An intelligent work presentation system based on data tags
By using a data-label-based intelligent artwork display system, which employs convolutional neural networks and recurrent neural network models for artwork classification and personalized recommendations, the system addresses the poor user experience issues in existing systems and achieves fast and accurate artwork recommendations and resource optimization.
Patent Information
- Application Number
- CN202510287894.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-12
- Publication Date
- 2026-01-02
- Estimated Expiration
- 2045-03-12
AI Technical Summary
The existing artwork display system cannot make accurate recommendations based on users' personalized needs, resulting in users having to spend a lot of time and effort to find works they are interested in, leading to a poor user experience.
The system employs a data-label-based intelligent artwork display system. It uses a pre-trained composite model of convolutional neural networks and recurrent neural networks to classify artworks, and combines user basic information and historical operation information to calculate interest and recommendation scores, providing users with personalized artwork displays.
It enables fast and accurate work categorization and personalized recommendations, improves user experience and platform user stickiness, optimizes resource utilization efficiency, and reduces operating costs.
Smart Images
Figure CN120179931B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of work display, and particularly relates to an intelligent work display system based on data tags. BACKGROUND
[0002] In today's era of digital information explosion, the number of various works is growing exponentially. How to effectively manage and display these works has become a problem to be solved.
[0003] The traditional work display method mainly relies on manual classification and arrangement, which has many drawbacks. On the one hand, the efficiency of manual classification is low, which requires a lot of time and labor cost. For massive work data, manual classification and annotation one by one not only has huge workload, but also is easy to make mistakes and omissions. On the other hand, manual classification is subjective, and different people may have different understanding and classification standards for works, resulting in inaccurate and inconsistent classification results.
[0004] Although some existing work display systems introduce certain information management means, they only simply store and display work information electronically, and lack in-depth analysis and mining of work information. Therefore, they cannot accurately recommend and display works according to the personalized needs of users, resulting in poor user experience when users search for works of interest. SUMMARY
[0005] The purpose of the present application is to provide an intelligent work display system based on data tags, which solves the technical problem that accurate work recommendation and display cannot be made according to the personalized needs of users, resulting in poor user experience when users search for works of interest.
[0006] The purpose of the present application can be achieved by the following technical solutions:
[0007] An intelligent work display system based on data tags, comprising:
[0008] A work data collection module for collecting relevant data of each work, the relevant data including basic information of the work, detailed description of the work, and relevant pictures of the work; the basic information of the work including work name, author and creation time;
[0009] A label generation module for pre-processing the data collected by the data collection module and then inputting the pre-processed data into a work classification model to output category labels of the work;
[0010] A data storage module for storing the work data and corresponding category labels in a database;
[0011] The intelligent display module analyzes the user's work category preference according to the basic information and the historical operation information of the user, matches the work category preference of the user with corresponding work category labels, and provides personalized work display for the user.
[0012] In a further aspect, the work classification model is a composite model of a pre-trained convolutional neural network model and a recurrent neural network model.
[0013] In a further aspect, the process of providing personalized work display for the user includes:
[0014] Obtaining the basic information of the current user, including the preferred work category and the author information followed by the user;
[0015] Based on the historical operation information of the current user, the interest degree RK of the current user for each work category is analyzed; K represents the Kth work category.
[0016] After arranging in descending order of the interest degree RK, the work categories in the first N positions are selected to form a preliminary work recommendation sequence.
[0017] Based on the interest degree RK of the current user for each work category, the heat coefficient H of each work, and the difference ΔT between the creation time of each work and the current time, the recommendation score PS of each work is comprehensively analyzed; S represents the Sth work.
[0018] The preliminary work sequence is reordered according to the recommendation score PS, and the works in the first M positions are selected to form a work recommendation sequence.
[0019] In a further aspect, the historical operation information of the current user includes: the number of times NL of browsing, the number of times Nd of liking, the number of times Np of commenting, and the number of times NS of collecting the Kth work category; the total number of historical browsing times NL of the current user z , the number of times Nd of liking z , the number of times Np of commenting z , and the number of times NS of collecting z .
[0020] The expression of the interest degree RK of each work category is:
[0021]
[0022] In the formula, n represents the number of preset time periods, ξ represents the ξth preset time period, θ ξ represents the weight coefficient corresponding to the ξth preset time period, α, β, γ, δ represent weight factors, and α+β+γ+δ=1, rK ξ represents the interest index of each work category in the ξth preset time period.
[0023] In a further aspect, the weight coefficient θ corresponding to the ξth preset time period ξ By the following process:
[0024] The interest index rK of each work category of the ξth preset time period and the last preset time period is taken as input into the following formula: ξ , rK ξ-1 The following formula is taken as input:
[0025] ΔrK = rK ξ - rK ξ-1 The difference ΔrK of the interest index of each work category is calculated.
[0026] If ΔrK > 0, when ΔrK ≥ ΔrK o+ , the following is performed: Otherwise,
[0027] If ΔrK < 0, when ΔrK ≤ ΔrK o- , the following is performed: Otherwise,
[0028] In the formula, ω is an adjustment coefficient, is an initial weight coefficient.
[0029] In a further aspect, the recommended score PS of each work is calculated by the following formula: In the formula, ρ1, ρ2, and ρ3 are reference coefficients, and ρ1 + ρ2 + ρ3 = 1.
[0030] In a further aspect, the expression of the work heat coefficient H is as follows:
[0031] In the formula, m represents the number of work heat related parameters, ξ j (t) represents the time-varying curve of each work heat related parameter, t0 and t1 represent the start time and end time of the monitoring time period, respectively, and ν j represents a preset proportion coefficient of each work heat related parameter; j represents the jth item; the work heat related parameters include the number of views, likes, comments, and collections of the current work.
[0032] In a further aspect, the intelligent display module further includes a user feedback collection unit configured to collect feedback information of the user on the displayed works, and the feedback information includes a recommendation satisfaction degree.
[0033] When the recommendation satisfaction degree is lower than a satisfaction threshold interval, the number of work categories in the preliminary work recommendation sequence is increased; when the recommendation satisfaction degree belongs to the satisfaction threshold interval, the number of work categories in the preliminary work recommendation sequence is maintained; and when the recommendation satisfaction degree is higher than the satisfaction threshold interval, the number of work categories in the preliminary work recommendation sequence is reduced.
[0034] Advantages of the present application:
[0035] (1) The present application collects data through the work data collection module, and classifies by using the composite model of the label generation module, replacing manual classification; compared with the traditional way, the classification time is greatly shortened, and the error rate is reduced;
[0036] (2) In the present application, the intelligent display module analyzes the user's basic information and historical operation information, combines the work popularity and creation time, and provides personalized work display for the user, improves the user experience, enhances the user's good impression and trust for the platform, and improves the user stickiness and activity. BRIEF DESCRIPTION OF DRAWINGS
[0037] The present application will be further described below in conjunction with the accompanying drawings.
[0038] Figure 1 The system logic diagram of the present application. DETAILED DESCRIPTION
[0039] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0040] Please refer to Figure 1 The present application is an intelligent work display system based on data label, which comprises:
[0041] A work data collection module is used to collect relevant data of each work, and the relevant data includes basic information of the work, detailed description of the work, and relevant pictures of the work; the basic information of the work includes the work name, the author and the creation time;
[0042] A label generation module is used to pre-process the data collected by the data collection module, and then input into the work classification model to output the category label of the work; the image and text data are pre-processed to meet the input requirements of the convolutional neural network CNN model and the recurrent neural network RNN model; for example, scaling and normalization are performed on the image, and word segmentation, word embedding and other operations are performed on the text;
[0043] A data storage module is used to store the work data and the corresponding category label in the database;
[0044] The intelligent display module analyzes the user's work category preference according to the user's basic information and historical operation information, matches the user's work category preference with corresponding work category labels, and provides personalized work display for the user.
[0045] In the application, in order to solve the problem that the existing work display adopts manual work classification before work display, the program is complex and subjective factors lead to high classification error rate, therefore, through work information collection and preprocessing, the category label of each work is output by the label generation module, the category label exists in a relational database such as MySQL, Oracle or a non-relational database such as MongoDB, Redis for retrieval and extraction, which can obviously make the work classification faster and more accurate, thereby creating conditions for subsequent different users to provide more preferred work display, and achieving the dual effects of improving user satisfaction and work display efficiency.
[0046] The work classification model is a composite model of a pre-trained convolutional neural network model and a recurrent neural network model. In the application, the label generation module uses a composite model of a pre-trained convolutional neural network (CNN) model and a recurrent neural network (RNN) model, which fully combines the advantages of CNN in image feature extraction and RNN in text sequence processing, can more accurately process multi-modal data, and generate reasonable category labels. It should be noted that the pre-trained convolutional neural network model is one of ResNet, VGG or Inception series models, and the recurrent neural network model is one of LSTM and GRU models. The composite model adopts early fusion, late fusion or attention mechanism-based fusion to fuse the output features of the convolutional neural network model and the recurrent neural network model.
[0047] The process of providing personalized work display for the user includes:
[0048] Obtain the basic information of the current user, including the preferred work category and the author information concerned;
[0049] Based on the historical operation information of the current user, the interest degree RK of the current user to each work category is analyzed; K represents the Kth work category.
[0050] After arranging in descending order of interest degree RK, the work categories in the first N positions are selected to form a preliminary work recommendation sequence;
[0051] Based on the interest degree RK of the current user to each work category, the heat coefficient H of each work, and the difference ΔT between the creation time of each work and the current time, the recommendation score PS of each work is comprehensively analyzed; S represents the Sth work.
[0052] The works in the top M positions are selected to form a work recommendation sequence according to the recommended score PS.
[0053] The present application can preliminarily understand the preference direction of the user by obtaining the basic information of the user such as the preferred work category and the author information concerned; the interest degree RK obtained by analyzing the historical operation information of the user can more accurately grasp the interest degree of the user to different work categories; for example, the system will give a higher interest degree to a category of works which the user often browses, and the works in this category will be preferentially considered in subsequent recommendation, which makes the recommended works more in line with the actual interests of the user, greatly improves the probability of the user finding interesting works, and improves the user experience; at the same time, the work recommendation sequence is formed by arranging the work categories in descending order according to the interest degree RK and selecting the top N work categories, which narrows the recommendation range and focuses on the categories more interesting to the user; then, the recommended score PS is calculated by comprehensively considering the interest degree RK of the user, the work heat coefficient H and the difference between the work creation time and the current time, which further optimizes the recommendation result; the heat coefficient H can reflect the popularity of the work, and newly created works can also obtain appropriate weight in the recommendation through the time difference factor, so that the recommended works not only meet the interests of the user, but also have certain heat and timeliness, improving the quality and efficiency of the recommendation;
[0054] Moreover, when the user can quickly and accurately find works that meet his or her interests, the user will increase the like and trust for the platform, and thus be more willing to use the platform frequently, improving the stickiness and activity of the user; for example, if the user can always find new works of interest on the platform, the user will be more inclined to use the platform for a long time to obtain work information and perform related operations, which has a positive significance for the long-term stable development of the platform; similarly, the personalized recommendation method avoids blind recommendation and reduces the situation that the user browses a large number of uninteresting works, so that the resources of the platform such as server resources and storage resources can be more effectively utilized, and the platform can focus on recommending works that the user may be interested in, improving the utilization efficiency of resources and reducing operating costs.
[0055] The historical operation information of the current user includes: the number of times NL of browsing, the number of times Nd of liking, the number of times Np of commenting and the number of times NS of collecting the Kth work category; the total number of times NL of browsing of the current user z , the number of times Nd of liking z , the number of times Np of commenting z and the number of times NS of collecting z ;
[0056] The expression of the interest degree RK of each work category is:
[0057]
[0058] In the formula, n represents the number of preset time periods, ξ represents the ξth preset time period, θ ξ represents the weight coefficient corresponding to the ξth preset time period, α, β, γ, and δ represent weight factors, are determined based on historical data and empirical data, and α+β+γ+δ=1, rK ξ represents the interest index of each work category in the ξth preset time period.
[0059] The present application provides a method for obtaining the interest degree RK of each work category, which comprises the following steps: calculating the interest degree RK of each work category; the above method comprehensively considers the browsing, liking, commenting, and collecting behaviors of a user on different work categories in different preset time periods, dynamically adjusts the calculation of the interest degree of the user on each work category through the weight factors and the preset time period weight coefficients, more accurately reflects the change of the user interest with time, and makes the calculation result of the interest degree more accurate, thereby providing more accurate data support for personalized recommendation.
[0060] The weight coefficient θ ξ corresponding to the ξth preset time period is obtained through the following process:
[0061] The interest index rK ξ of each work category in the ξth preset time period and the interest index rK ξ-1 of each work category in the previous preset time period are brought into the following formula:
[0062] ΔrK=rK ξ -rK ξ-1 The difference ΔrK of the interest index of each work category is calculated;
[0063] If ΔrK>0, when ΔrK≥ΔrK o+ , θ , otherwise,
[0064] If ΔrK<0, when ΔrK≤ΔrK o- , θ , otherwise,
[0065] In the formula, ω is an adjustment coefficient, is determined based on historical data and empirical data analysis, is an initial weight coefficient, ΔrK o+ , and ΔrK o- is a preset warning value.
[0066] The present application provides a method for obtaining the weight coefficient θ ξ corresponding to each preset time period, specifically, according to ΔrK=rK ξ -rK ξ-1The difference value of the interest index of each work category is calculated, and the difference value is represented as ΔrK, which represents the change of the work category interest index of adjacent preset time periods, and the weight coefficient is dynamically adjusted, so that the fluctuation of the user interest can be captured in time; when the user interest rises or falls greatly, the weight is adjusted accordingly, so that the calculation of the user interest degree can better reflect the latest interest tendency of the user, and the personalized recommendation effect is further optimized.
[0067] The recommendation score PS of each work is calculated by the formula: The recommendation score PS of each work is calculated by the formula:
[0068] In the application, RK represents the interest degree of the user to the Kth work category, which reflects the personalized preference of the user, and the higher the RK value is, the more intense the user's interest in the category is; H is a work popularity coefficient, which is used to measure the degree of popularity of the work among the public, and the larger the H value is, the higher the popularity of the work is; ΔT represents the difference value of the work creation time from the current time, and the larger the ΔT is, the longer the work is from the current time; the formula takes these three factors into consideration, and comprehensively weighs the personalized demand of the user, the popularity of the work and the timeliness; the multiplication of the work popularity coefficient H and the reciprocal of the creation time difference value ΔT means that the higher the work popularity is and the newer the creation time is, the greater the score added in the recommendation score is; plus the interest degree RK of the user to the work category, the final recommendation score PS can more comprehensively reflect the recommendation value of the work to the user; so that the works can be sorted according to the PS value, and the works with high PS value are preferentially recommended, so that the recommended works not only meet the user's interest, but also have a certain degree of popularity and freshness, the recommendation result is optimized, and the attraction and value of the recommended works to the user are improved.
[0069] The expression of the work popularity coefficient H is:
[0070] In the formula, m represents the number of work popularity related parameters, ξ j (t) represents the time-varying curve of each work popularity related parameter, t0 and t1 represent the starting time and the ending time of the monitoring period respectively, and v j represents a preset proportional coefficient of each work popularity related parameter, which is determined based on historical data; j represents the jth; the work popularity related parameters include but are not limited to the number of browsing, liking, commenting and collecting of the current work.
[0071] In the application, by comprehensively considering the browsing, liking, commenting and collecting of the work, which are closely related to the work popularity, the degree of attention and popularity of the work on the platform can be objectively and comprehensively measured; compared with simply judging the work popularity according to a single index, the comprehensive calculation method is more accurate, and the deviation of the work popularity evaluation caused by the limitation of the single index is avoided; by The cumulative change amount of each parameter in the monitoring period is summed and averaged, so that the work heat coefficient H can reflect the dynamic change of the work heat with time, so as to capture the fluctuation of the work heat, and timely evaluate and adjust the work heat rising or falling, and provide more time-effective data support for work recommendation.
[0072] The intelligent display module further comprises a user feedback collection unit for collecting feedback information of the user on the displayed works, and the feedback information comprises recommendation satisfaction, and the recommendation satisfaction is reflected by user scoring, and the score is set to 1-10 points, and the higher the score is, the higher the recommendation satisfaction is.
[0073] When the recommendation satisfaction is lower than the satisfaction threshold interval, the number of works of the primary work recommendation sequence is increased; when the recommendation satisfaction belongs to the satisfaction threshold interval, the number of works of the primary work recommendation sequence is maintained; and when the recommendation satisfaction is higher than the satisfaction threshold interval, the number of works of the primary work recommendation sequence is reduced.
[0074] It should be noted that the calculation formula and the parameters participating in the operation in the present application are all pre-processed by dimensionless processing, and the process of dimensionless processing is well known in the industry, which is not described here.
[0075] The above describes one embodiment of the present application in detail, but the content described is only a preferred embodiment of the present application, and cannot be considered as limiting the scope of the present application. Any equivalent changes and improvements made according to the scope of the present application should still belong to the scope of the present application.
Claims
1. A data tag based smart portfolio presentation system, characterized in that, The application relates to an intelligent work display method and device. The works data collection module is used for collecting relevant data of various works, wherein the relevant data comprises basic information of the works, detailed description of the works and relevant pictures of the works; the basic information of the works comprises a work name, an author and a creation time; The label generation module is used for pre-processing the data collected by the data collection module and then inputting the data into a work classification model to output a work category label; The data storage module is used for storing the work data and the corresponding category label in a database; The intelligent display module is used for analyzing a work category preference of a user according to basic information and historical operation information of the user, matching the work category preference of the user with the corresponding work category label and providing personalized work display for the user. The process of providing personalized work display for the user comprises the following steps: Basic information of a current user is acquired, including a preferred work category and author information concerned; Based on historical operation information of the current user, the interest degree RK of the current user to each work category is analyzed; K represents the Kth work category. After being arranged in descending order according to the interest degree RK, the work categories in the first N positions are selected to form a preliminary work recommendation sequence; Based on the interest degree RK of the current user to each work category, a heat coefficient H of each work and a difference value AT between the creation time of each work and the current time, the recommendation score PS of each work is comprehensively analyzed; S represents the Sth work. The preliminary work sequence is reordered according to the recommendation score PS, and the work in the first M positions is selected to form a work recommendation sequence. The historical operation information of the current user includes: the number of times of browsing NL, the number of times of liking Nd, the number of times of commenting Np and the number of times of collecting NS of the Kth work category; the total number of times of browsing NL of the current user z , the number of times of liking Nd z , the number of times of commenting Np z and the number of times of collecting NS z The expression of the interest degree RK of each work category is as follows: In the formula, n represents the number of preset time periods, ξ represents the ξth preset time period, θ ξ represents the weight coefficient corresponding to the ξth preset time period, α, β, γ, and δ represent weight factors, and α+β+γ+δ=1, rK ξ represents the interest index of each work category in the ξth preset time period. The recommended score PS of each work is calculated by the formula: ; in the formula, p1, p2, and p3 represent reference coefficients, and p1+p2+p3=1.
2. The data tag based smart presentation system of claim 1, wherein, The work classification model is a composite model of a pre-trained convolutional neural network model and a recurrent neural network model.
3. The data tag based smart presentation system of claim 1, wherein, The weight coefficient θ corresponding to the ξth preset time period ξ By the following process: The interest index rK of each work category of the (ξ-1)th preset time period is obtained by the following formula: ξ rK ξ-1 Substituting into the following formula: ΔrK = rK ξ -rK ξ-1 The difference ΔrK of the interest index of each work category is calculated. If ΔrK> 0, then when ΔrK≥ ΔrK o+ Otherwise, If ΔrK< 0, then when ΔrK≤ ΔrK o- Otherwise, In the formula, ω is an adjustment coefficient, is an initial weight coefficient, ΔrK o+ is an initial weight coefficient, ΔrK o- is a preset early warning value.
4. The data tag based smart presentation system of claim 3, wherein, The expression of the work heat coefficient H is as follows: wherein m represents the number of terms of the work heat related parameters, ξ j (t) represents the curve of change of each term of the work heat related parameters with time, t0 and t1 represent the starting time and the ending time of the monitoring period respectively, ν j represents the preset proportional coefficient of each term of the work heat related parameters; j represents the jth term.
5. The data tag based smart presentation system of claim 1, wherein, The intelligent display module further comprises a user feedback collection unit used for collecting feedback information of the user on the displayed works, and the feedback information comprises a recommendation satisfaction degree. When the recommendation satisfaction degree is lower than a satisfaction threshold interval, the work categories in the preliminary work recommendation sequence are increased; when the recommendation satisfaction degree belongs to the satisfaction threshold interval, the work categories in the preliminary work recommendation sequence are kept; and when the recommendation satisfaction degree is higher than the satisfaction threshold interval, the work categories in the preliminary work recommendation sequence are reduced.
Citation Information
Patent Citations
Method and device for recommending service information
CN103927673A
Video recognition method and device and electronic equipment
CN112149632A