Model training and blog post quality evaluation method and device, equipment and storage medium

Abstract

This application provides a method, apparatus, device, and storage medium for training a model and evaluating blog post quality. The method includes: acquiring multiple training samples based on exposure logs of a blog post recommendation system; preprocessing the blog post information and corresponding blogger user information of each training sample to generate blog post-side behavioral features and user-side behavioral features corresponding to each training sample; refining the blog post-side behavioral features and user-side behavioral features to obtain revised blog post features and revised user features corresponding to each training sample; and adjusting the model parameters based on the blog post-side behavioral features, user-side behavioral features, revised blog post features, and revised user features corresponding to each training sample during iterative training of the initial blog post quality evaluation model using multiple training samples, to obtain the trained blog post quality evaluation model. This application enables more accurate blog post recommendations.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, device, and storage medium for training a model and evaluating blog quality. Background Technology

[0002] Currently, recommender systems are one of the most effective ways to solve information overload. In relational stream distribution scenarios, recommender systems generally consist of two stages: the recall stage and the ranking stage. In the recall stage, multiple methods such as user profiling, collaborative filtering, graph networks, and approximate search are used to fully explore blog posts that users may be interested in, and the blog posts recalled from multiple paths are truncated and merged according to time order or serpentine sorting. In the ranking stage, deep learning-based recommendation algorithms provide personalized recommendations to the recalled blog posts to stimulate users' consumption desires.

[0003] However, by fully considering the multiple interests of users and the changes in interests, the blog posts recalled through multiple links have serious dimensional problems in the quality calculation results of blog posts in different interest areas due to the different interest areas and user bases they target. This affects the accurate recommendation of blog posts. Summary of the Invention

[0004] This application provides a method, apparatus, device, and storage medium for training a model and evaluating blog post quality, in order to achieve more accurate blog post recommendations.

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

[0006] This application provides a training method for a blog post quality assessment model. The method includes: acquiring multiple training samples based on exposure logs from a blog post recommendation system; each training sample includes blog post information and the corresponding user information of the blogger; the blogger is the user who published the sample blog post; preprocessing the blog post information and the corresponding user information of the blogger in each training sample to generate blog post-side behavioral features and user-side behavioral features corresponding to each training sample; wherein the blog post-side behavioral features characterize the interaction of the sample blog post within a preset first time period; and the user-side behavioral features characterize the posting activity of the blogger within a preset second time period. The interaction status of blog posts is analyzed; by modifying the blog post-side behavioral features and user-side behavioral features, modified blog post features and user-side modified features are obtained for each training sample; the confidence level of the modified blog post features is higher than that of the blog post-side behavioral features, and the confidence level of the user-side modified features is higher than that of the user-side behavioral features; during the iterative training of the initial blog post quality assessment model using multiple training samples, the model parameters are adjusted based on the blog post-side behavioral features, user-side behavioral features, modified blog post features, and user-side modified features corresponding to each training sample until the blog post quality assessment model converges, resulting in the trained blog post quality assessment model; the blog post quality assessment model is used to assess the quality of blog posts.

[0007] In some possible implementations, the above method further includes: after acquiring multiple training samples, preprocessing the blog post information of the sample blog posts in each training sample and the user information of the corresponding sample blogger to generate short-term features of the blog posts and user information features corresponding to each training sample; the short-term features of the blog posts are used to characterize the interaction of the sample blog posts in a preset third time period, the third time period being shorter than the first time period, and the user information features are used to characterize the user profile of the sample blogger.

[0008] In some possible implementations, during the iterative training of the initial blog post quality assessment model using multiple training samples, the model parameters are adjusted based on the blog post-side behavioral features, user-side behavioral features, blog post correction features, and user correction features corresponding to each training sample until the blog post quality assessment model converges, resulting in a trained blog post quality assessment model. This includes: determining the labeling information corresponding to each sample blog post based on the blog post-side behavioral features, user-side behavioral features, blog post correction features, and user correction features corresponding to each training sample; the labeling information is used to label the sample blog post as a positive or negative sample; inputting the short-term features and user information features corresponding to each training sample into the blog post quality assessment model to obtain the quality assessment result output by the blog post quality assessment model; the quality assessment result includes a first quality score corresponding to the sample blog post and a prediction result indicating whether the sample blog post is a positive or negative sample; the first quality score is used to characterize the user's degree of interest in the sample blog post; obtaining a comparison result by comparing the prediction result and labeling information corresponding to each sample blog post; and tuning the parameters of the blog post quality assessment model based on the comparison result of each sample blog post until the blog post quality assessment model converges.

[0009] In some possible implementations, based on the blog post-side behavior features, user-side behavior features, blog post correction features, and user correction features corresponding to each training sample, the labeling information corresponding to each sample blog post is determined, including: performing linear fusion processing on the blog post-side behavior features, user-side behavior features, blog post correction features, and user correction features corresponding to each training sample to obtain the initial quality score corresponding to the sample blog post in each training sample; the initial quality score is used to characterize the degree of user interest in the sample blog post without processing by the blog post quality assessment model; if the initial quality score of the sample blog post meets a preset threshold, the sample blog post is labeled as a positive sample; if the initial quality score of the sample blog post does not meet the preset threshold, the sample blog post is labeled as a negative sample.

[0010] In some possible implementations, the blog post-side behavioral features and user-side behavioral features are modified to obtain the modified blog post features and user-side modified features corresponding to each training sample. This includes: modifying the credibility of the blog post-side behavioral features according to the Wilson confidence interval to obtain the modified blog post features corresponding to each training sample; and modifying the credibility of the user-side behavioral features according to the Wilson confidence interval to obtain the modified user-side features corresponding to each training sample.

[0011] This application provides a method for evaluating the quality of blog posts. The method includes: obtaining multiple historical blog posts that have been exposed and the bloggers who published them based on the exposure logs of a blog post recommendation system; inputting the blog post information of each historical blog post and the user information of its corresponding historical blogger into a trained blog post quality evaluation model to obtain a second quality score for each historical blog post output by the model; training the blog post quality evaluation model using the training method described above, with the second quality score representing the degree of user interest in the historical blog posts; determining target blog posts among the multiple exposed historical blog posts based on the second quality score; and using the target blog posts for coarse or fine ranking.

[0012] In some possible implementations, the target blog post is determined from multiple exposed historical blog posts based on the second quality score, including: calculating a third quality score for each historical blog post according to a preset decay mapping relationship, a preset time period, and the second quality score; the third quality score is used to characterize the user's interest in the historical blog post after removing the time effect; the decay mapping relationship is the mapping relationship between the second quality score and the third quality score; the blog post information of each historical blog post and the third quality score corresponding to each historical blog post are associated and saved to the trained blog post quality assessment model; and the target blog post is determined from multiple exposed historical blog posts based on the third quality score.

[0013] In some possible implementations, the target blog post is determined from multiple historical blog posts based on the third quality score, including: selecting multiple recalled blog posts from multiple historical blog posts through multi-link recall; obtaining the third quality score corresponding to each recalled blog post in the trained blog post quality assessment model; and selecting the target blog post from the multiple recalled blog posts based on the third quality score corresponding to each recalled blog post.

[0014] This application provides a training device for a blog post quality assessment model. The device includes: a sample data acquisition module for acquiring multiple training samples based on exposure logs from a blog post recommendation system; wherein each training sample includes blog post information of a sample blog post and user information of its corresponding blogger, the blogger being the user who published the sample blog post; and a sample data preprocessing module for preprocessing the blog post information and corresponding blogger user information of each training sample to generate blog post-side behavioral features and user-side behavioral features corresponding to each training sample; wherein the blog post-side behavioral features characterize the interaction of the sample blog post within a preset first time period; and the user-side behavioral features characterize the interaction of the sample blogger within a preset second time period. The system tracks the interaction with published blog posts; a sample data processing module is used to refine the blog post-side behavioral features and user-side behavioral features to obtain modified blog post features and user-side modified features for each training sample; the confidence level of the modified blog post features is higher than that of the blog post-side behavioral features, and the confidence level of the user-side modified features is higher than that of the user-side behavioral features; a training module is used to iteratively train the initial blog post quality assessment model using multiple training samples, adjusting the model parameters based on the blog post-side behavioral features, user-side behavioral features, modified blog post features, and user-side modified features for each training sample until the blog post quality assessment model converges, resulting in a trained blog post quality assessment model; the blog post quality assessment model is used to assess the quality of blog posts.

[0015] In some possible implementations, the sample data preprocessing module is further used to preprocess the blog post information of the sample blog posts in each training sample and the user information of the corresponding sample blogger after acquiring multiple training samples, so as to generate short-term features of the blog posts and user information features corresponding to each training sample; the short-term features of the blog posts are used to characterize the interaction of the sample blog posts in a preset third time period, which is shorter than the first time period, and the user information features are used to characterize the user profile of the sample blogger.

[0016] In some possible implementations, a training module is used to determine the labeling information corresponding to each sample blog post based on the blog post-side behavior features, user-side behavior features, blog post correction features, and user correction features corresponding to each training sample; the labeling information is used to label the sample blog post as a positive or negative sample; the short-term features of the blog post and user information features corresponding to each training sample are input into the blog post quality assessment model to obtain the quality assessment result output by the blog post quality assessment model; the quality assessment result includes a first quality score corresponding to the sample blog post and a prediction result of whether the sample blog post is a positive or negative sample; the first quality score is used to characterize the degree of user interest in the sample blog post; a comparison result is obtained by comparing the prediction result and labeling information corresponding to each sample blog post; based on the comparison result of each sample blog post, the parameters of the blog post quality assessment model are tuned until the blog post quality assessment model converges.

[0017] In some possible implementations, the training module is used to perform linear fusion processing on the blog post-side behavior features, user-side behavior features, blog post correction features, and user correction features corresponding to each training sample to obtain the initial quality score corresponding to the sample blog post in each training sample; the initial quality score is used to characterize the degree of user interest in the sample blog post without processing by the blog post quality assessment model; if the initial quality score of the sample blog post meets a preset threshold, the sample blog post is marked as a positive sample; if the initial quality score of the sample blog post does not meet the preset threshold, the sample blog post is marked as a negative sample.

[0018] In some possible implementations, the sample data processing module is used to perform confidence correction on the blog post-side behavioral features based on the Wilson confidence interval, to obtain the blog post correction features corresponding to each training sample; and to perform confidence correction on the user-side behavioral features based on the Wilson confidence interval, to obtain the user correction features corresponding to each training sample.

[0019] This application provides a blog post quality assessment device, comprising: a data acquisition module for acquiring multiple historical blog posts that have been exposed and the historical bloggers who published them, based on exposure logs from a blog post recommendation system; a data processing module for inputting the blog post information of each historical blog post and the user information of its corresponding historical blogger into a trained blog post quality assessment model to obtain a second quality score for each historical blog post output by the blog post quality assessment model; the blog post quality assessment model is trained using the same training method as the blog post quality assessment model, and the second quality score is used to characterize the degree of user interest in the historical blog posts; and a filtering module for determining target blog posts from multiple exposed historical blog posts based on the second quality score; the target blog posts are used for coarse ranking or fine ranking of blog posts.

[0020] In some possible implementations, the filtering module is used to calculate a third quality score for each historical blog post based on a preset decay mapping relationship, a preset time period, and a second quality score; the third quality score is used to characterize the user's interest in the historical blog post after removing the influence of time period; the decay mapping relationship is the mapping relationship between the second quality score and the third quality score; the blog post information of each historical blog post and the third quality score corresponding to each historical blog post are associated and saved to the trained blog post quality evaluation model; and the target blog post is determined from multiple exposed historical blog posts based on the third quality score.

[0021] In some possible implementations, the filtering module is used to filter out multiple recalled blog posts from multiple exposed historical blog posts through multi-link recall; obtain the third quality score corresponding to each recalled blog post in the trained blog post quality assessment model; and select the target blog post from the multiple recalled blog posts based on the third quality score corresponding to each recalled blog post.

[0022] This application provides an electronic device, including: a memory for storing executable instructions; and a processor for executing the executable instructions stored in the memory to implement the method provided in this application.

[0023] This application provides a computer storage medium storing executable instructions, which, when executed by a processor, implement the training method or the blog post quality assessment method of the blog post quality assessment model provided in this application.

[0024] This application provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the training method or blog quality assessment method of the blog post quality assessment model provided in this application.

[0025] This application has the following beneficial effects:

[0026] In this application, by modifying the blog post-side behavioral features and user-side behavioral features, modified blog post features and user-side modified features are obtained for each training sample. The confidence level of the modified blog post features is higher than that of the blog post-side behavioral features, and the confidence level of the modified user features is higher than that of the user-side behavioral features. This can solve the problem of inconsistent dimensions between sample blog posts, achieve consistency of training samples, provide a more accurate data processing foundation for the blog post recommendation process, and thus achieve more accurate blog post recommendations.

[0027] Furthermore, based on the preset attenuation mapping relationship, preset time limit, and second quality score, the third quality score corresponding to each historical blog post is calculated. This allows for the correction of blog posts with strong time sensitivity, providing a more intuitive representation of the quality of historical blog posts. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the architecture of the recommendation system provided in the embodiments of this application;

[0029] Figure 2 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application;

[0030] Figure 3 This is an optional flowchart illustrating the training method of the blog post quality assessment model provided in this application embodiment;

[0031] Figure 4 This is an optional flowchart illustrating the training method and the blog post quality assessment method provided in the embodiments of this application. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0033] 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.

[0034] If the application documents contain similar descriptions such as "first / second", the following explanation shall be added: In the following description, the terms "first / second / third" are used only to distinguish similar objects and do not represent a specific order of objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0036] With the widespread adoption of smart devices and the rapid development of internet technology, modern society has fully entered the mobile internet era. As the Web 2.0 model becomes mainstream, internet users are not only consumers of online content but also producers and disseminators of online information. The amount of information on the internet is exploding; for example, social media platforms like Weibo can see over 100 million posts daily. Proactively recommending high-quality information that aligns with users' interests is an indispensable and even crucial part of the entire operation when faced with this massive amount of information.

[0037] Currently, recommender systems are one of the most effective ways to solve information overload. In relational stream distribution scenarios, recommender systems generally consist of two stages: the recall stage and the ranking stage. In the recall stage, multiple methods such as user profiling, collaborative filtering, graph networks, and approximate search are used to fully explore blog posts that users may be interested in, and the blog posts recalled from multiple paths are truncated and merged according to time order or serpentine sorting. In the ranking stage, deep learning-based recommendation algorithms provide personalized recommendations to the recalled blog posts to stimulate users' consumption desires.

[0038] However, by fully considering users' multiple interests and changes in interests and recalling blog posts through multiple channels, the quality calculation results of blog posts recalled through different strategies have serious dimensional problems in measuring the degree of user preference, thus affecting the accurate recommendation of blog posts.

[0039] To address the aforementioned issues, this application provides a training method for a blog post quality assessment model to achieve more accurate blog post recommendations.

[0040] The following describes exemplary applications of the electronic devices provided in the embodiments of this application. These electronic devices can be implemented as various types of user terminals such as laptops, tablets, desktop computers, and mobile devices (e.g., mobile phones, wearable smartwatches, dedicated messaging devices), or as servers. The following will describe exemplary applications when the electronic device is implemented as a server.

[0041] See Figure 1 , Figure 1 This is a schematic diagram of the architecture of the recommendation system provided in the embodiments of this application. Electronic devices (electronic devices 40-1 and 40-2 are shown as examples) are connected to server 200 through network 300. Network 300 can be a wide area network or a local area network, or a combination of both.

[0042] In some possible implementations, user A can publish material information A through terminal 40-1, and user B can publish material information B through terminal 40-2. Material information A and material information B are uploaded to server 200 via network 300. Server 200 can store material information A and material information B in database 500. To achieve precise material recommendation based on user interests, server 200 can sort the materials and push them to the user with the highest degree of matching with the user's interests. The materials pushed to the user can be displayed on the graphical interface of electronic device 400 (graphical interfaces 41-1 and 41-2 are shown as examples).

[0043] In some embodiments, server 200 may be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, and big data and artificial intelligence platforms. The terminal may be a smartphone, tablet, laptop, desktop computer, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited in this embodiment of the invention.

[0044] See Figure 2 , Figure 2 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Figure 2 The illustrated electronic device 400 can be the aforementioned terminal and / or server 200. Electronic device 400 includes: at least one processor 410, memory 450, at least one network interface 420, and user interface 430. The various components in electronic device 400 are coupled together via a bus system 440. It is understood that the bus system 440 is used to implement communication between these components. In addition to a data bus, the bus system 440 also includes a power bus, a control bus, and a status signal bus. However, for clarity, ... Figure 2 The general labeled all buses as Bus System 440.

[0045] Processor 410 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0046] User interface 430 includes one or more output devices 431 that enable the presentation of media content, including one or more speakers and / or one or more visual displays. User interface 430 also includes one or more input devices 432, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0047] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state storage, hard disk drives, optical disk drives, etc. The memory 450 may optionally include one or more storage devices physically located away from the processor 410.

[0048] The memory 450 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), and the volatile memory may be random access memory (RAM). The memory 450 described in this application embodiment is intended to include any suitable type of memory.

[0049] In some embodiments, memory 450 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, as illustrated below.

[0050] Operating system 451 includes system programs for handling various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, driver layer, etc., for implementing various basic business functions and handling hardware-based tasks;

[0051] The network communication module 452 is used to reach other computing devices via one or more (wired or wireless) network interfaces 420, exemplary network interfaces 420 including: Bluetooth, WiFi, and Universal Serial Bus (USB), etc.

[0052] Presentation module 453 is configured to enable the presentation of information (e.g., a user interface for operating peripheral devices and displaying content and information) via one or more output devices 431 (e.g., a display screen, a speaker, etc.) associated with user interface 430;

[0053] The input processing module 454 is used to detect and translate one or more user inputs or interactions from one or more input devices 432.

[0054] In some embodiments, the training apparatus for the blog post quality assessment model provided in this application can be implemented in software. Figure 2 The training device 455 and the blog post quality assessment device 456, which are stored in memory 450, are shown. They can be software in the form of programs and plug-ins. The training device 455 includes the following software modules: sample data acquisition module 4551, sample data preprocessing module 4552, sample data processing module 4553, and training module 4554. The blog post quality assessment device 456 includes the following software modules: data acquisition module 4561, data processing module 4562, and screening module 4563.

[0055] These modules are logically structured, and therefore can be combined or further broken down arbitrarily according to their implemented functions. The functions of each module will be explained below.

[0056] In other embodiments, the training device and the evaluation device for the blog post quality assessment model provided in this application can be implemented in hardware. As an example, the device provided in this application can be a processor in the form of a hardware decoding processor, which is programmed to execute the training method and the evaluation method for the blog post quality assessment model provided in this application. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0057] In some embodiments, the recommendation system provided in this application can be a blog post recommendation system. The training method of the blog post quality assessment model provided in this application is described below using a blog post recommendation system as an example.

[0058] The training method of the blog post quality assessment model provided in this application embodiment will be described below with reference to exemplary applications and implementations of the electronic devices provided in this application embodiment.

[0059] See Figure 3 , Figure 3 This is a schematic diagram of an optional flowchart of the training method for the blog post quality assessment model provided in this application embodiment. The following will combine... Figure 3 The steps shown illustrate the training method for the blog post quality assessment model.

[0060] It should be noted that, for ease of explanation, the training device for the blog post quality assessment model can be simply referred to as the training device.

[0061] S301. Based on the exposure logs of the blog post recommendation system, obtain multiple training samples. Each training sample includes the blog post information of the sample blog post and the user information of the corresponding sample blogger; the sample blogger is the user who published the sample blog post.

[0062] Understandably, the training device can access the exposure logs of the blog post recommendation system to collect blog post information and user information of the bloggers who published the sample blog posts, and use these as training samples. During the collection process, to meet the data volume requirements for training data, blog post information from multiple sample blog posts can be collected according to actual needs, and correspondingly, user information of the bloggers corresponding to multiple sample blog posts will also be collected. A sample blogger can publish only one sample blog post or multiple sample blog posts; this embodiment does not specifically limit this.

[0063] In some embodiments, the blog post information of multiple sample blog posts and the user information of their corresponding sample bloggers can be stored in the database of the blog post recommendation system.

[0064] S302, preprocess the blog post information and the corresponding user information of the sample blogger in each training sample to generate blog post-side behavioral features and user-side behavioral features for each training sample. The blog post-side behavioral features characterize the interaction of the sample blog post within a preset first time period; the user-side behavioral features characterize the interaction of the blog posts published by the sample blogger within a preset second time period.

[0065] Understandably, the training device can obtain blog post information and corresponding user information of sample bloggers from the database of the blog post recommendation system. Then, it performs data preprocessing on the blog post information and user information to obtain blog post-side behavioral features and user-side behavioral features. Specifically, each sample blog post corresponds to one blog post-side behavioral feature, and each sample blogger corresponds to one user-side behavioral feature; since each sample blogger corresponds to one sample blogger, correspondingly, each sample blog post also corresponds to one user-side behavioral feature.

[0066] In some embodiments, blog post-side behavioral features are used to characterize the interaction of sample blog posts within a preset first time period; user-side behavioral features are used to characterize the interaction of blog posts published by sample bloggers within a preset second time period.

[0067] In one example, taking blog post A as the sample blog post, blogger A as the sample blogger, and a preset first time period of three days and a preset second time period of one month as an example, the blog post-side behavioral characteristics of blog post A represent the interaction data of blog post A over the past three days, including exposure, clicks, favorites, reposts, comments, likes, and viewing time. The user-side behavioral characteristics of blogger A who published blog post A represent the interaction data of blogger A's historical blog posts over the past month, including exposure, clicks, favorites, reposts, comments, likes, and viewing time. It should be noted that in the above example, the preset first time period of three days and the preset second time period of one month are only one example. An appropriate time length can be selected according to actual needs, and this application embodiment does not specifically limit this.

[0068] In some possible implementations, after acquiring multiple training samples, the training method of the above-mentioned blog post quality assessment model may further include: preprocessing the blog post information and the corresponding user information of the sample blogger in each training sample to generate short-term features of the blog post and user information features corresponding to each training sample. The short-term features of the blog post are used to characterize the interaction of the sample blog post within a preset third time period, which is shorter than the first time period. The user information features are used to characterize the user profile of the sample blogger.

[0069] Understandably, the training device can obtain blog post information and corresponding user information of sample bloggers from the database of the blog post recommendation system. Then, it performs data preprocessing on the blog post information and user information to obtain short-term features of the blog posts and user information features. Each sample blog post corresponds to one short-term feature and one user information feature.

[0070] In one example, taking blog post B as the sample blog post, blogger B as the sample blogger, and a preset third time period of one hour as an example, the short-term features of blog post B represent the interaction data of blog post B in the past hour, including exposure, clicks, favorites, reposts, comments, likes, and viewing time. The user information features of blogger B who published blog post B represent the user profile of blogger B, including the number of followers, interests, and active periods. It should be noted that the short-term features of the blog post refer to the short-term characteristics of the sample blog post, while the blog post-side behavioral features refer to the long-term characteristics of the sample blog post. Therefore, the third time period corresponding to the short-term features of the blog post is shorter than the first time period corresponding to the blog post-side behavioral features.

[0071] S303, by modifying the blog post-side behavioral features and user-side behavioral features, modified blog post features and modified user features are obtained for each training sample. The confidence level of the modified blog post features is higher than that of the blog post-side behavioral features, and the confidence level of the modified user features is higher than that of the user-side behavioral features.

[0072] Understandably, in blog post recommendation systems, sample blog posts with different content will correspond to different interest areas. For example, sample blog post A corresponds to the interest area of ​​"entertainment celebrities," sample blog post B corresponds to the interest area of ​​"finance," and sample blog post C corresponds to the interest area of ​​"pets." Compared to the interest area of ​​"entertainment celebrities," "finance" and "pets" are more niche, targeting a smaller audience (i.e., users other than the sample bloggers in the blog post recommendation system). In this case, the interaction data such as exposure, clicks, favorites, reposts, comments, likes, and viewing time for "finance" and "pets" will be significantly lower than those for "entertainment celebrities," thus affecting the recommendation of sample blog posts B and C by the blog post recommendation system. Therefore, it is necessary to modify the blog post-side behavioral features and user-side behavioral features to obtain blog post modified features and user modified features with higher confidence levels.

[0073] In some possible implementations, S303 may include: adjusting the credibility of blog post-side behavioral features based on Wilson confidence intervals to obtain blog post-corrected features for each training sample; and adjusting the credibility of user-side behavioral features based on Wilson confidence intervals to obtain user-corrected features for each training sample.

[0074] Understandably, by using the pre-defined Wilson confidence interval formula, the credibility of blog post-side behavioral features and user-side behavioral features can be corrected to obtain blog post-corrected features and user-corrected features.

[0075] In this embodiment, Wilson confidence intervals are used to correct the confidence of blog post-side behavioral features and user-side behavioral features. This can provide better confidence interval estimates for small sample data (i.e., sample blog posts and sample bloggers in niche interest areas), effectively alleviating the problem of feature dilution of sample blog posts and sample bloggers in niche interest areas.

[0076] In some embodiments, the Wilson confidence interval is used to modify the blog post behavior features to obtain the expression (1) of the modified blog post features as follows:

[0077]

[0078] In the formula, For the corrected features of the i-th blog post (i.e., the corrected features of the blog post corresponding to the i-th sample blog post), p i Let n represent the behavior rate corresponding to the behavior feature of the i-th blog post. i Let z represent the exposure volume corresponding to the behavioral characteristics of the i-th blog post, and z is a confidence level statistic under a preset normal distribution; where behavioral rates include: click-through rate, view rate, interaction rate, etc.

[0079] In some embodiments, z in expression (1) can be obtained by a preset standard normal distribution table, for example, at a 95% confidence level, z takes the value of 1.96.

[0080] In some embodiments, the expression (2) for calculating the behavior rate in the i-th blog post behavior feature is as follows:

[0081] p i =x i / n i (2)

[0082] In the formula, x i This represents the number of behaviors corresponding to the behavioral characteristics of the i-th blog post; where the number of behaviors includes: clicks, views, interactions, etc.

[0083] In this embodiment of the application, when a sample blog post has just been published, there is no exposure or corresponding interaction yet. In order to evaluate the quality of the sample blog post, user-side behavioral features and user-correction features can be used to evaluate the quality of the newly published sample blog post, so as to effectively alleviate the cold start problem caused by the lack of blog post-side features (i.e., blog post-side behavioral features and blog post correction features).

[0084] S304, during the iterative training of the initial blog post quality assessment model using multiple training samples, the model parameters are adjusted based on the blog post-side behavioral features, user-side behavioral features, blog post correction features, and user correction features corresponding to each training sample, until the blog post quality assessment model converges, resulting in the trained blog post quality assessment model. This model is used to assess the quality of blog posts.

[0085] Understandably, once the blog post behavior characteristics, user behavior characteristics, blog post correction characteristics, and user correction characteristics are determined, a blog post quality assessment model can be trained based on these characteristics.

[0086] In some possible implementations, S304 above may include: determining labeling information for each sample blog post based on the blog post-side behavior features, user-side behavior features, blog post correction features, and user correction features corresponding to each training sample; the labeling information is used to label the sample blog post as a positive or negative sample; inputting the short-term features of the blog post and user information features corresponding to each training sample into the blog post quality assessment model to obtain the quality assessment result output by the blog post quality assessment model; the quality assessment result includes a first quality score corresponding to the sample blog post and a prediction result for whether the sample blog post is a positive or negative sample; the first quality score is used to characterize the degree of user interest in the sample blog post; obtaining a comparison result by comparing the prediction result and labeling information corresponding to each sample blog post; and tuning the parameters of the blog post quality assessment model based on the comparison result of each sample blog post until the blog post quality assessment model converges.

[0087] Understandably, each sample blog post can be scored based on blog post-side behavioral features, user-side behavioral features, blog post correction features, and user correction features. Then, based on the scoring results (i.e., the initial quality score), the sample blog posts are divided into positive and negative samples. Next, the short-term features and user information features corresponding to each training sample are input into the blog post quality assessment model to obtain the quality assessment result output by the model. The quality assessment result includes the prediction result for the sample blog post; by comparing the prediction result with the labeling information, a comparison result is obtained. Based on the comparison result, the parameters of the blog post quality assessment model are adjusted until a fully trained blog post quality assessment model is obtained.

[0088] In some possible implementations, determining the labeling information for each sample blog post based on the blog post-side behavior features, user-side behavior features, blog post correction features, and user correction features corresponding to each training sample may include: performing linear fusion processing on the blog post-side behavior features, user-side behavior features, blog post correction features, and user correction features corresponding to each training sample to obtain the initial quality score corresponding to the sample blog post in each training sample; wherein, the initial quality score is used to characterize the degree of user interest in the sample blog post without processing by the blog post quality assessment model; if the initial quality score of the sample blog post meets a preset threshold, the sample blog post is labeled as a positive sample; if the initial quality score of the sample blog post does not meet the preset threshold, the sample blog post is labeled as a negative sample.

[0089] Understandably, by using linear fusion processing to assign corresponding weights to each feature and then performing summation calculations, it is possible to score each sample blog post and obtain the scoring result (i.e., the initial quality score); the initial quality scores are then sorted, and the multiple sample blog posts are divided into positive and negative samples based on the sorting results.

[0090] In some embodiments, the expression (3) for linearly fusing blog post behavior features, user behavior features, blog post correction features, and user correction features is as follows:

[0091] S i =αM wi +βM hi +γA hi +δA wi (3)

[0092] In the formula, S i M represents the initial quality score of the i-th sample blog post. wi M represents the modified feature of the i-th blog post. hi Let A represent the behavioral characteristics of the i-th blog post. hi Let A represent the i-th user-side behavioral feature. wi Let α represent the user-modified feature, β represent the linear fusion weight coefficient of the corresponding blog post modified feature, γ represent the linear fusion weight coefficient of the corresponding blog post-side behavior feature, and δ represent the linear fusion weight coefficient of the corresponding user-side behavior feature.

[0093] In some embodiments, α, β, γ and δ are preset linear fusion weight coefficients with a value range of [0.1]. The values ​​of the linear fusion weight coefficients can be adjusted according to actual needs. For example, α, β, γ and δ can be 0.2, 0.5, 0.2 and 0.1 respectively. This application embodiment does not specifically limit the values ​​of the linear fusion weight coefficients.

[0094] In some embodiments, the above-mentioned marking of a sample blog post as a positive sample when its initial quality score meets a preset threshold, and marking it as a negative sample when its initial quality score does not meet the preset threshold, may include: sorting multiple initial quality scores according to an equidistant binning and percentile delineation method; marking sample blog posts with initial quality scores greater than the 80th percentile as positive samples; and marking sample blog posts with initial quality scores less than the 80th percentile as negative samples.

[0095] In some embodiments, sample labeling may further include difficult positive samples and difficult negative samples. Therefore, the above-mentioned labeling of sample blog posts as positive samples when their initial quality scores meet a preset threshold, and labeling them as negative samples when their initial quality scores do not meet the preset threshold, may include: sorting multiple initial quality scores according to equidistant binning and percentile delineation; labeling sample blog posts with initial quality scores greater than the 80th percentile as positive samples; randomly selecting 15% of sample blog posts between the 70th and 80th percentiles and labeling them as difficult positive samples, and labeling the remaining 85% of sample blog posts between the 70th and 80th percentiles as difficult negative samples; and labeling sample blog posts with initial quality scores less than the 70th percentile as negative samples.

[0096] In this embodiment of the application, by constructing difficult positive samples and difficult negative samples, the learning difficulty of the microblog quality assessment model can be effectively increased, overfitting can be prevented, and the generalization ability of the microblog quality assessment model can be greatly improved.

[0097] In some embodiments, after determining positive and negative samples, the short-term features and user information features corresponding to each sample blog post can be input into the blog post quality assessment model to obtain the quality assessment result output by the model. The quality assessment result includes a first quality score corresponding to the sample blog post and a prediction result indicating whether the sample blog post is a positive or negative sample. The first quality score is used to characterize the user's level of interest in the sample blog post. After obtaining the quality assessment result, it is determined whether the prediction result of the sample blog post is the same as the labeling information of the sample blog post to obtain a comparison result. Based on the comparison result, the parameters of the blog post quality assessment model are adjusted until the model converges.

[0098] In this embodiment, positive and negative samples are determined based on blog post behavior features, user behavior features, blog post correction features, and user correction features. When using short-term blog post features and user information features as training data for the blog post quality assessment model, since short-term blog post features and user information features do not participate in the division of positive and negative samples, there will be no feature leakage problem in model training.

[0099] In some embodiments, the blog post quality assessment model can be a recommendation model, such as the DeepFM (deepfactorization machine) model. The DeepFM model not only retains the strong memory ability of traditional models and strengthens the feature combination of shallow networks, but also integrates the advantages of the generalization ability of deep learning models. Furthermore, the DeepFM model has a simple structure and is relatively easy to train and deploy in engineering, making it suitable for widespread application in recommendation scenarios.

[0100] In some embodiments, the DeepFM model is used as an example for blog post quality assessment. The DeepFM model includes an FM layer, a Deep layer, a fully connected layer, and an activation function layer. During training, the outputs of the FM layer and the Deep layer of the DeepFM model can be concatenated and input into the fully connected layer and the activation function layer. The activation function layer uses the sigmoid activation function. The activation function layer then outputs the model's prediction results.

[0101] In this embodiment of the application, the DeepFM model is used as an example for blog post quality assessment. The output of the FM layer and the output of the Deep layer are concatenated and input into the fully connected layer and the activation function layer. This can alleviate the problem of the magnitude explosion caused by the addition method, which results in most of the output value (i.e., logis) falling into the sigmoid saturation region.

[0102] In some embodiments, the training method of the blog post quality assessment model may include: improving the loss function of the blog post quality assessment model by weighting the loss of positive samples.

[0103] In one example, the expression (4) for the loss function of the improved blog post quality assessment model is as follows:

[0104]

[0105] In the formula, p is the probability value that the blog post predicted by the blog post quality assessment model is a positive sample, and p can also represent the first quality score of the sample blog post. ω is the preset positive sample weighting coefficient, a is the preset positive and negative sample importance factor, b is the preset hard sample weighting factor, and logis is the output value of the activation function layer. The value range of a is [0,1], with a preferred value of 0.9615, and the preferred value of b is 2.

[0106] This application provides a blog post quality assessment method. After obtaining a trained blog post quality assessment model, the method may include: obtaining multiple historical blog posts that have been exposed and the bloggers who published them, based on the exposure logs of a blog post recommendation system; inputting the blog post information of each historical blog post and the user information of its corresponding blogger into the trained blog post quality assessment model to obtain a second quality score for each historical blog post output by the model; training the blog post quality assessment model using the method described above, with the second quality score representing the user's level of interest in the historical blog posts; determining target blog posts among the multiple exposed historical blog posts based on the second quality score; and using the target blog posts for coarse or fine ranking.

[0107] Understandably, after obtaining the trained blog post quality assessment model, it can be deployed online. This involves inputting the blog post information and the corresponding user information from those historical blog posts into the model to obtain a second quality score. Based on this second quality score, multiple historical blog posts are ranked. Then, those corresponding to second quality scores that meet a preset threshold are selected as target blog posts. These target posts can then proceed to the next stage of the blog post recommendation process: either coarse ranking or fine ranking.

[0108] In some possible implementations, the above-mentioned method of determining the target blog post from multiple exposed historical blog posts based on the second quality score includes: calculating a third quality score corresponding to each historical blog post according to a preset decay mapping relationship, a preset time period, and the second quality score; the third quality score is used to characterize the user's interest in the historical blog post after removing the influence of time period; the decay mapping relationship is the mapping relationship between the second quality score and the third quality score; associating and saving the blog post information of each historical blog post and the third quality score corresponding to each historical blog post to the trained blog post quality evaluation model; and determining the target blog post from multiple exposed historical blog posts based on the third quality score.

[0109] Understandably, when using it online, considering the time-sensitivity of blog posts, the quality scores (i.e., the second quality scores) of historical blog posts output by the trained blog post quality assessment model can be time-decayed to remove the impact of the time-sensitivity of blog posts on recommendations.

[0110] In one example, taking exposed historical blog posts as including posts D, E, and F, with the blogger who published post D as blogger D, post E as blogger E, and post F as blogger F, the following steps are taken: The blog post information for posts D, E, and F, and the user information for bloggers D, E, and F are obtained from the exposure logs of the blog post recommendation system. The blog post information for each historical post and the user information of the corresponding blogger are input into the trained blog post quality assessment model for quality evaluation, resulting in a second quality score for each of posts D, E, and F. Through a pre-defined decay mapping process, the second quality score of each historical post is subject to time-related decay processing, yielding a third quality score for each of posts D, E, and F. Then, the blog post information and third quality score of each historical blog post are associated and stored in the database of the trained blog post quality assessment model; next, based on the third quality score of each historical blog post, multiple historical blog posts are sorted to select the target blog post.

[0111] In some embodiments, the expression (5) for the above attenuation mapping relationship is as follows:

[0112]

[0113] In the formula, c is the decay rate, the time decay interval is [f, d], f is the publication time of the historical blog post, d is the time when the popularity of the historical blog post disappears, m is the preset blog post validity period, l is the decay value, g is the publication time of the historical blog post, p is the quality score (i.e., the second quality score) of the historical blog post output by the trained blog post quality assessment model, and S d The third mass fraction is defined as follows. The preferred time decay interval is [0.1, 1].

[0114] In this embodiment of the application, the third quality score obtained after time decay processing can remove the impact of the timeliness of the blog post.

[0115] In some embodiments, each time a new historical blog post is exposed, it is input into the trained blog post quality assessment model for quality evaluation, i.e., calculating the third quality score of the historical blog post. Each historical blog post may be exposed multiple times, and the third quality score of the historical blog post can be calculated every time it is exposed. The latest obtained third quality score replaces the previously stored third quality score in the blog post quality assessment model. In other words, the third quality score of the historical blog post is updated every time it is exposed.

[0116] In some embodiments, the latest third quality score of historical blog posts is obtained through real-time updates. Then, based on the third quality score, personalized strategies can be developed to support historical blog posts in vertical fields (such as beauty, sports and fitness) according to the level of the historical blogger and the type of historical blog posts, so as to improve the exposure and user reach of historical blog posts in the relationship stream.

[0117] In some possible implementations, the above-mentioned determination of the target blog post from multiple exposed historical blog posts based on the third quality score may include: filtering multiple recalled blog posts from multiple exposed historical blog posts through multi-link recall; obtaining the third quality score corresponding to each recalled blog post in the trained blog post quality assessment model; and selecting the target blog post from the multiple recalled blog posts based on the third quality score corresponding to each recalled blog post.

[0118] Understandably, during the online application of a trained blog post quality assessment model, a multi-link recall approach can be used to initially select N blog posts that meet preset requirements from M historical blog posts in the blog post recommendation system, where N < M. Then, the third quality score corresponding to each recalled blog post is obtained from the database of the trained blog post quality assessment model. Finally, based on the N third quality scores, the N recalled blog posts are ranked, and the top n recalled blog posts with third quality scores meeting preset thresholds are selected as target blog posts. These n target blog posts will then enter the coarse-ranking or fine-ranking stage of the blog post recommendation process for further selection.

[0119] Figure 4 This is a schematic diagram of an optional process for training a blog post quality assessment model and evaluating blog post quality, as provided in the embodiments of this application. The following is a detailed explanation of the process. Figure 4 The steps shown illustrate the training method for the blog post quality assessment model.

[0120] The following explanation uses the DeepFM model as an example to illustrate the blog post quality assessment model.

[0121] S401, the training device obtains the blog post information of multiple sample blog posts and the corresponding sample blogger's user information from the exposure log of the blog post recommendation system, and stores the blog post information and user information in the database of the blog post recommendation system.

[0122] S402, the training device obtains the blog post information of the sample blog posts and the user information of the sample bloggers from the database, and performs data preprocessing on the above blog post information and user information to obtain blog post-side behavioral features, user-side behavioral features, blog post short-term features and user information features; then, it performs correction processing on the blog post-side behavioral features and user-side behavioral features to obtain the blog post correction features and user correction features corresponding to each sample blog post.

[0123] S403, the training device performs linear fusion processing on blog post-side behavioral features, user-side behavioral features, blog post correction features, and user correction features to obtain the initial quality score corresponding to each sample blog post; and labels multiple sample blog posts as positive and negative samples based on the initial quality scores.

[0124] S404, the training device iteratively trains the blog post quality assessment model based on the short-term features of the blog post, user information features, and the labeling of positive or negative samples corresponding to each sample blog post, so as to obtain the trained blog post quality assessment model; then, the blog post quality assessment device uses the trained blog post quality assessment model to perform quality assessment on each exposed historical blog post, so as to obtain the third quality score corresponding to each historical blog post.

[0125] S405, the blog post quality assessment device retrieves recalled blog posts from multiple historical blog posts through a multi-link recall method, and obtains the third quality score of the recalled blog posts from the database of the trained blog post quality assessment model.

[0126] S406, the blog post quality assessment device sorts the recalled blog posts based on the third quality score of the recalled blog posts to obtain the target blog posts, and sends the target blog posts to the fine sorting stage.

[0127] At this point, the recommendation system has completed the training method for the blog post quality assessment model and the blog post quality assessment method described above.

[0128] In this embodiment, by modifying the blog post-side behavioral features and user-side behavioral features, modified blog post features and modified user features are obtained for each training sample. The confidence level of the modified blog post features is higher than that of the blog post-side behavioral features, and the confidence level of the modified user features is higher than that of the user-side behavioral features. This can solve the problem of inconsistent dimensions between sample blog posts, achieve consistency of training samples, provide a more accurate data processing foundation for the blog post recommendation process, and thus achieve more accurate blog post recommendations.

[0129] Furthermore, based on the preset attenuation mapping relationship, preset time limit, and second quality score, the third quality score corresponding to each historical blog post is calculated. This allows for the correction of blog posts with strong time sensitivity, providing a more intuitive representation of the quality of historical blog posts.

[0130] The following continues to describe the exemplary structure of the training device 455 for the blog post quality assessment model provided in this application embodiment as a software module. In some embodiments, such as Figure 2As shown, the software modules in the training device 455 of the blog post quality assessment model stored in the memory 450 may include: a sample data acquisition module 4551, used to acquire multiple training samples based on the exposure logs of the blog post recommendation system; wherein each training sample includes the blog post information of the sample blog post and the user information of the corresponding sample blogger, the sample blogger being the user who published the sample blog post; a sample data preprocessing module 4552, used to preprocess the blog post information of the sample blog post and the user information of the corresponding sample blogger in each training sample, generating blog post-side behavioral features and user-side behavioral features corresponding to each training sample; wherein the blog post-side behavioral features are used to characterize the interaction of the sample blog post within a preset first time period; the user-side behavioral features are used to characterize the interaction of the sample blogger within a preset first time period. The interaction data of blog posts published during the second time period; the sample data processing module 4553 is used to modify the blog post-side behavioral features and user-side behavioral features to obtain the blog post modified features and user modified features corresponding to each training sample; the confidence level of the blog post modified features is higher than that of the blog post-side behavioral features, and the confidence level of the user modified features is higher than that of the user-side behavioral features; the training module 4554 is used to adjust the model parameters based on the blog post-side behavioral features, user-side behavioral features, blog post modified features, and user modified features corresponding to each training sample during the iterative training of the initial blog post quality assessment model using multiple training samples, until the blog post quality assessment model converges, and the trained blog post quality assessment model is obtained; the blog post quality assessment model is used to assess the quality of blog posts.

[0131] In some possible implementations, the sample data preprocessing module 4552 is further configured to, after acquiring multiple training samples, preprocess the blog post information of the sample blog posts in each training sample and the user information of the corresponding sample blogger, and generate short-term features of the blog posts and user information features corresponding to each training sample; the short-term features of the blog posts are used to characterize the interaction of the sample blog posts in a preset third time period, the third time period being shorter than the first time period, and the user information features are used to characterize the user profile of the sample blogger.

[0132] In some possible implementations, the training module 4554 is used to determine the labeling information corresponding to each sample blog post based on the blog post-side behavior features, user-side behavior features, blog post correction features, and user correction features corresponding to each training sample; the labeling information is used to label the sample blog post as a positive or negative sample; the short-term features of the blog post and user information features corresponding to each training sample are input into the blog post quality assessment model to obtain the quality assessment result output by the blog post quality assessment model; the quality assessment result includes the first quality score corresponding to the sample blog post and the prediction result of whether the sample blog post is a positive or negative sample; the first quality score is used to characterize the degree of user interest in the sample blog post; a comparison result is obtained by comparing the prediction result and labeling information corresponding to each sample blog post; based on the comparison result of each sample blog post, the parameters of the blog post quality assessment model are tuned until the blog post quality assessment model converges.

[0133] In some possible implementations, the training module 4554 is used to perform linear fusion processing on the blog post-side behavior features, user-side behavior features, blog post correction features, and user correction features corresponding to each training sample to obtain the initial quality score corresponding to the sample blog post in each training sample; the initial quality score is used to characterize the degree of user interest in the sample blog post without processing by the blog post quality assessment model; if the initial quality score of the sample blog post meets a preset threshold, the sample blog post is marked as a positive sample; if the initial quality score of the sample blog post does not meet the preset threshold, the sample blog post is marked as a negative sample.

[0134] In some possible implementations, the sample data processing module 4553 is used to perform confidence correction on the blog post-side behavioral features based on the Wilson confidence interval to obtain the blog post correction features corresponding to each training sample; and to perform confidence correction on the user-side behavioral features based on the Wilson confidence interval to obtain the user correction features corresponding to each training sample.

[0135] This application provides a blog post quality assessment device 456, which includes: a data acquisition module 4561, used to acquire multiple historical blog posts that have been exposed and the historical bloggers who published them, based on the exposure logs of a blog post recommendation system; a data processing module 4562, used to input the blog post information of each historical blog post and the user information of its corresponding historical blogger into a trained blog post quality assessment model to obtain a second quality score for each historical blog post output by the blog post quality assessment model; the blog post quality assessment model is trained using the training method described above, and the second quality score is used to characterize the degree of user interest in the historical blog posts; and a filtering module 4563, used to determine target blog posts among the multiple historical blog posts that have been exposed based on the second quality score; the target blog posts are used for coarse ranking or fine ranking of blog posts.

[0136] In some possible implementations, the filtering module 4563 is used to calculate a third quality score corresponding to each historical blog post based on a preset decay mapping relationship, a preset time period, and a second quality score; the third quality score is used to characterize the user's interest in the historical blog post after removing the influence of time period; the decay mapping relationship is the mapping relationship between the second quality score and the third quality score; the blog post information of each historical blog post and the third quality score corresponding to each historical blog post are associated and saved to the trained blog post quality evaluation model; and the target blog post is determined from multiple exposed historical blog posts based on the third quality score.

[0137] In some possible implementations, the screening module 4563 is used to screen out multiple recalled blog posts from multiple exposed historical blog posts through multi-link recall; obtain the third quality score corresponding to each recalled blog post in the trained blog post quality assessment model; and select the target blog post from the multiple recalled blog posts based on the third quality score corresponding to each recalled blog post.

[0138] This application provides a computer program product or computer program that includes computer instructions stored in a computer storage medium. A processor of a computer device reads the computer instructions from the computer storage medium and executes the computer instructions, causing the computer device to perform the training method or blog post quality assessment method of the blog post quality assessment model described in this application.

[0139] This application provides a computer storage medium storing executable instructions. When these executable instructions are executed by a processor, they cause the processor to execute the training method or blog quality assessment method of the blog post quality assessment model provided in this application. For example... Figure 3 The training method and the blog post quality assessment method of the blog post quality assessment model are shown.

[0140] In some embodiments, the computer storage medium may be a memory such as FRAM, ROM, PROM, EEPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0141] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0142] As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hypertext Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborating files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0143] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0144] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. A training method for a blog post quality assessment model, characterized in that, The method includes: Based on the exposure logs of the blog post recommendation system, multiple training samples are obtained; each training sample includes the blog post information of the sample blog post and the user information of the corresponding sample blogger, wherein the sample blogger is the user who published the sample blog post; The blog post information and the corresponding user information of the sample blogger in each training sample are preprocessed to generate blog post-side behavioral features and user-side behavioral features for each training sample. The blog post-side behavioral features are used to characterize the interaction of the sample blog post within a preset first time period. The user-side behavioral features are used to characterize the interaction of the blog posts published by the sample blogger within a preset second time period. By modifying the blog post-side behavior features and the user-side behavior features, blog post modified features and user modified features corresponding to each training sample are obtained; the confidence level of the blog post modified features is higher than that of the blog post-side behavior features, and the confidence level of the user modified features is higher than that of the user-side behavior features. During the iterative training of the initial blog post quality assessment model using the multiple training samples, the model parameters are adjusted based on the blog post-side behavioral features, user-side behavioral features, blog post correction features, and user correction features corresponding to each training sample, until the blog post quality assessment model converges, resulting in a trained blog post quality assessment model; the blog post quality assessment model is used to assess the quality of blog posts.

2. The method according to claim 1, characterized in that, The method further includes: After acquiring multiple training samples, the blog post information and the corresponding user information of the sample blogger in each training sample are preprocessed to generate short-term features of the blog post and user information features corresponding to each training sample. The short-term features of the blog post are used to characterize the interaction of the sample blog post in a preset third time period, which is shorter than the first time period. The user information features are used to characterize the user profile of the sample blogger.

3. The method according to claim 2, characterized in that, In the process of iteratively training the initial blog post quality assessment model using the multiple training samples, the model parameters are adjusted based on the blog post-side behavioral features, user-side behavioral features, blog post correction features, and user correction features corresponding to each training sample, until the blog post quality assessment model converges, resulting in a trained blog post quality assessment model, including: Based on the blog post-side behavior features, user-side behavior features, blog post correction features, and user correction features corresponding to each training sample, the labeling information corresponding to each sample blog post is determined; the labeling information is used to mark the sample blog post as a positive sample or a negative sample. The short-term features and user information features of each training sample are input into the blog post quality assessment model to obtain the quality assessment result output by the model. The quality assessment result includes a first quality score corresponding to the sample blog post and a prediction result of whether the sample blog post is a positive or negative sample. The first quality score is used to characterize the degree of user interest in the sample blog post. The comparison results are obtained by comparing the prediction results and tagging information corresponding to each sample blog post; Based on the comparison results of each sample blog post, the parameters of the blog post quality assessment model are adjusted until the blog post quality assessment model converges.

4. The method according to claim 3, characterized in that, The process involves determining the tagging information for each sample blog post based on the blog post-side behavior features, user-side behavior features, blog post correction features, and user correction features corresponding to each training sample. This includes: For each training sample, the blog post-side behavioral features, user-side behavioral features, blog post correction features, and user correction features are linearly fused to obtain the initial quality score for each sample blog post. The initial quality score is used to characterize the degree of user interest in the sample blog post before it has been processed by the blog post quality assessment model. If the initial quality score of the sample blog post meets the preset threshold, the sample blog post is marked as a positive sample. If the initial quality score of the sample blog post does not meet the preset threshold, the sample blog post will be marked as a negative sample.

5. The method according to claim 1, characterized in that, The process involves modifying the blog post-side behavioral features and the user-side behavioral features to obtain modified blog post features and modified user features for each training sample, including: Based on the Wilson confidence interval, the blog post behavior features are reliably corrected to obtain the blog post correction features corresponding to each training sample. Based on the Wilson confidence interval, the credibility of the user-side behavioral features is corrected to obtain the user-corrected features corresponding to each training sample.

6. A method for evaluating the quality of blog posts, characterized in that, The method includes: Based on the exposure logs of the blog post recommendation system, obtain multiple historical blog posts that have been exposed and the historical bloggers who published the aforementioned historical blog posts; The blog post information of each historical blog post and the user information of the corresponding historical blogger are input into the trained blog post quality assessment model to obtain the second quality score corresponding to each historical blog post output by the blog post quality assessment model; the blog post quality assessment model is trained using the method described in any one of claims 1 to 5, and the second quality score is used to characterize the user's degree of interest in the historical blog post. Based on the second quality score, a target blog post is determined from the plurality of exposed historical blog posts; the target blog post is used for rough or fine blog post ranking.

7. The method according to claim 6, characterized in that, The step of determining the target blog post from the plurality of exposed historical blog posts based on the second quality score includes: Based on the preset decay mapping relationship, the preset time period, and the second quality score, a third quality score is calculated for each historical blog post; the third quality score is used to characterize the user's interest in the historical blog post after removing the influence of time period; the decay mapping relationship is the mapping relationship between the second quality score and the third quality score. The blog post information of each historical blog post and the corresponding third quality score of each historical blog post are associated and saved to the trained blog post quality assessment model. The target blog post is determined from the plurality of previously exposed historical blog posts based on the third quality score.

8. The method according to claim 7, characterized in that, The step of determining the target blog post from the plurality of previously exposed historical blog posts based on the third quality score includes: Through multi-link recall, multiple recall blog posts were selected from the multiple exposed historical blog posts. In the trained blog post quality assessment model, obtain the third quality score corresponding to each recalled blog post; The target blog post is selected from the plurality of recalled blog posts based on the third quality score corresponding to each recalled blog post.

9. A blog post quality assessment device, characterized in that, The device includes: The data acquisition module is used to obtain multiple historical blog posts that have been exposed and the bloggers who published those historical blog posts based on the exposure logs of the blog post recommendation system. The data processing module is used to input the blog post information of each historical blog post and the user information of the corresponding historical blogger into the trained blog post quality assessment model to obtain the second quality score corresponding to each historical blog post output by the blog post quality assessment model; the blog post quality assessment model is trained using the method described in any one of claims 1 to 5, and the second quality score is used to characterize the degree of user interest in the historical blog post. The filtering module is used to determine the target blog post from the plurality of exposed historical blog posts based on the second quality score; the target blog post is used for rough ranking or fine ranking of blog posts.

10. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable instructions; A processor, when executing executable instructions or computer programs stored in the memory, implements the method of any one of claims 1 to 5 or the method of any one of claims 6 to 8.

Images