News Recommendation Method and System for Dynamically Configuring Recall Strategies Based on User Profiles

Abstract

The present invention relates to a news recommendation method and system for dynamically configuring a recall strategy based on a user profile, including collecting historical behavior data of a user to construct a user profile and extracting user profile features; dynamically configuring a recall strategy based on the user profile features, and selecting an optimal recall channel by calculating the weights and scores of recall channels; recalling news from a news database according to the user profile features and the recall strategy determined based on the optimal recall channel, generating a personalized recommendation list through a sorting model, and optimizing the recommendation result in real time according to user behavior feedback. By collecting and analyzing user behavior data, constructing an accurate user profile, dynamically configuring a recall strategy, and optimizing the recommendation result according to real-time user feedback, the present invention can significantly improve the accuracy of news recommendation and user satisfaction, and enhance the user experience.

Description

Technical Field

[0001] The present invention relates to the technical field of news recommendation, and in particular, to a news recommendation method and system based on dynamically configuring a recall strategy according to a user profile. Background Art

[0002] In modern information society, users' demand for news and information is increasing day by day. However, in the face of a vast amount of information, how to push the news that best meets the interests and needs of users to users has become an important technical challenge. Existing news recommendation systems usually rely on fixed recommendation algorithms and strategies, and it is difficult to respond to changes in users' interests in real time, resulting in low accuracy of recommendation results and low user satisfaction.

[0003] Traditional recommendation methods mainly include collaborative filtering, content-based recommendation, and hybrid recommendation strategies. These methods have the following deficiencies: First, collaborative filtering methods rely on a large amount of user behavior data and it is difficult to provide accurate recommendations in cold start situations; Second, although content-based recommendation methods can solve the cold start problem to a certain extent, their capture of users' interests is relatively static and it is difficult to dynamically reflect changes in users' interests; Finally, although hybrid recommendation methods combine multiple recommendation strategies, their complexity is high and real-time performance is poor, and it is difficult to make dynamic adjustments according to users' immediate feedback. Summary of the Invention

[0004] The purpose of the present invention is to provide a news recommendation method and system based on dynamically configuring a recall strategy according to a user profile to solve the problems of poor accuracy and real-time performance of existing technical results.

[0005] To achieve one of the above-mentioned invention purposes, an embodiment of the present invention provides a news recommendation method based on dynamically configuring a recall strategy according to a user profile. The method includes:

[0006] Collecting historical behavior data of users to construct a user profile and extracting user profile features;

[0007] Based on user profile features, dynamically configuring a recall strategy, and selecting an optimal recall channel by calculating the weights and scores of recall channels;

[0008] Recalling news from a news database according to user profile features and a recall strategy determined based on the optimal recall channel, generating a personalized recommendation list through a sorting model, and optimizing the recommendation result in real time according to user behavior feedback.

[0009] As a further improvement of an embodiment of the present invention, the method further includes that the "collecting historical behavior data of users to construct a user profile and extracting user profile features" includes:

[0010] The historical behavior data of the collected user includes data on user browsing records, click records, search records, subscription records, sharing behaviors, comments, likes, and stay time;

[0011] Preprocess the historical behavior data of the user to construct a user portrait; the preprocessing includes data cleaning and filtering of noise data;

[0012] Extract features from the preprocessed historical behavior data of the user to generate m user portrait features ; where, is the total number of user portrait features;

[0013] The feature extraction includes extracting the user's interest points, active time periods, active regions, and preference categories.

[0014] As a further improvement of an embodiment of the present invention, the method further includes that the dynamic configuration of the recall strategy based on the user portrait features includes,

[0015] Based on user portrait features to determine recall channels ;

[0016] Recall channel is determined by m_i user portrait features where m_i is the number of user portrait features on which the i-th recall channel is based, and m_i ≤ m, and i represents the index of the recall channel;

[0017] Each user portrait feature u_ij corresponds to an update time of .

[0018] As a further improvement of an embodiment of the present invention, the method further includes calculating the weights of the recall channels, including,

[0019] Calculate the weight of each user portrait feature, and the calculation formula is:

[0020]

[0021] where, is the weight of the user portrait feature , is the j-th user portrait feature of the i-th recall channel, is the current time, is the update time of u_ij, is the time influence period of the user portrait feature;

[0022] Calculate the global recall weight of all user portrait features, and the calculation formula is:

[0023]

[0024] Among them, is the global recall weight of the i-th channel, is an empirical constant, and its value ranges from [0, 1], is the recommended click-through rate of the i-th recall channel within the time impact period, is the maximum click-through rate of all recall channels within the time impact period, is the minimum click-through rate of all recall channels within the time impact period.

[0025] As a further improvement of an embodiment of the present invention, the method further includes that calculating the score of the recall channel includes,

[0026] Calculating the personalized recall weight of user k, and the calculation formula is:

[0027]

[0028] Among them, is the weight of the j-th user portrait feature of the i-th recall channel of user k, is the user portrait feature of user k is the corresponding update time, is the personalized weight of the i-th recall channel of user k;

[0029] Based on the global recall weight and the personalized recall weight, calculating the comprehensive recall score of the recall channel, and the calculation formula is:

[0030]

[0031] Among them, is the comprehensive recall score of the i-th recall channel of user k, is the basic score of the i recall channels; is a constant for determining whether the recall channel is an offline channel. When the recall channel is an offline channel, , otherwise ω = 0; is the offline recall coefficient, is the real-time recall coefficient, is an empirical constant, and its value range is [0, 1].

[0032] As a further improvement of an embodiment of the present invention, the method further includes that the selection of the optimal recall channel includes,

[0033] For the recall channel r_i according to the recall score Perform reverse sorting and select the top n recall channels with the highest recall scores as the optimal recall channels for user k.

[0034] As a further improvement of an embodiment of the present invention, the method further includes that "recalling news from the news database according to the user portrait features and the recall strategy determined based on the optimal recall channels, and generating a personalized recommendation list through a sorting model" includes,

[0035] Preliminarily screen candidate news from the news database according to the user portrait features, and obtain the recall news list for each user according to the recall strategy;

[0036] By calculating the comprehensive recall scores of each candidate news item in the recall news list, prioritize each candidate news item, and select a part of the news items with the highest scores to generate a recommendation list.

[0037] To achieve one of the above-mentioned invention purposes, an embodiment of the present invention further provides a news recommendation system for dynamically configuring recall strategies based on user portraits, and the system includes an acquisition module, a calculation module, and a recommendation module;

[0038] The acquisition module is used to collect the historical behavior data of users to construct user portraits and extract user portrait features;

[0039] The calculation module is used to dynamically configure recall strategies based on user portrait features, and select the optimal recall channels by calculating the weights and scores of recall channels;

[0040] The recommendation module is used to recall news from the news database according to the user portrait features and the recall strategy determined based on the optimal recall channels, generate a personalized recommendation list through a sorting model, and optimize the recommendation results in real time according to user behavior feedback.

[0041] To achieve one of the above-mentioned invention purposes, an embodiment of the present invention further provides an electronic device, including a memory and a processor, characterized in that a computer program that can run on the processor is stored in the memory, and when the program is executed on the processor, the steps in the above-mentioned news recommendation method for dynamically configuring recall strategies based on user portraits are implemented.

[0042] To achieve one of the above-mentioned invention purposes, an embodiment of the present invention further provides a storage medium, and the storage medium stores a computer program, characterized in that when the computer program is executed by a processor, the steps in the above-mentioned news recommendation method for dynamically configuring recall strategies based on user portraits are implemented.

[0043] Compared with the prior art, the present invention provides a news recommendation method and system for dynamically configuring a recall strategy based on a user profile. By collecting and analyzing user behavior data, constructing an accurate user profile, dynamically configuring the recall strategy, and optimizing the recommendation results according to the user's real-time feedback, the accuracy of news recommendation and user satisfaction can be significantly improved, and the user experience can be enhanced. Brief Description of the Drawings

[0044] Figure 1 is the overall flowchart of the news recommendation method for dynamically configuring a recall strategy based on a user profile according to the present invention.

[0045] Figure 2 is the schematic architecture diagram of the news recommendation system for dynamically configuring a recall strategy based on a user profile according to the present invention. Detailed Embodiments

[0046] The present invention will be described in detail below with reference to the specific embodiments shown in the drawings. However, these embodiments do not limit the present invention, and any structural, methodical, or functional transformation made by those of ordinary skill in the art based on these embodiments is included within the protection scope of the present invention.

[0047] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present invention and should not be construed as limiting the present invention.

[0048] In the first embodiment of the present invention, an embodiment of the present invention provides a news recommendation method for dynamically configuring a recall strategy based on a user profile, as Figure 1 shown, the method includes,

[0049] S1: Collect the historical behavior data of the user to construct a user profile and extract the user profile features;

[0050] S2: Dynamically configure the recall strategy based on the user profile features, and select the optimal recall channel by calculating the weights and scores of the recall channels;

[0051] S3: Recall news from the news database according to the user profile features and the recall strategy determined based on the optimal recall channel, generate a personalized recommendation list through a sorting model, and optimize the recommendation results in real time according to the user behavior feedback.

[0052] In a specific embodiment of the present invention, collecting the historical behavior data of the user to construct a user profile and extracting the user profile features is specifically,

[0053] Collecting the historical behavior data of the user, including the data of the user's browsing records, click records, search records, subscription records, sharing behaviors, comments, likes, and stay time;

[0054] Preprocessing the historical behavior data of the user to construct a user portrait; the preprocessing includes data cleaning and filtering out noise data;

[0055] Performing feature extraction on the preprocessed historical behavior data of the user to generate m user portrait features ; where is the total number of user portrait features;

[0056] The feature extraction includes extracting the user's points of interest, active time periods, active regions, and preference categories.

[0057] It should be noted that before performing feature extraction, it is necessary to preprocess the historical behavior data of the user. The preprocessing includes data cleaning and filtering out noise data to ensure the accuracy and reliability of subsequent analysis. Data cleaning mainly removes invalid, incorrect, or duplicate data, and filtering out noise data excludes interfering data irrelevant to the construction of the user portrait. After completing the preprocessing, enter the feature extraction stage. Feature extraction includes extracting key features from the user's behavior data, such as the user's points of interest (i.e., the types of content the user most frequently browses and clicks), active time periods (the time periods when the user is most active in a day), active regions (the geographical locations where the user most frequently conducts activities), and preference categories (the content categories the user is most interested in). These features can reflect the user's behavior patterns and interest tendencies from multiple perspectives, jointly constituting the user portrait, which details each user's interests and behavior patterns and provides an accurate data basis for subsequent recommendation algorithms.

[0058] In a specific embodiment of the present invention, based on the user portrait features, a recall strategy is dynamically configured. Specifically,

[0059] Based on user portrait features, determine recall channels ;

[0060] The recall channels are determined by m_i user portrait features , where m_i is the number of user portrait features on which the i-th recall channel is based, and m_i ≤ m, and i represents the index of the recall channel;

[0061] Each user portrait feature u_ij corresponds to an update time of .

[0062] It should be noted that the update time corresponding to each user portrait feature reflects the latest degree of user behavior data. By analyzing the update time, the timeliness of each feature can be evaluated to ensure that the recall channel can dynamically respond to changes in user interests and provide the latest and most relevant news content.

[0063] In a specific embodiment of the present invention, the weight of the recall channel is calculated. Specifically,

[0064] Calculate the weight of each user portrait feature. The calculation formula is:

[0065]

[0066] Where, is the weight of the user portrait feature ; is the j-th user portrait feature of the i-th recall channel, is the current time, is the update time of u_ij, is the time influence period of the user portrait feature;

[0067] Calculate the global recall weight of all user portrait features. The calculation formula is:

[0068]

[0069] Where, is the global recall weight of the i-th channel, is an empirical constant, and its value is between [0, 1], is the recommended click-through rate of the i-th recall channel within the time influence period, is the maximum click-through rate of all recall channels within the time influence period, is the minimum click-through rate of all recall channels within the time influence period.

[0070] It should be noted that in order to more comprehensively evaluate the collinearity of each recall channel and provide more reliable weights for subsequent fusion sorting, it is necessary to comprehensively consider the timeliness of user portrait features, the overall performance of the channel, and the adjustment of empirical parameters;

[0071] Furthermore, by calculating the weight of the user portrait feature, the contribution degree of each user portrait feature to the current recommendation is measured. From the formula, it can be seen that the weight of the user portrait feature decays linearly over time. The newer the user, the higher the weight. When the update time of the user portrait feature exceeds , its weight is regarded as 0 and no longer affects the recommendation.

[0072] Further, by calculating the global recall weight, the overall weight of each recall channel is calculated, comprehensively considering the weight of user portrait features and the overall performance of the channel; the first item calculates the average value of the weights of all user portrait features, reflecting the overall quality of the user portrait features of this channel, and the second item calculates the relative position of the click-through rate of this channel among all channels, reflecting the overall performance of this channel; by adjusting the value of β, the influence ratio of the weight of user portrait features and the click-through rate of the channel in the final weight can be controlled; when β is close to 1, the weight of user portrait features has a greater influence; when β is close to 0, the click-through rate of the channel has a greater influence.

[0073] In a specific embodiment of the present invention, the score of the recall channel is calculated, specifically,

[0074] Calculate the personalized recall weight of user k, and the calculation formula is:

[0075]

[0076] where, is the weight of the jth user portrait feature of the ith recall channel of user k, is the user portrait feature of user k corresponding update time, is the personalized weight of the ith recall channel of user k;

[0077] Based on the global recall weight and the personalized recall weight, calculate the comprehensive recall score of the recall channel, and the calculation formula is:

[0078]

[0079] where, is the comprehensive recall score of the ith recall channel of user k, is the basic score of the ith recall channel; is a constant for judging whether the recall channel is an offline channel. When the recall channel is an offline channel, , otherwise ω = 0; is the offline recall coefficient, is the real-time recall coefficient, is an empirical constant, and its value range is [0, 1].

[0080] It should be noted that by calculating the comprehensive recall score, the basic score of the recall channel, the coefficients of offline and real-time recall, and the influence of global and personalized weights are considered. The final comprehensive recall score is calculated by the method of weighted average. This method can more accurately evaluate the comprehensive score of each recall channel by combining global and personalized weights and considering the different characteristics of offline and real-time recall, so as to select the optimal recall channel in the news recommendation process, improving the accuracy of the recommendation result and user satisfaction.

[0081] In a specific embodiment of the present invention, selecting the optimal recall channel is specifically as follows:

[0082] Sort the recall channels \(r_i\) in descending order according to the recall score and select the top \(n\) recall channels with the highest recall scores as the optimal recall channels for user \(k\).

[0083] It should be noted that when sorting, descending order is adopted, that is, the recall channels with higher scores are arranged in the front. In this way, it can be intuitively seen which recall channels have higher recommendation value and effect based on the current user portrait features and historical behavior data. Select the top \(n\) recall channels with the highest scores from the sorted recall channels as the optimal recall channels for user \(k\). These optimal recall channels will be used as the main channels for news recall in the subsequent recommendation process. By selecting the top \(n\) channels with the highest scores, it can be ensured that the recommended news content can satisfy the user's interests and preferences as much as possible, thus improving the accuracy of the recommendation result and user satisfaction.

[0084] In a specific embodiment of the present invention, according to the user portrait features and the recall strategy determined based on the optimal recall channel, recall news from the news database and generate a personalized recommendation list through a sorting model, specifically as follows:

[0085] Preliminarily screen candidate news from the news database according to the user portrait features, and obtain the recall news list for each user according to the recall strategy;

[0086] By calculating the comprehensive recall score of each candidate news item in the recall news list, prioritize each candidate news item, and select a part of the news items with the highest scores to generate a recommendation list.

[0087] It should be noted that the matching degree between each news and the user's interests is determined by calculating the comprehensive recall score of the candidate news items. The calculation of the comprehensive recall score synthesizes various factors, including the relevance of the news content, the timeliness of the news, the user's historical behavior data, etc. Through such calculation, each candidate news will obtain a specific score, reflecting its recommendation value for the current user. All candidate news items are sorted according to the comprehensive recall score. When sorting, reverse sorting is adopted, that is, the news with higher comprehensive recall score is ranked in the front. This can ensure that the most relevant and user-interest-compliant news is preferentially presented to the user. Select a part of the news items with the highest comprehensive recall score to generate a personalized recommendation list. Determine the specific number of recommended news according to the user's reading habits and the demand for the number of news. The generated recommendation list will preferentially display the news with the highest score, ensuring the high quality and high relevance of the recommended content, thereby enhancing the user's reading experience and satisfaction.

[0088] In Embodiment 2 of the present invention, the present invention provides a news recommendation system based on dynamically configuring a recall strategy according to a user portrait, as Figure 2 shown, the system includes a collection module 1, a calculation module 2, and a recommendation module 3;

[0089] The collection module 1 is used to collect the user's historical behavior data to construct a user portrait and extract user portrait features;

[0090] The calculation module 2 is used to dynamically configure a recall strategy based on the user portrait features, and select an optimal recall channel by calculating the weights and scores of the recall channels;

[0091] The recommendation module 3 is used to recall news from the news database according to the user portrait features and the recall strategy determined based on the optimal recall channel, generate a personalized recommendation list through a sorting model, and optimize the recommendation result in real time according to the user behavior feedback.

[0092] In Embodiment 3 of the present invention, the present invention provides an electronic device, including a memory and a processor, characterized in that a computer program that can run on the processor is stored in the memory, and when the program is executed on the processor, the steps in the above-mentioned news recommendation method based on dynamically configuring a recall strategy according to a user portrait are implemented.

[0093] In Embodiment 4 of the present invention, the present invention provides a storage medium, the storage medium stores a computer program, characterized in that when the computer program is executed by a processor, the steps in the above-mentioned news recommendation method based on dynamically configuring a recall strategy according to a user portrait are implemented.

[0094] In summary, the news recommendation method and system based on dynamically configuring recall strategies according to user portraits provided by the present invention can significantly improve the accuracy of news recommendations and user satisfaction, and enhance the user experience by collecting and analyzing user behavior data, constructing accurate user portraits, dynamically configuring recall strategies, and optimizing recommendation results according to real-time user feedback.

[0095] It should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative manner of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0096] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described modules can refer to the corresponding processes in the foregoing method embodiments and will not be described herein again.

[0097] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0098] In addition, the functional modules in each embodiment of the present application can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or in the form of a hardware plus software functional module.

[0099] The above integrated modules implemented in the form of software functional modules can be stored in a computer-readable storage medium. The above software functional modules are stored in a storage medium and include several instructions to enable a computer system (which can be a personal computer, a server, or a network system, etc.) or a processor to execute some steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A news recommendation method for dynamically configuring a recall strategy based on user portraits, characterized in that: include, Collect users' historical behavior data to build user portraits and extract user portrait features; Based on user profile features, dynamically configure the recall strategy and select the optimal recall channel by calculating the weight and score of the recall channel; The dynamic configuration recall strategy based on user portrait features includes: based on User portrait features are determined Recall Channel ; Recall Channel From m_i user portrait features Determine, where m_i is the number of user portrait features based on the i-th recall channel, and m_i≤m, i represents the index of the recall channel; The update time corresponding to each user portrait feature u_ij is ; Calculating the weight of the recall channel includes, Calculate the weight of each user portrait feature using the following formula: in, User portrait features The weight of is the j-th user portrait feature of the i-th recall channel, is the current time, is the update time of u_ij, The time impact period of user portrait features; Calculate the global recall weight of all user portrait features using the following formula: in, is the global recall weight of the ith channel, is an empirical constant, with a value between [0, 1], is the recommendation click rate of the i-th recall channel within the time impact period, is the maximum click rate within the time impact period of all recall channels, The minimum click rate within the time impact period of all recall channels; According to the user portrait characteristics and the recall strategy determined based on the optimal recall channel, news is recalled from the news database, a personalized recommendation list is generated through the sorting model, and the recommendation results are optimized in real time based on user behavior feedback.

2. The news recommendation method based on user portrait dynamic configuration recall strategy according to claim 1 is characterized by: The "collecting historical behavior data of users to construct user portraits and extracting user portrait features" includes: The collected user historical behavior data includes user browsing history, click history, search history, subscription history, sharing behavior, comments and likes, and stay time data; Preprocessing the historical behavior data of the user to construct a user profile; The preprocessing includes data cleaning and filtering noise data; Perform feature extraction on the preprocessed user's historical behavior data to generate m user portrait features ;in, is the total number of user portrait features; The feature extraction includes extracting the user's points of interest, active time periods, active areas, and preference categories.

3. The news recommendation method based on user portrait dynamic configuration recall strategy according to claim 1 is characterized by: Calculating the recall channel score includes, Calculate the personalized recall weight of user k. The calculation formula is: in, is the weight of the j-th user portrait feature of the i-th recall channel of user k, User portrait features for user k The corresponding update time, is the personalized weight of the i-th recall channel of user k; Based on the global recall weight and the personalized recall weight, the comprehensive recall score of the recall channel is calculated using the following formula: in, is the comprehensive recall score of the i-th recall channel of user k, is the basic score of the i-th recall channel; A constant to determine whether the recall channel is an offline channel. When the recall channel is an offline channel, , otherwise ω=0; is the offline recall coefficient, is the real-time recall coefficient, is an empirical constant, ranging from [0, 1].

4. The news recommendation method based on user portrait dynamic configuration recall strategy according to claim 3 is characterized by: The selecting of the optimal recall channel includes: For the recall channel r_i, according to the recall score Sort in descending order and select the first n recall channels with the highest recall scores as the optimal recall channels for user k.

5. The news recommendation method based on user portrait dynamic configuration recall strategy according to claim 1 is characterized by: The "recalling news from the news database based on user portrait features and a recall strategy determined based on an optimal recall channel, and generating a personalized recommendation list through a sorting model" includes: Preliminarily screening candidate news from a news database according to the user portrait features, and obtaining a recalled news list for each user according to the recall strategy; By calculating the comprehensive recall score of each candidate news item in the recalled news list, each candidate news item is prioritized, and a portion of news items with the highest scores are selected to generate a recommendation list.

6. A news recommendation system for dynamically configuring a recall strategy based on user portraits, used to implement the steps in the news recommendation method for dynamically configuring a recall strategy based on user portraits as claimed in claim 1, characterized in that: It includes acquisition module, calculation module and recommendation module; The collection module is used to collect the user's historical behavior data to construct a user portrait and extract user portrait features; The calculation module is used to dynamically configure the recall strategy based on the user portrait features, and select the optimal recall channel by calculating the weights and scores of the recall channels; The recommendation module is used to recall news from the news database according to user portrait features and a recall strategy determined based on the optimal recall channel, generate a personalized recommendation list through a sorting model, and optimize the recommendation results in real time according to user behavior feedback.

7. An electronic device, comprising a memory and a processor, characterized in that: The memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps in the news recommendation method for dynamically configuring a recall strategy based on a user portrait as described in any one of claims 1 to 5 are implemented.

8. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the news recommendation method for dynamically configuring a recall strategy based on a user portrait as described in any one of claims 1 to 5 are implemented.

Images