust the recall strategy of the multimedia information based on the exposure history of the multimedia information corresponding to the historical browsing information and the tier information of the target object.

[0015] This invention also provides an electronic device, the electronic device comprising: Memory, used to store executable instructions; The processor, when executing executable instructions stored in the memory, implements the aforementioned method for determining the quality indicators of the target object.

[0016] This invention also provides a computer-readable storage medium storing executable instructions, which, when executed by a processor, implement the aforementioned method for determining the quality indicators of a target object.

[0017] The embodiments of the present invention have the following beneficial effects: 1) This invention acquires basic historical data from all target objects; extracts target object feature vectors of different dimensions based on the basic historical data; determines the first target object of the training samples based on the search click logs of the target objects; obtains the training sample set corresponding to the first target object based on the target object feature vectors of different dimensions; and trains the quality indicator determination model based on the training sample set to determine the model parameters of the quality indicator determination model. Thus, by using finer-grained feature extraction of target objects from different dimensions, the quality indicators of target objects can be more comprehensively characterized, making the quality indicators of target objects more in line with the needs of recall strategies.

[0018] 2) The quality index determination model is used to predict the quality index of all target objects, resulting in predictions for the quality index of all target objects. This improves the accuracy of the predicted quality index of target objects, better identifies the partial order relationships between different target objects, and makes the retrieval of multimedia information based on the quality index of target objects more accurate. Attached Figure Description

[0019] Figure 1 A schematic diagram illustrating a usage scenario of the method for determining the quality indicators of a target object provided in an embodiment of the present invention; Figure 2 A schematic diagram of the composition structure of the target object quality index determination device provided in an embodiment of the present invention; Figure 3 This is an optional flowchart illustrating the method for determining the quality indicators of a target object provided in an embodiment of the present invention. Figure 4 This is an optional flowchart illustrating the method for determining the quality indicators of a target object provided in an embodiment of the present invention. Figure 5 This is a schematic diagram illustrating the search results of the target search engine in an embodiment of the present invention. Figure 6 This is a schematic diagram of click data consisting of search terms and corresponding search results in an embodiment of the present invention; Figure 7 This is a schematic diagram illustrating the composition of click data in an embodiment of the present invention; Figure 8 This is a schematic diagram of the model structure for determining quality indicators in this invention; Figure 9 This is a schematic diagram illustrating the effect of determining the quality indicators of the target object in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0022] Before providing a further detailed description of the embodiments of the present invention, the nouns and terms involved in the embodiments of the present invention will be explained, and the nouns and terms involved in the embodiments of the present invention shall be interpreted as follows.

[0023] 1) In response to, used to indicate the conditions or states on which the operation performed depends. When the conditions or states on which it depends are met, one or more operations performed may be performed in real time or with a set delay. Unless otherwise specified, there is no restriction on the order in which the multiple operations are performed.

[0024] 2) Based on, used to indicate the conditions or states on which the operation is performed depends. When the conditions or states on which it depends are met, one or more operations can be performed in real time or with a set delay. Unless otherwise specified, there is no restriction on the order in which the multiple operations are performed.

[0025] 3) Model training involves multi-class classification learning on the image dataset. This model can be built using deep learning frameworks such as TensorFlow and Torch, employing multiple layers of neural networks like CNNs to form a multi-class classification model. The model input is a three-channel or original-channel matrix generated from images read using tools like OpenCV. The model output is the multi-class probability, ultimately outputting the quality index judgment result of the target object through algorithms such as softmax. During training, the model approximates the correct trend using objective functions such as cross-entropy.

[0026] 4) Neural Network (NN): Artificial Neural Network (ANN), also known as neural network or neural network-like network, is a mathematical or computational model in the fields of machine learning and cognitive science that imitates the structure and function of biological neural networks (the central nervous system of animals, especially the brain) and is used to estimate or approximate functions.

[0027] 5) Recommendation Accuracy: The recommended multimedia content has a certain effect over a period of time, and this effect is measured by the user's interest in the multimedia content. Accuracy plays an important role in online user retention, clicks, and CTR on the client side.

[0028] 6) Multimedia information, including various forms of information available on the Internet, such as advertising information, video files, recommended multimedia information, news information, etc., presented on clients or smart devices.

[0029] 7) Gradient Boosting Decision Tree (GBDT): This is an iterative decision tree algorithm composed of multiple decision trees. The conclusions of all trees are summed to obtain the final answer. XGBoost is a type of gradient boosting tree model that uses a second-order Taylor expansion of the loss function and optimizes the loss function using the information of the second derivative. It greedily selects whether to split nodes based on whether the loss function decreases. Furthermore, XGBoost incorporates regularization, learning rate adjustment, column sampling, and approximate optimal split points to prevent overfitting.

[0030] 8) Logistic regression (LR) model is a log-odds model used to predict the results of binary classification of input features.

[0031] In this invention, embodiments can be implemented using cloud technology. Cloud technology refers to a hosting technology that unifies hardware, software, and network resources within a wide area network (WAN) or local area network (LAN) to achieve data computation, storage, processing, and sharing. It can also be understood as a general term for network technologies, information technologies, integration technologies, management platform technologies, and application technologies based on cloud computing business models. The backend services of network systems require substantial computing and storage resources, such as video websites, image websites, and many portal websites; therefore, cloud technology needs cloud computing as its support.

[0032] It's important to note that cloud computing is a computing model that distributes computing tasks across a resource pool comprised of numerous computers, enabling various application systems to access computing power, storage space, and information services as needed. The network providing these resources is called the "cloud." From the user's perspective, resources in the "cloud" are infinitely scalable, readily available, and can be used on demand, expanded at any time, and paid for based on usage. As the foundational providers of cloud computing capabilities, they establish cloud resource pool platforms, often referred to as cloud platforms or Infrastructure as a Service (IaaS). These platforms deploy various types of virtual resources within the resource pool for external customers to choose from. The cloud resource pool primarily includes: computing devices (which can be virtualized machines containing operating systems), storage devices, and network devices.

[0033] For example, one type of AI cloud service can provide a quality indicator determination service for target objects. Specifically, the cloud server encapsulates the target object quality indicator determination program provided in this application embodiment. Users access the cloud service through a terminal (running a client, such as an instant messaging client, live streaming client, short video client, social media client, etc.) to call the target object quality indicator determination service. This allows the cloud-deployed server to call the encapsulated target object quality indicator determination program, determine which of the numerous target objects are high-quality works, and perform corresponding operations on the information based on the target object quality indicator determination results. For example, during the recall phase, works of target objects with higher quality indicator prediction results are prioritized; during the ranking phase, target objects with lower identified quality indicator prediction results are down-weighted or filtered.

[0034] Figure 1 This is a schematic diagram illustrating a usage scenario of the method for determining the quality indicators of a target object provided in an embodiment of the present invention. (See attached diagram.) Figure 1The terminals (including terminals 10-1 and 10-2) are equipped with corresponding clients capable of playing embedded multimedia information. The terminals connect to server 200 via network 300, which can be a wide area network (WAN), a local area network (LAN), or a combination of both. Data transmission is achieved using a wireless link. The multimedia information includes, but is not limited to, videos, images, GIF animations, and advertising information. The types of multimedia information obtained by the terminals (including terminals 10-1 and 10-2) from the corresponding server 200 via network 300 can be the same or different. For example, the terminals (including terminals 10-1 and 10-2) can obtain video advertisements or image advertisements from the same industry from the corresponding server 200 via network 300. The specific types are not limited in this application. Server 200 can store different multimedia information, including advertising multimedia information in different dynamic formats, such as GIF, MP4, and MOV.

[0035] During the process of the terminal (terminal 10-1 and / or terminal 10-2) obtaining and displaying the corresponding service with embedded multimedia information from the server 200 via network 300, the user can perform different operations on the multimedia information presented in the multimedia information playback window through the terminal (terminal 10-1 and / or terminal 10-2), generating different user usage process data records. For example, when the multimedia information is a video advertisement, the user can share and / or like the exposed video advertisement while watching the information, or click on it. When the multimedia information is a dynamic GIF advertisement, during the exposure of the advertisement through the terminal (terminal 10-1 and / or terminal 10-2), the user can forward and / or comment on the advertisement, or jump to the corresponding product purchase link page through the GIF advertisement.

[0036] As an example, when server 200 determines which multimedia information to recommend to user terminal 10-1 or 10-2, it needs to adjust the recommended multimedia information in a timely manner, such as replacing any multimedia information in the set to be recommended, to adapt to the viewing needs of audiences corresponding to different target industries. Taking video advertising multimedia information as an example, the quality index determination model provided by this invention can be applied to video advertising playback. In video advertising playback, different video advertising multimedia information from different data sources is usually processed, and finally, the corresponding different multimedia information and the corresponding recommended video to be recommended are presented on the user interface (UI). The accuracy and timeliness of the characteristics of different multimedia information directly affect the user experience. The background database of video playback receives a large amount of video data from different sources every day. The different multimedia information obtained for recommending multimedia information to audiences corresponding to the target industry can also be called by other applications (e.g., the recommendation results of the video advertising recommendation process are migrated to the long video recommendation process or the news recommendation process). Of course, the quality index determination model matching the audiences corresponding to the target industry can also be migrated to different video recommendation processes (e.g., webpage video recommendation process, mini-program video recommendation process, or long video client video recommendation process).

[0037] As an example, server 200 is used to deploy a corresponding quality indicator determination model to implement the target object quality indicator determination method provided by this invention, or to deploy a target object quality indicator determination device to implement the target object quality indicator determination method. Specifically, it can acquire basic historical data from all target objects; extract target object feature vectors of different dimensions based on the basic historical data; determine the first target object of the training samples according to the search click logs of the target objects; obtain the training sample set corresponding to the first target object according to the target object feature vectors of different dimensions; train the quality indicator determination model based on the training sample set, determine the model parameters of the quality indicator determination model, so as to adjust the recall strategy of multimedia information in the target industry through the quality indicator determination model, and recommend multimedia information through the recall strategy, and display the multimedia information created by the target object with a high output quality indicator prediction result through the terminal (terminal 10-1 and / or terminal 10-2), so that users can obtain a better recommendation experience. Taking multimedia information as an example, the quality index determination model provided by this invention can be applied to video ad playback. During video ad playback, different multimedia information from different data sources is typically processed, and ultimately presented on the user interface (UI) along with the corresponding multimedia information and the multimedia information to be recommended in the corresponding video ad recommendation process. The accuracy and timeliness of the characteristics of different multimedia information directly affect the user experience. The video playback backend database receives a large amount of multimedia information data from different sources every day. The obtained multimedia information, which is used to recommend multimedia information to the target industry's corresponding audience, can also be called by other applications (e.g., the recommendation results of the video ad recommendation process can be migrated to the recommendation process in an instant messaging client or a news recommendation process). Of course, multimedia information created by users with high quality index prediction results can also be migrated to different video recommendation processes (e.g., webpage video recommendation process, mini-program video recommendation process, or video recommendation process in an instant messaging client). The recommended video ads can meet the user's viewing needs.

[0038] The method for determining the quality indicators of the target object provided in this application embodiment is based on artificial intelligence (AI). AI is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess perception, reasoning, and decision-making capabilities.

[0039] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0040] In the embodiments of this application, the main artificial intelligence software technologies involved include the aforementioned speech processing technologies and machine learning. For example, it may involve Automatic Speech Recognition (ASR) technology in speech technology, including speech signal preprocessing, speech signal frequency analyzing, speech signal feature extraction, speech signal feature matching / recognition, and speech training.

[0041] For example, this could involve machine learning (ML), a multidisciplinary field encompassing probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning typically includes techniques such as deep learning, which includes artificial neural networks, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and deep neural networks (DNNs).

[0042] It is understood that the method for determining the quality indicators of the target object and the voice processing provided in this application can be applied to intelligent devices. Intelligent devices can be any device with information display functions, such as intelligent terminals, smart home devices (such as smart speakers, smart washing machines, etc.), smart wearable devices (such as smartwatches), in-vehicle intelligent central control systems (which display multimedia information to users through applets that perform different tasks), or AI intelligent medical devices (which display treatment cases by showing multimedia information), etc.

[0043] The structure of the target object quality index determination device according to an embodiment of the present invention will be described in detail below. The target object quality index determination device can be implemented in various forms, such as a dedicated terminal with multimedia information recommendation processing function, or a server equipped with target object quality index determination device processing function, for example, the preceding... Figure 1 Server 200 in the middle. Figure 2 This is a schematic diagram of the composition of the device for determining the quality indicators of a target object provided in an embodiment of the present invention. It can be understood that... Figure 2 This is only an exemplary structure of the device for determining the quality indicators of the target object, and not the entire structure. It can be implemented as needed. Figure 2 The structure shown may be part or all of the structure.

[0044] The target object quality index determination device provided in this embodiment of the invention includes: at least one processor 201, a memory 202, a user interface 203, and at least one network interface 204. The various components in the target object quality index determination device are coupled together via a bus system 205. It can be understood that the bus system 205 is used to realize the connection and communication between these components. In addition to a data bus, the bus system 205 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 2 The general labeled all buses as Bus System 205.

[0045] The user interface 203 may include a monitor, keyboard, mouse, trackball, click wheel, buttons, touchpad, or touch screen.

[0046] It is understood that memory 202 can be volatile memory or non-volatile memory, or both. In this embodiment of the invention, memory 202 is capable of storing data to support the operation of a terminal (such as 10-1). Examples of this data include any computer programs used to operate on the terminal (such as 10-1), such as operating systems and applications. The operating system includes various system programs, such as the framework layer, core library layer, driver layer, etc., used to implement various basic services and handle hardware-based tasks. Applications can include various applications.

[0047] In some embodiments, the target object quality index determination device provided in this invention can be implemented using a combination of hardware and software. For example, the target object quality index determination device provided in this invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the training method of the quality index determination model provided in this invention. For instance, the hardware decoding processor can employ one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0048] As an example of the target object quality index determination device provided in this embodiment of the invention, which is implemented by combining software and hardware, the target object quality index determination device provided in this embodiment of the invention can be directly embodied as a combination of software modules executed by processor 201. The software modules can be located in a storage medium, which is located in memory 202. Processor 201 reads the executable instructions included in the software modules in memory 202 and combines them with necessary hardware (e.g., including processor 201 and other components connected to bus 205) to complete the training method of the quality index determination model provided in this embodiment of the invention.

[0049] As an example, processor 201 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0050] As an example of the hardware implementation of the quality index determination device for the target object provided in the embodiments of the present invention, the device provided in the embodiments of the present invention can be directly executed by a processor 201 in the form of a hardware decoding processor. For example, it can be executed by one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components to implement the training method of the quality index determination model provided in the embodiments of the present invention.

[0051] In this embodiment of the invention, the memory 202 is used to store various types of data to support the operation of the target object quality index determination device. Examples of such data include: any executable instructions for operation on the target object quality index determination device, such as executable instructions that can be included in the executable instructions, and a program implementing the training method of the quality index determination model of this embodiment of the invention.

[0052] In other embodiments, the target object quality index determination device provided in this invention can be implemented in software. Figure 2A device for determining the quality index of a target object stored in memory 202 is shown. This device can be software in the form of programs and plugins, and includes a series of modules. As an example of a program stored in memory 202, it may include the device for determining the quality index of the target object. The device for determining the quality index of the target object includes the following software modules: Information transmission module 2081 and information processing module 2082. When the software modules in the target object quality index determination device are read into RAM and executed by processor 201, the training method for the quality index determination model provided in this embodiment of the invention will be implemented. The functions of each software module in the target object quality index determination device include: The information transmission module 2081 is used to acquire basic historical data from all target objects.

[0053] The information processing module 2082 is used to extract feature vectors of target objects in different dimensions based on basic historical data.

[0054] The information processing module 2082 is used to determine the first target object of the training sample based on the search click log of the target object.

[0055] The information processing module 2082 is used to obtain the training sample set corresponding to the first target object based on the target object feature vectors of different dimensions.

[0056] The information processing module 2082 is used to train the quality index determination model based on the training sample set and determine the model parameters of the quality index determination model.

[0057] The information processing module 2082 is used to predict the quality indicators of all target objects through the quality indicator determination model, and obtain the prediction results of the quality indicators of all target objects.

[0058] according to Figure 2 The illustrated electronic device, in one aspect of this application, also provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. The processor of the electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform various alternative implementations of the method for determining the quality indicators of the target object, as provided in different embodiments and combinations thereof.

[0059] Before introducing the multimedia information recommendation method provided in this application, the shortcomings of multimedia information recommendation in related technologies will be briefly explained. When performing multimedia information recommendation in related technologies, the following methods can be used: Combining expert business experience with a small number of key features, a weighted scoring formula is used. Taking WeChat official account information recommendation as an example: Related technologies utilize expert business experience to extract a small number of core features of the official account, such as the number of followers, the number of articles published, and the number of reads. These features are then weighted using a formula for scoring, for example, "score = a Follower count + b Number of posts + c The scoring method uses "read count" and sets different thresholds for different account levels. Multiple scoring formulas may be needed to address the specific characteristics of each account. The drawback of this method is that the granularity of account assessment is too coarse, making it impossible to accurately evaluate each account uniformly. The accuracy of the quality indicators at each level is insufficient, resulting in inadequate differentiation between accounts and preventing the recall strategy from fully utilizing the account evaluation quality.

[0060] To solve the above problems, combined with Figure 2 The illustrated device for determining the quality index of the target object describes the method for determining the quality index of the target object provided in this embodiment of the invention. See also: Figure 3 , Figure 3 This is an optional flowchart illustrating the method for determining the quality indicators of a target object provided in an embodiment of the present invention. It can be understood that... Figure 3 The steps shown can be performed by various electronic devices that operate the target object quality indicator determination device, such as a server or server cluster equipped with the target object quality indicator determination device, wherein the dedicated terminal equipped with the target object quality indicator determination device can be a preceding step. Figure 2 The illustrated embodiment is an electronic device with a target object quality index determination device. The following section addresses... Figure 3 The steps shown are explained.

[0061] Step 301: The target object quality index determination device acquires basic historical data from all target objects.

[0062] In this application, each target object can correspond to at least one account. By default, each target object in each application scenario corresponds to one account. Each account can send multimedia information in different forms. The content structure of each account's information is used to characterize the content style of the information. For example, when the information is text, its content structure may include at least one of the following: title length, text length, number of images, and image-to-text ratio. When the information is video, its content structure may include at least one of the following: title length, video duration, and video-related descriptive information. By extracting features from the content structure of the information, relevant features identifying the quality of the information's content structure are obtained. The account features of the information can be characterized based on the account's level. The account level can be divided according to the account's activation rate, account activity, and the number or frequency of information published by the account. Generally, the account level is positively correlated with the recommendation volume of its published information; for example, the recommendation volume of information published by an authoritative account is greater than that of information published by a general account. The content understanding features of the information are used to indicate whether the information belongs to the quality category that is blocked from recommendation in the second stage. For example, if the information belongs to at least one of the quality categories such as meaningless title, pieced-together article, or advertising text, all creative information of that account can be blocked.

[0063] The basic historical data obtained in step 301 is the sum of data from various industries' multimedia information recommendation processes. For example, it can include the sum of all basic data across multiple multimedia information recommendation environments such as product recommendations, advertising recommendations, e-commerce advertising recommendations, news recommendations, and short video recommendations. When obtaining basic historical data, it is possible to effectively extract data from the raw logs of user usage data, such as extracting the target device ID (user account), multimedia information type, multimedia information browsing duration, and the multimedia information recommendation environment to obtain basic historical data from different dimensions.

[0064] It is understood that in the specific implementation of this application, user-related data such as basic historical data in the media information recommendation environment are involved. When the embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0065] Step 302: The target object quality index determination device extracts target object feature vectors of different dimensions based on basic historical data.

[0066] In some embodiments of the present invention, unlike the feature segmentation dimensions in related technologies, the target object feature vectors of different dimensions in this application mainly include five dimensions: basic attribute feature dimension, article content feature dimension, interaction data feature dimension, service capability feature dimension, and search performance feature dimension. The process of extracting historical data of different dimensions from basic historical data based on different dimensional information is described below, with reference to... Figure 4 , Figure 4 An optional flowchart illustrating the method for determining the quality indicators of a target object provided in an embodiment of the present invention specifically includes the following steps: Step 401: When the dimensional information is the basic attribute feature dimension, extract the data of the objects of interest, authentication data, and official entity number information from the basic historical data.

[0067] Within the basic attribute feature dimension, the system can read the target object's follow list and user level profiles, join and calculate the total number of followers of the target object, as well as the number and proportion of followers at each level, i.e., follower-related features. It reads the target object's authentication unit and type, setting different authentication weights for different types; for example, authentication as a government agency or public institution is set to a weight of 1, while authentication as a regular enterprise is set to a weight of 0.4, etc., and uses these authentication weights as authentication-related features. Finally, it obtains the target object's nickname and the authentication entity name, calculates the overlap between the two texts, and generates an official entity account probability feature.

[0068] Step 402: When the dimension information is the content feature dimension of the article, extract the historical article list, original article tag, article category target tag, and article quality tag from the basic historical data.

[0069] Among them, the article content feature dimension can read the target object's historical article list, original article tag, article category target tag, and internally detected article quality level data, and statistically generate the target object's total historical article count, original article count, original article ratio, average article level, and proportion of each article level features.

[0070] The articles are aggregated by category, and the focus feature H of the target object's articles is generated by calculating the entropy value based on the proportion of each category. See Formula 1 for details. Formula 1 Where pi is the proportion of category i to the total number of categories.

[0071] Read the most recent posting time of the target object and convert it into the number of days from the current day, i.e., the duration of the hiatus.

[0072] Step 403: When the dimension information is the interactive data feature dimension, extract the number of reads, forwards, comments, and favorites from the basic historical data.

[0073] Among the interactive data feature dimensions, the system reads the number of reads, shares, comments, and favorites of the target object's historical posts, and summarizes and statistically generates the static historical total number of reads, shares, comments, and favorites for the target object. It also reads the user's reading, sharing, commenting, and favorites logs of posts about the target object within a recent period (optional 30 days), summarizes them using the target object as a keyword, and generates the recent dynamic reads, shares, comments, and favorites features for the target object.

[0074] Step 404: When the dimension information is a service capability feature dimension, extract the number of menus and the number of menu visits from the basic historical data.

[0075] Among the service capability characteristics, the menu content and access logs of the target object can be read to generate menu quantity and access volume characteristics. A dead link detection interface (or plugin) is called to detect links bound to the menu, generating menu availability characteristics. A dead link refers to a link where the server address has changed and the link at the current address cannot be found. Dead links can include both protocol dead links and content dead links. Protocol dead links can include those explicitly indicated by the page's TCP / HTTP protocol status, such as 404, 403, and 503 statuses. Content dead links can include pages where the server returns a normal status, but the content has been changed to something unrelated to the original content, such as "not found," "deleted," or "requires permissions."

[0076] If at least one of the search results contains an abnormal link, and a user clicks on that abnormal link but does not see the corresponding content, it will negatively impact the user's search experience. For example, a user searches for information on a search website, and the search results display several search links. If any of these search links are abnormal, it will negatively affect the user's experience using the search website. Similarly, a user searches for information using the search function of a social media app, and the search results display several search links. If any of these search links are abnormal, it will negatively affect the user's experience using the search function of the social media app.

[0077] In the application scenario of search, some of the search results obtained by users are invalid or have expired. Identifying and cleaning these invalid or expired results requires the platform to have the capability to identify web page link content. Webpage dead link identification is an important foundation for the survival of the platform, and the capability of identifying and filtering dead webpage content directly affects the ecology of the entire platform and user experience. Extracting features in the service capability dimension through step 404 can optimize the number of menus and the number of menu visits of the target object, and improve user experience.

[0078] Step 405: when the dimension information is the search performance feature dimension, extract the click-through rate parameter and the average click stay duration parameter from the basic historical data.

[0079] Wherein, in the search performance feature dimension, the search logs within the corresponding detection time period can be first separated from the basic historical data, the search logs for a period of time (which record the corresponding query when the target object is searched, and the target object ID of the user who searched the official account) are read, the data of <query, target object ID, exposed article ID, whether clicked, click stay duration> is obtained, aggregation is performed on the target object ID, and the click-through rate of exposed articles and the average click stay duration feature of the target object are generated.

[0080] So far, through step 401 to step 405, historical data of five different dimensions have been extracted from the basic historical data. However, for the use requirements of different search engines, the corresponding business requirements of the model can also be determined according to the quality index, and the target object feature vectors of different dimensions that need to be crossed are determined; feature crossing is performed on the target object feature vectors of different dimensions that need to be crossed to obtain a crossed feature vector. For example: in the "Look" search engine of an instant messaging client, when recommending information, the number of reads of the target object can be divided by the number of follower users to obtain the correlation between the number of reads and the number of follower users as the crossed feature vector. The crossed feature can better determine the relationship between the number of reads of the target object and the number of follower users, so as to screen target objects with a large number of followers and a large number of reads; on the contrary, in the recommendation of the "Search" search engine of a short video client, the ratio of the video playback duration to the number of complete playbacks can be used as the crossed feature vector to reflect the complete playback rate, so as to screen and recommend target objects with a high complete playback rate.

[0081] After extracting five different dimensions of historical data and crossed feature vectors from the basic historical data through step 401 to step 405, step 303 is continued to determine the training sample set.

[0082] Step 303: the quality index determining apparatus for a target object determines a first target object of a training sample according to search click logs of the target object.

[0083] In step 301 and step 302, the historical data of five different dimensions and the cross feature vectors are derived from all target objects, but only data of some target objects is required in the training phase of the quality indicator determination model. Therefore, through step 303, which target objects are the first target objects participating in training can be determined according to the search click logs of the target objects. In some embodiments of the present invention, determining the first target objects of training samples can be implemented in the following manner: screening random search terms and low-frequency search terms according to the search click logs of the target objects; determining second target objects corresponding to the random search terms; determining third target objects corresponding to the low-frequency search terms; combining the second target objects and the third target objects to obtain the first target objects of the training samples. Wherein, when reading the search click logs, <query, target object id, exposed article id, display position>, target objects to which topN multimedia information belongs are selected for labeling respectively from random queries and low-frequency queries (the definition of low frequency can be determined according to different search engine usage environments, for example, in the "Look" search engine of an instant messaging client, a weekly search volume less than 7 is low frequency, and in the "Search" search engine of a short video client, a weekly search volume less than 30 is low frequency). The labeling gear of quality indicators can be set according to business, for example, it is divided into 5 gears (respectively representing 0: severely low quality, 1: slightly low quality, 2: generally normal, 3: authoritative, 4: very authoritative). Wherein, the labeling volume can be flexibly adjusted according to the total number of target objects to meet the usage requirements of different search engines.

[0084] In some embodiments of the present invention, for servers with insufficient computing power, the search click logs of a target search engine often contain a large amount of useless information, which causes the process of screening random search terms and low-frequency search terms to consume a large amount of computing power and take a long time. To solve this problem, denoising can be performed on the search click logs of the target search engine to obtain search click data, and the search click data is used to screen random search terms and low-frequency search terms. The specific method includes: determining a target search engine matched with the quality indicator determination model; triggering a corresponding word segmentation library according to search term parameter information carried in the search click logs of the target search engine; performing word segmentation processing on the search click logs of the target search engine through the word dictionary of the triggered word segmentation library to form different word-level search terms and sentence-level search terms; determining search results respectively corresponding to different word-level search terms and sentence-level search terms; performing denoising processing on different word-level search terms and sentence-level search terms to form search click data corresponding to the search click logs of the target search engine. Wherein, in combination with the description of the foregoing embodiments, different terminal devices (such as the foregoing Figure 1The terminals 10-1 and / or 10-2 shown can provide a search bar for entering keywords on their respective search interfaces (e.g., web pages, information search apps, and search mini-programs in instant messaging clients), and a search button for performing data searches on those keywords. When a user enters a keyword in the search bar, and the terminal device detects a click on the search button, it triggers the server to initiate a corresponding word segmentation instruction. This word segmentation instruction carries the keyword from the search bar, and the server receives the instruction. Alternatively, the terminal device can display popular search keywords on the search interface. When a click on a popular search keyword is detected, the terminal device sends the word segmentation instruction to the server, carrying the popular search keyword, and the server receives the instruction. It should be noted that this embodiment of the invention does not limit the triggering method of the word segmentation instruction. (Refer to...) Figure 5 , Figure 5 This is a schematic diagram of the search effect of the target search engine in an embodiment of the present invention. When a user enters any search term, different search results (different types of multimedia information) can be displayed. When a user clicks to trigger any search result, they can access the corresponding public account and view the multimedia information created by the target object.

[0085] refer to Figure 6 , Figure 6 This is a schematic diagram of click data composed of search terms and corresponding search results in an embodiment of the present invention. The user's behavior data is obtained, and query-doc click data is constructed by associating query-doc pairs with user click behavior, through user logs stored on the server. Figure 6 As shown, the same document may be clicked by a user with different search queries, thus linking the query and the document. Here, the query refers to the search term entered by the user in the search system, which is usually a relatively short text; the document refers to the result returned by the search system, which can be a target object, such as a public account or mini program, or an article.

[0086] In some embodiments of the present invention, search results in the click data can be determined based on the sorting of the click data; according to the search results, the log information of the search engine is traversed, and search terms matching the search results are determined based on the behavior records of different target users; the search terms corresponding to the same search result are combined to determine the corresponding search term set. (See reference) Figure 7 , Figure 7This is a schematic diagram illustrating the click data structure in an embodiment of the present invention. For doc1, if it has been clicked by query1, query2, etc., then query1, query2, etc., are constructed into a query list. This ensures the comprehensiveness of the search terms corresponding to the same search result, provides more accurate training samples, and helps determine the quality indicators to improve the model's training accuracy.

[0087] In some embodiments of the present invention, the name of the segmentation library used for segmenting the search term text can also be determined. Based on the name of the segmentation library, the parameters of the segmentation library matching the word-level feature vector corresponding to the search term text are determined. Specifically, when determining the name of the segmentation library used for segmenting the search term text, the list of segmentation library identifiers stored in the cloud server can be traversed based on the search term parameter information carried by the search click logs of the target search engine. When the search term parameter information matches a segmentation library identifier in the list, it indicates that the segmentation library word dictionary corresponding to the current segmentation library identifier can perform segmentation processing on the click logs. At this time, the identifier of the segmentation library is recorded, and the name and parameters of the segmentation library are searched in the cloud server. The parameters of the segmentation library include: the type of segmentation library, the name of the segmentation library, and the version of the segmentation library. Since the word-level feature vectors generated when processing the same text information using different word segmentation libraries are not entirely the same, the parameters of the word segmentation library that match the word-level feature vectors corresponding to the search command text are determined based on the name of the word segmentation library. This determines the parameters of the word segmentation library used to segment the search command text. For example, if the search command text is "album XXX and songs by singer XXX", after processing with word segmentation library A, a set of word-level feature vectors A (album XXX; singer XXX's mp3) corresponding to the search command text is formed; after processing with word segmentation library B, a set of word-level feature vectors B (album XXX; singer XXX; mp3) corresponding to the search command text is formed; and after processing with word segmentation library A1, a set of word-level feature vectors A1 (album; XXX; singer; XXX; songs) corresponding to the search command text is formed.

[0088] Step 304: The target object quality index determination device obtains the training sample set corresponding to the first target object based on the target object feature vectors of different dimensions.

[0089] Step 305: The target object quality index determination device trains the quality index determination model based on the training sample set and determines the model parameters of the quality index determination model.

[0090] During training, a regression task was chosen over a classification task for the model training objective. The GBDT algorithm and regression objective were used, along with a training sample set and labels, for model training.

[0091] in, Figure 8 This diagram illustrates the model structure of the quality index determination model in this invention. The model employs a pairwise method to rank the objects to be recommended. Specifically, the pairwise method solves the ranking problem by approximating it as a classification problem, with each input sample being a label-document pair. For multiple result documents in a single query, any two documents are combined to form a document pair as input samples. This involves learning a binary classifier. For each input document pair A and B (the origin of the pairwise method), the binary classifier assigns a classification label of 1 or 0 based on whether A has a better relevance than B. By classifying all document pairs, a partial order relation is obtained, thus constructing the ranking relation for the entire document set. The principle of this type of method is to reduce the number of inverted document pairs in the ranking for a given entire document set S, thereby reducing ranking errors and optimizing the ranking results.

[0092] like Figure 8 As shown, the quality indicator determination model includes: a decision tree algorithm processing network and a regression classifier network. The decision tree algorithm network (GBDT Gradient Boosting Decision Tree) can be denoted as follows for a given dataset D with sample size n and variable dimension m:

[0093] Fitting the data using an additive ensemble tree model can be expressed as Equation 2: Formula 2; in Representing the function space A function that represents a tree model, containing information such as the specific tree structure and leaf nodes.

[0094] Then, the objective function is minimized, that is: , in, For regularization terms, denoted as , The number of leaf nodes. For leaf node coefficients, Indicates the difficulty of node splitting. Represents the L2 regularization coefficient. Let be the loss function, representing and The deviation between them.

[0095] Since we need a set of all tree models, but cannot obtain them all at once, we can adopt the following approach: First, fix the model obtained in the previous (t-1) training iteration. Then, in the next (t) training iteration, train based on the previously fixed result to obtain the corresponding t-th tree. Continue this process, training sequentially. The prediction result in the t-th iteration is expressed as in Formula 3: Formula 3, The objective function is:

[0096] Using a second-order Taylor expansion,

[0097] in , After removing the constant term, we get:

[0098] definition For the j-th leaf node, Expanding, we get:

[0099] Therefore, we can conclude that: ; The final objective function is:

[0100] The objective function This objective function serves as a standard to measure the quality of a tree structure; a smaller value indicates a better structure. The optimal split point is selected using this objective function to construct the classification and regression tree (CARTC).

[0101] Finally, for the Logistic Regression Classifier (LR), logistic regression is based on linear regression and applies a logistic function. For a two-class problem, its classification discriminant function is:

[0102] Where w represents the model parameters for determining the quality index.

[0103] For determining the model parameters of a quality index model, maximum likelihood estimation is used. This involves finding a set of parameters such that, under this set of parameters, the likelihood (probability) of the data is maximized. In a logistic regression model, likelihood can be expressed as:

[0104] Taking the logarithm yields the log-likelihood:

[0105] Using the above equation as the objective function, the parameters are solved using the gradient descent method.

[0106] Step 306: The target object quality index determination device predicts the quality index of all target objects through the quality index determination model, and obtains the quality index prediction results of all target objects.

[0107] Once the quality metrics are determined and the model training is complete, it can be deployed on the server of the instant messaging client or a server cluster. This allows for calculations on all target objects in the search engine, determining the tier information for each target object. It's important to note that during tier information calculation, the percentage of samples in each tier of the training set can be statistically analyzed. The predicted scores in the training set are then sorted in descending order, and the threshold for each tier is determined based on this percentage. For example, if the percentages of 0, 1, 2, 3, and 4 are 35%, 25%, 25%, 10%, and 5%, respectively, then the score at the 5th percentile after descending sorting is used as the threshold for dividing the 3rd and 4th tiers. This ensures that target objects exist in every tier (five tiers).

[0108] Finally, the recall strategy for multimedia information can be adjusted based on the tier information and quality index prediction results of each target object, and multimedia information can be recommended through the adjusted recall strategy. For example, target object A in an instant messaging client has the following characteristics: {Followers: 980,000, Percentage of high-quality followers: 60%, Authentication type: Public unit, Authentication weight: 1, Official account probability: 98%, Historical posts: 1767, Number of original articles: 1076, Post category focus: 76%, Average article views: 100,000, Number of reposts: 10,032, Number of comments: 30,022, ... (All other features mentioned above)}. When input into a trained quality index determination model, the predicted score is 4.6, placing it in the top 5% of target objects, classified as a tier 5 target object. Therefore, target object A has a tier 5 target object a.

[0109] The following sections explain the recall strategies for different types of multimedia information. Figure 9 This is a schematic diagram illustrating the effect of determining the quality index of the target object in an embodiment of the present invention. When adjusting the recall strategy for video news in the news recommendation field in a short video playback interface using the quality index determination model provided in this application, the effect is achieved through... Figure 8The quality metric determination model shown can recommend different video news to users watching video news in video news slots 1, 2, and 3. By dynamically adjusting the recall strategy, video news in the news recommendation field can be delivered in the order of video news A, B, and C. Specifically, when a user searches for video news "XXX", video news slot 1 displays video news A, video news slot 2 displays video news B, and video news slot 3 displays video news C. This ensures that users receive video news information produced by target objects with high predicted quality metrics, resulting in a better user experience and increasing click-through rates for better video news delivery.

[0110] like Figure 9 As shown, when playing video news, the exposure parameters during playback can be sent to the detection server so that the server can obtain these parameters. These exposure parameters are then used as evaluation parameters for the playback effect of multimedia information, and the target exposure parameters are searched based on the adjustment results of the recall strategy. For example, if the exposure parameters for video news slots 1, 2, and 3 are 100, 85, and 70 times respectively, the recommendation effect of the video news can be determined using these exposure parameters. Based on the adjustment results of the recall strategy, if the target exposure parameters for video news A, B, and C are 65, 75, and 102 times respectively, then video news A can be moved to video news slot 3, video news B to video news slot 2, and video news C to video news slot 1, flexibly meeting the needs of video news delivery.

[0111] At the same time, such as Figure 9 As shown, when dynamically adjusting the playback strategy for time-sensitive short videos, historical browsing information of the target industry's corresponding audience can be obtained; based on the historical browsing information of the target industry's corresponding audience, the exposure history of time-sensitive short videos corresponding to the historical browsing information is determined; based on the exposure history of time-sensitive short videos corresponding to the historical browsing information, the playback strategy for time-sensitive short videos is dynamically adjusted to achieve the desired effect. Figure 9 For example, due to different user preferences, video news from any target industry can be blocked. Therefore, if it is determined that video news B was previously blocked in the browsing history of viewer 1 corresponding to the target industry, video news A can be replaced by other video news information (such as video news C) by dynamically adjusting the playback strategy. Similarly, if it is determined that video news C was previously blocked in the browsing history of viewer 2 corresponding to the target industry, video news A can be replaced by other video news information (such as video news D) by dynamically adjusting the playback strategy. This is to conform to the usage habits of the target industry's corresponding viewers and provide users with a better user experience.

[0112] Beneficial technical effects: 1) This invention acquires basic historical data from all target objects; extracts feature vectors of target objects in different dimensions based on the basic historical data; determines the first target object of the training samples based on the search click logs of the target objects; obtains the training sample set corresponding to the first target object based on the feature vectors of the target objects in different dimensions; and trains the quality indicator determination model based on the training sample set to determine the model parameters of the quality indicator determination model. Thus, by using more granular feature extraction of target objects from different dimensions, the quality indicators of target objects can be more comprehensively characterized, making the quality indicators of target objects more in line with the needs of recall strategies.

[0113] 2) The quality index determination model predicts the quality indices of all target objects, resulting in predictions for all target objects' quality indices. This leads to higher accuracy in the predictions of target object quality indices, better identification of partial order relationships between different target objects, and more accurate recall of multimedia information based on the target object's quality indices.

[0114] 3) For servers with insufficient computing power, the search click logs of the target search engine can be denoised to obtain search click data. By using the search click data to filter random search terms and low-frequency search terms, the floating-point calculation load of the server can be effectively reduced, so that the method for determining the quality indicators of the target object provided in this application can be widely promoted.

[0115] The above are merely embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for determining the quality indicators of a target object, characterized in that, The method includes: Retrieve basic historical data from all target objects; Based on the aforementioned historical data, feature vectors of target objects in different dimensions are extracted. Identify a target search engine that matches the quality indicator determination model; Based on the search term parameter information carried in the search click logs of the target search engine, the corresponding word segmentation library is triggered; The search click logs of the target search engine are segmented using the triggered word segmentation dictionary to form different word-level search terms and sentence-level search terms; Determine the search results corresponding to the different word-level search terms and sentence-level search terms; The different word-level and sentence-level search terms are subjected to noise reduction processing to form search click data corresponding to the search click logs of the target search engine; Based on the search click data, filter random search terms and low-frequency search terms; determine the second target object corresponding to the random search terms; determine the third target object corresponding to the low-frequency search terms; The second target object and the third target object are combined to obtain the first target object of the training sample; Based on the target object feature vectors of different dimensions, obtain the training sample set corresponding to the first target object; Based on the training sample set, the quality index determination model is trained to determine the model parameters of the quality index determination model; The quality index determination model is used to predict the quality index of all target objects, and the prediction results of the quality index of all target objects are obtained.

2. The method according to claim 1, characterized in that, The step of extracting target object feature vectors of different dimensions based on the aforementioned historical data includes: Determine the different dimensions of the feature vector of the target object, wherein the different dimensions include at least one of the following: basic attribute feature dimension, article content feature dimension, interaction data feature dimension, service capability feature dimension, and search performance feature dimension; Based on the different dimensions of information, extract historical data of different dimensions from the basic historical data; The historical data of different dimensions are subjected to feature processing to obtain feature vectors of the target object in different dimensions.

3. The method according to claim 2, characterized in that, The method further includes: Based on the quality indicators, determine the business requirements corresponding to the model, and determine the target object feature vectors of different dimensions that need to be cross-referenced. The feature vectors of the target objects with different dimensions that need to be crossed are subjected to feature crossing to obtain the cross feature vector.

4. The method according to claim 2, characterized in that, The step of extracting historical data of different dimensions from the basic historical data based on the different dimensions of information includes: When the dimensional information is the basic attribute feature dimension, the data of the objects of interest, authentication data, and official entity number information are extracted from the basic historical data; When the dimensional information is the content feature dimension of the article, extract the historical article list, original article tag, article category target tag, and article quality tag from the basic historical data; When the dimensional information is an interactive data feature dimension, the number of reads, forwards, comments, and favorites are extracted from the basic historical data; When the dimension information is a service capability feature dimension, the number of menu items and the number of menu visits are extracted from the basic historical data. When the dimensional information is a search performance feature dimension, the click-through rate parameter and the average click dwell time parameter are extracted from the basic historical data.

5. The method according to claim 1, characterized in that, The method further includes: Determine the name of the word segmentation library used when segmenting the search term text; Based on the name of the word segmentation library, determine the parameters of the word segmentation library that match the word-level feature vector corresponding to the search term text, wherein the parameters of the word segmentation library include: The type of word segmentation library, the name of the word segmentation library, and the version of the word segmentation library.

6. The method according to claim 1, characterized in that, The method further includes: Based on the predicted quality indicators of the target objects, calculate the gear information for each target object; Based on the grade information of each target object and the prediction results of the quality indicators, the recall strategy for multimedia information is adjusted, and multimedia information is recommended through the adjusted recall strategy.

7. The method according to claim 6, characterized in that, The method further includes: When the multimedia information is a video advertisement The exposure parameters of the video advertisement during playback are sent to the detection server so that the detection server can obtain the exposure parameters of the video advertisement. The exposure parameter is used as an evaluation parameter for the playback effect of the multimedia information, and the target exposure parameter is found based on the adjustment result of the recall strategy.

8. The method according to claim 6, characterized in that, The method further includes: Obtain historical browsing information of the target industry's corresponding audience; Based on the historical browsing information of the audience corresponding to the target industry, determine the exposure history of multimedia information corresponding to the historical browsing information; Based on the exposure history of the multimedia information corresponding to the historical browsing information and the tier information of the target object, the recall strategy for the multimedia information is dynamically adjusted.

9. A device for determining the quality indicators of a target object, characterized in that, The device includes: The information transmission module is used to acquire basic historical data from all target objects; The information processing module is used to extract feature vectors of target objects in different dimensions based on the aforementioned basic historical data; The information processing module is further configured to: determine a target search engine that matches the quality indicator determination model; trigger a corresponding word segmentation library based on the search term parameter information carried in the search click logs of the target search engine; perform word segmentation processing on the search click logs of the target search engine using the triggered word segmentation library dictionary to form different word-level search terms and sentence-level search terms; determine the search results corresponding to the different word-level search terms and sentence-level search terms respectively; perform noise reduction processing on the different word-level search terms and sentence-level search terms to form search click data corresponding to the search click logs of the target search engine; filter random search terms and low-frequency search terms based on the search click data; determine a second target object corresponding to the random search terms; determine a third target object corresponding to the low-frequency search terms; and combine the second target object and the third target object to obtain the first target object of the training sample. The information processing module is further configured to obtain a training sample set corresponding to the first target object based on the target object feature vectors of different dimensions. The information processing module is further configured to train the quality index determination model based on the training sample set, and determine the model parameters of the quality index determination model. The information processing module is also used to predict the quality indicators of all target objects through the quality indicator determination model, and obtain the quality indicator prediction results of all target objects.

10. The apparatus according to claim 9, characterized in that, The information processing module is further configured to determine different dimensions of the feature vector of the target object, wherein the different dimensions include at least one of the following: basic attribute feature dimension, article content feature dimension, interaction data feature dimension, service capability feature dimension, and search performance feature dimension. The information processing module is also used to extract historical data of different dimensions from the basic historical data based on the different dimensions of information. The information processing module is also used to perform feature processing on the historical data of different dimensions to obtain feature vectors of target objects of different dimensions.

11. The apparatus according to claim 10, characterized in that, The information processing module is also used to determine the business requirements corresponding to the model based on the quality indicators, and to determine the target object feature vectors of different dimensions that need to be crossed. The information processing module is also used to perform feature crossing on the feature vectors of the target objects of different dimensions that need to be crossed, to obtain the cross feature vector.

12. The apparatus according to claim 10, characterized in that, The information processing module is also used to extract data of the object of interest, authentication data, and official entity number information from the basic historical data when the dimension information is the basic attribute feature dimension; The information processing module is also used to extract historical posting list, posting originality tag, posting category target tag, and posting quality tag from the basic historical data when the dimension information is the posting content feature dimension; The information processing module is also used to extract the number of reads, forwards, comments, and favorites from the basic historical data when the dimension information is an interactive data feature dimension; The information processing module is also used to extract the number of menus and the number of menu visits from the basic historical data when the dimension information is a service capability feature dimension. The information processing module is also used to extract click-through rate parameters and average click dwell time parameters from the basic historical data when the dimension information is a search performance feature dimension.

13. The apparatus according to claim 9, characterized in that, The information processing module is also used to determine the name of the word segmentation library used when performing word segmentation processing on the search term text; The information processing module is further configured to determine the parameters of a word segmentation library that matches the word-level feature vector corresponding to the search term text, based on the name of the word segmentation library. The parameters of the word segmentation library include: the type of word segmentation library, the name of the word segmentation library, and the version of the word segmentation library.

14. The apparatus according to claim 9, characterized in that, The information processing module is also used to calculate the gear information of each target object based on the quality index prediction results of the target object; The information processing module is also used to adjust the recall strategy for multimedia information based on the gear information of each target object and the prediction result of the quality index, and to recommend multimedia information through the adjusted recall strategy.

15. The apparatus according to claim 14, characterized in that, The information processing module is also used to send the exposure parameters of the video advertisement to the detection server when the multimedia information is a video advertisement, so that the detection server can obtain the exposure parameters of the video advertisement. The information processing module is also used to use the exposure parameter as an evaluation parameter for the playback effect of the multimedia information, and to find the target exposure parameter based on the adjustment result of the recall strategy.

16. The apparatus according to claim 14, characterized in that, The information processing module is also used to obtain the historical browsing information of the audience corresponding to the target industry; The information processing module is also used to determine the multimedia information exposure history corresponding to the historical browsing information based on the historical browsing information of the audience corresponding to the target industry. The information processing module is also used to dynamically adjust the recall strategy of the multimedia information based on the exposure history of the multimedia information corresponding to the historical browsing information and the tier information of the target object.

17. A computer program product comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by the processor, they implement the method for determining the quality indicators of the target object as described in any one of claims 1 to 8.

18. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable instructions; A processor, when executing executable instructions stored in the memory, implements the method for determining the quality indicators of a target object as described in any one of claims 1 to 8.

19. A computer-readable storage medium storing executable instructions, characterized in that, When the executable instructions are executed by the processor, they implement the method for determining the quality indicators of the target object as described in any one of claims 1 to 8.

Images