Personalized content recommendation method based on deep learning driving

Abstract

The invention relates to the technical field of content recommendation, and discloses a personalized content recommendation method based on deep learning driving, which comprises the following steps: collecting historical data behaviors of a user, acquiring the number of historical data behaviors collected for the user, setting a behavior quantity threshold value, and when the number of historical data behaviors is smaller than the behavior quantity threshold value, recommending the personalized content to the user. And marking the user as an inactive user, and executing a sparse data processing strategy. According to the method, discretized historical behavior data are mapped to a continuous and dense low-dimensional vector space by using an embedding technology in deep learning, potential semantic information and internal association of user behaviors can be better captured, and user basic attributes are matched by adopting a Siamese network, so that the user behavior can be more accurately and rapidly identified. Personalized recommendation can be generated with the help of behavior data of similar active users, and troubles caused by data sparsity are relieved.

Description

Technical Field

[0001] The present invention relates to the technical field of content recommendation, and particularly to a personalized content recommendation method driven by deep learning. Background Art

[0002] The personalized content recommendation method is a technology and strategy that provides customized content or services by analyzing users' interests, behaviors, and preferences. This method aims to improve the user experience and engagement, and common application scenarios include e-commerce product recommendations, news and information recommendations, video and music recommendations, etc.

[0003] It mainly extracts content features based on the content that users have liked or consumed, and then recommends new content with similar features. However, the problem of data sparsity generally exists in most recommendation systems, especially for less active users. Due to insufficient historical behavior data, it is difficult for the model to capture the true interests of users, which makes the recommendation method that simply relies on users' historical records ineffective. Summary of the Invention

[0004] Aiming at the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a personalized content recommendation method driven by deep learning, which can reasonably and accurately analyze the sparse data of inactive users to generate personalized content that meets the expected user preferences.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A personalized content recommendation method driven by deep learning, the method includes the following steps:

[0006] Collect the historical data behaviors of users, and obtain the number of historical data behaviors collected for the user. Set a behavior volume threshold. When the number of historical data behaviors is less than the behavior volume threshold, mark it as an inactive user and execute the sparse data processing strategy;

[0007] The sparse data processing strategy includes establishing an active user attribute library. The active user attribute library includes the historical data behaviors and basic attributes of active users, and obtains the basic attributes of inactive users. Use a Siamese network to compare and match the basic attribute features of active users with those of inactive users, and select the active user with the highest similarity to the basic attributes of the inactive user and mark it as a similar user;

[0008] Extract the content features from the historical data behaviors generated by inactive users, compare them with the content features in the historical data behaviors generated by all similar users, mark the similar users with the most similar content features to those of the inactive users as template users and obtain the highest feature similarity, set a similarity threshold, compare the highest feature similarity with the similarity threshold, and make corresponding responses according to the comparison results;

[0009] If the highest feature similarity is greater than or equal to the similarity threshold, temporarily regard the inactive user as a template user and recommend personalized content of the template user to the inactive user until the historical data behaviors generated by the inactive user reach the behavior volume threshold; if the highest feature similarity is less than the similarity threshold, generate personalized content recommendations based on both the content features extracted from the historical data behaviors of the inactive user and the content features of the template user.

[0010] In some embodiments, when the highest feature similarity is less than the similarity threshold, obtain the data volume ratio by dividing the number of historical data behaviors of the inactive user by the behavior volume threshold, and when making personalized content recommendations for the inactive user, recommend content that conforms to the user's own content features according to the ratio of the data volume ratio, and the remaining personalized content is recommended based on the content features of the template user.

[0011] In some embodiments, when the highest feature similarity is greater than or equal to the similarity threshold, obtain the average daily active times of the inactive user based on the number of historical data behaviors of the inactive user, and obtain the adjusted active times within a preset time after implementing the sparsity data processing strategy, compare the adjusted active times with the average daily active times, and make corresponding responses according to the comparison results.

[0012] In some embodiments, the specific way to obtain the average daily active times is to obtain the number of days from the day when the inactive user established the account to the present, and divide the number of historical data behaviors of the inactive user by the number of days from the day when the account was established to the present to obtain the average daily active times; the specific way to obtain the adjusted active times is to obtain the number of historical data behaviors generated by the inactive user within the preset time after implementing the sparsity data processing strategy, and divide the number of historical data behaviors by the preset time to obtain the adjusted active times.

[0013] In some embodiments, when the adjusted active times are greater than the average daily active times, continue to generate personalized content for the user based on the content features of the template user without performing additional operations; when the adjusted active times are less than or equal to the average daily active times, perform personalized content optimization, increase the proportion of the inactive user's own content features in the personalized recommendation, and reduce the dependence on the features of the template user.

[0014] In some embodiments, the specific way of performing personalized content optimization is to generate half of the generated content volume according to the content characteristics of the inactive user and the other half according to the content characteristics of the template user, or to consider the highest feature similarity of the user as less than the similarity threshold.

[0015] In some embodiments, when the adjusted active times are greater than the daily average active times, an adjusted improvement ratio is obtained. The specific way to obtain the adjusted improvement ratio is to find the difference between the adjusted active times and the daily average active times to get the daily extra times, and then divide the daily extra times by the daily average active times to obtain the adjusted improvement ratio. Then, a threshold for the improvement amount is set, and the adjusted improvement ratio is compared with the threshold for the improvement amount, and corresponding actions are taken according to the comparison result.

[0016] In some embodiments, if the adjusted improvement ratio is greater than or equal to the threshold for the improvement amount, continue to generate personalized content for the user according to the content characteristics of the template user without performing additional operations; if the adjusted improvement ratio is less than the threshold for the improvement amount, further user behavior judgment should be performed.

[0017] In some embodiments, performing further user behavior judgment specifically includes: generating 10% of the content volume in the generated content according to the content characteristics of the user, and marking this part of the content volume as user characteristic content, and increasing the priority of the user characteristic content so that the user can be preferentially recommended to the user characteristic content when using the product. If the predicted value of the historical data behavior times when browsing the user characteristic content is greater than the predicted value of the historical data behavior times when browsing the rest of the generated content, then the adjusted active times of the user should be considered as less than or equal to the daily average active times, and personalized content optimization should be performed to reduce the dependence on the template user characteristics; if the predicted value of the historical data behavior times when browsing the user characteristic content is less than or equal to the predicted value of the historical data behavior times when browsing the rest of the generated content, then continue to generate personalized content for the user according to the content characteristics of the template user without performing additional operations.

[0018] The present invention further provides a computer-readable storage medium storing a computer program, and the computer program is executed by a processor to implement the above-mentioned personalized content recommendation method driven by deep learning.

[0019] The technical solution provided by the present invention has the following beneficial effects compared with the prior art:

[0020] First, by using the embedding technology in deep learning, the present invention maps the discretized historical behavior data into a continuous and dense low-dimensional vector space, which can better capture the potential semantic information and internal associations of user behavior, improve the accuracy of personalized recommendations. Aiming at the problem of insufficient historical data behavior of inactive users, by establishing an active user attribute library and using a Siamese network to match user basic attributes, it is possible to generate personalized recommendations with the assistance of the behavior data of similar active users and alleviate the troubles caused by data sparsity.

[0021] Second, when the content feature similarity between an inactive user and an active user is relatively high, the present invention can directly generate recommendations with the assistance of a template user. Conversely, it generates recommended content by mixing the features of the inactive user and the template user, taking into account the personal preferences of the user and the experience of similar users, and ensuring that the recommended content is more in line with the actual interests of the user.

[0022] Third, by comparing the adjusted active times with the daily average active times and further evaluating the adjusted improvement ratio, the present invention can monitor the recommendation effect in real time. When the data change is not significant or the improvement is insufficient, it can also specifically increase the proportion of the features of the inactive user in the recommendation. This feedback mechanism ensures the dynamic optimization of the recommendation strategy and the robustness of the system.

[0023] Fourth, by adopting the method of dividing the generated content ratio and calculating the corresponding historical data behavior times of the two through estimated values, when the estimated data indicates that the user responds more actively to the content of their own features, the system will further reduce the dependence on the features of the template user, realizing a recommendation centered on the user's personality. It can not only cope with external interferences such as the natural fluctuations of user behavior or the overall activity of the platform, but also judge the recommendation effect using real-time behavior data feedback. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic flowchart of a personalized content recommendation method based on deep learning drive of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0026] It is understood that the term "a" should be construed as "at least one" or "one or more". That is, in one embodiment, the number of an element may be one, while in other embodiments, the number of the element may be multiple. The term "a" should not be construed as a limitation on the quantity.

[0027] The present invention provides a personalized content recommendation method driven by deep learning, as Figure 1 shown. The method includes the following steps:

[0028] Step 1: Collect the historical data behaviors of users, including basic behaviors such as click times, browsing time, payment amount, favorite number, comment number, etc. Use data preprocessing techniques to clean, normalize, and standardize the data to make different behavior data comparable. Utilize the embedding technique in deep learning to map the discretized behavior data into a continuous and dense low-dimensional vector space. In this process, an Embedding layer can be used to represent various behaviors of users, so as to capture the potential meaning and the association between behaviors. And obtain the number of historical data behaviors collected for this user, set a behavior quantity threshold, and compare the number of historical data behaviors with the behavior quantity threshold. When the number of historical data behaviors is greater than or equal to the behavior quantity threshold, it indicates that this user has generated a large amount of historical data behaviors, and mark it as an active user. Content features will be extracted based on the user's historical data behaviors, and new content with similar content features will be recommended to the user; when the number of historical data behaviors is less than the behavior quantity threshold, it indicates that the amount of historical data behaviors generated by this user is low, and mark it as an inactive user. Since the historical data behaviors generated by this user are insufficient, a sparse data processing strategy is executed;

[0029] Step 2: The sparse data processing strategy includes establishing an active user attribute library. The active user attribute library includes the historical data behaviors and basic attributes of active users. The basic attributes include the user's gender, age, geographical location, and device information. And obtain the basic attributes of inactive users, and compare the basic attributes of inactive users with the active user attributes in the active user database. The comparison method is specifically to adopt architectures such as Siamese network or contrast learning to compare and match the basic attribute features of active users and inactive users. By calculating the cosine similarity or Euclidean distance, the similarity between users is measured, and the active user with the highest similarity to the basic attributes of this inactive user is selected, and this type of active user is marked as a similar user;

[0030] Step 3: Extract the content features from the historical data behaviors generated by the inactive users, compare them with the content features from the historical data behaviors generated by all similar users, mark the similar user with the most similar content features to the inactive user as the template user, and simultaneously obtain the content feature similarity between the inactive user and the template user and mark it as the highest feature similarity. The highest feature similarity represents the maximum similarity of the content features that can be extracted from the historical data behaviors between the similar user and the inactive user, and also represents that the content feature behaviors extracted from the historical data behaviors of this similar user and this inactive user are the closest. Set a similarity threshold, compare the highest feature similarity with the similarity threshold, and make corresponding responses according to the comparison results;

[0031] Step 4: If the highest feature similarity is greater than or equal to the similarity threshold, it indicates that there is an active user in the active user attribute library whose historical behavior data is close to that of the inactive user. Then, temporarily regard this inactive user as the template user and recommend personalized content based on the template user to this inactive user until the historical data behaviors generated by this inactive user reach the behavior volume threshold. If the highest feature similarity is less than the similarity threshold, it indicates that there is no active user in the active user attribute library whose historical behavior data is close to that of the inactive user. Then, generate personalized content recommendations based on the content features extracted from the historical data behaviors of this inactive user (hereinafter referred to as content features for short) and the content features of the template user;

[0032] Specifically, when the highest feature similarity is less than the similarity threshold, divide the number of historical data behaviors of the inactive user by the behavior volume threshold to obtain the data volume ratio. When making personalized content recommendations for the inactive user, recommend content that conforms to the user's own content features to the user according to the ratio of the data volume ratio, and the remaining personalized content is recommended based on the content features of the template user. For example, assume that the number of collected user historical data behaviors is 20 times, the behavior volume threshold is set to 50 times, and the data volume ratio can be obtained as 0.4. Since the number of user historical data behaviors is less than the behavior volume threshold, mark this user as an inactive user, match the template user in the active user attribute library, set the highest feature similarity between the inactive user and the template user to 50%, and set the similarity threshold to 75%. Since the highest feature similarity is less than the similarity threshold, when generating personalized content recommendations for this inactive user, 40% of the generated content is generated based on the content features of this inactive user, and the other 60% is generated based on the content features of the template user.

[0033] After the sparse data processing strategy, when the highest feature similarity is greater than or equal to the similarity threshold, personalized content generated according to the template user will be recommended to inactive users. At this time, the average daily active times of the inactive users should also be obtained based on the historical data behavior times of the inactive users, and the adjusted active times should be obtained within a preset time after the execution of the sparse data processing strategy. Then, compare the adjusted active times with the average daily active times. When the adjusted active times are greater than the average daily active times, it indicates that after the execution of the sparse data processing strategy, the historical data behavior times generated by the users have been improved, reflecting that the personalized content recommended according to the template user conforms to the preferences of this inactive user, and the user is willing to increase the browsing volume and usage rate of the product. Then, continue to generate personalized content for this user according to the content characteristics of the template user without additional operations. When the adjusted active times are less than or equal to the average daily active times, it indicates that after the execution of the sparse data processing strategy, the historical data behavior times generated by the users have not been improved or have decreased, reflecting that the personalized content recommended according to the template user does not conform to the preferences of this inactive user, and the user's willingness to browse and use the product has decreased. Then, perform personalized content optimization, increase the proportion of the inactive user's own content characteristics in personalized recommendations, and reduce the dependence on the template user's characteristics. The specific way to perform personalized content optimization is to generate half of the generated content volume according to the content characteristics of this inactive user and the other half according to the content characteristics of the template user, or regard the highest feature similarity of this user as less than the similarity threshold, and recommend content that conforms to the user's own content characteristics to this user according to the proportion of the data volume ratio, and the remaining personalized content is recommended according to the content characteristics of the template user.

[0034] The specific method for obtaining the average daily active times is to obtain the number of days from the account creation date of the inactive user to the present, and divide the historical data behavior times of the inactive user by the number of days from the account creation date to the present to get the average daily active times. The specific method for obtaining the adjusted active times is to obtain the historical data behavior times generated by the inactive user from after implementing the sparsity data processing strategy to the preset time, and divide the historical data behavior times by the preset time to get the adjusted active times. Taking the above embodiments as an example, it is set that the highest feature similarity between the inactive user and the template user is 80%. Since the similarity threshold is 75% and the highest feature similarity is greater than the similarity threshold, the personalized content generated according to the template user is recommended to the inactive user. It is set that the number of days from the account creation date of the inactive user to the present is 20 days, and the number of historical data behavior times generated is 20 times, then the average daily active times is 1 time. It is set that the preset time is 5 days, and the number of historical data behavior times generated by the inactive user from after implementing the sparsity data processing strategy to the preset time is 10 times, then the adjusted active times is 2 times. Since the adjusted active times is greater than the average daily active times, it reflects that the personalized content recommended according to the template user conforms to the preferences of this inactive user, and then continue to generate personalized content for this user according to the content characteristics of the template user.

[0035] Meanwhile, when the adjusted active times is greater than the average daily active times, the adjusted improvement ratio should also be obtained, and a threshold for the improvement amount is set. Compare the adjusted improvement ratio with the threshold for the improvement amount and make corresponding responses according to the comparison results. Specifically, if the adjusted improvement ratio is greater than or equal to the threshold for the improvement amount, it indicates that after implementing the sparsity data processing strategy, the historical data behavior times generated by the user have been significantly improved, reflecting that the personalized content recommended according to the template user conforms to the preferences of this inactive user, and continue to generate personalized content for this user according to the content characteristics of the template user without any additional operations; if the adjusted improvement ratio is less than the threshold for the improvement amount, it indicates that after implementing the sparsity data processing strategy, although the historical data behavior times generated by the user have been improved, the improvement amount is not obvious, and it cannot be determined that it is necessarily caused by the change in the generated content. This is because the small improvement amount is not sufficient to exclude the influence of other external factors or random fluctuations in user behavior, and it cannot be clearly proved that the improved recommendation strategy (i.e., the recommended content generated by the template user) is the real decisive factor. That is to say, the insufficient improvement may be due to the natural fluctuations of the user's own usage habits, the overall activity change of the platform, or other interference factors, rather than the direct effect of the generated content strategy. At this time, further user behavior judgment should be performed;

[0036] The specific way to obtain the adjusted improvement ratio is to subtract the daily average active times from the adjusted active times to get the daily extra times, and then divide the daily extra times by the daily average active times to obtain the adjusted improvement ratio. The execution of further user behavior judgment specifically includes: selecting a small part of the content volume (such as 10% of the content volume) in the generated content for generation according to the user content characteristics, and marking this part of the content volume as user characteristic content, and increasing the priority of the user characteristic content, so that when the user uses the product, the user characteristic content can be preferentially recommended to the user, and comparing the estimated value of the historical data behavior times generated when the user browses the user characteristic content with the estimated value of the historical data behavior times generated when the user browses the rest of the generated content. If the estimated value of the historical data behavior times generated when browsing the user characteristic content is greater than the estimated value of the historical data behavior times generated when browsing the rest of the generated content, it means that the user responds more actively to the content customized based on their own characteristics, and the click, stay or other data behavior indicators are better, which indicates that the user has a more obvious preference for the content of their own characteristics. Then, the adjusted active times of the user should be regarded as less than or equal to the daily average active times, and personalized content optimization should be executed to reduce the dependence on the template user characteristics; if the estimated value of the historical data behavior times generated when browsing the user characteristic content is less than or equal to the estimated value of the historical data behavior times generated when browsing the rest of the generated content, it means that the user does not show a more positive response to the content customized based on their own characteristics, which indicates that the user has insufficient preference for the content of their own characteristics. Then, continue to generate personalized content for the user according to the content characteristics of the template user without additional operations; for example, assume that the number of days from the day when an inactive user established an account to the present is 20 days, and the number of historical data behavior times generated is 20 times, then the daily average active times is 1 time. Assume that the preset time is 5 days, and the number of historical data behavior times generated by the inactive user within the preset time after executing the sparsity data processing strategy is 6 times, then the adjusted active times is 1.2 times. The adjusted active times is greater than the daily average active times, and the daily extra times can be obtained as 0.2 times, and the adjusted improvement ratio is 20%. Assume that the improvement amount threshold is 35%. Since the adjusted improvement ratio is less than the improvement amount threshold, further user behavior judgment is executed. 10% of the content volume in the generated content is generated according to the user content characteristics, and this part of the content volume is marked as user characteristic content, and the remaining 90% of the content volume is still generated according to the content characteristics of the template user. Assume that the user generates 5 historical data behavior times when browsing the user characteristic content, and generates 10 historical data behavior times when browsing the rest of the generated content (the content generated according to the template user characteristics). From this, the estimated value of the historical data behavior times generated when browsing the user characteristic content can be obtained as 5÷0.1 = 50, and the estimated value of the historical data behavior times generated when browsing the rest of the generated content is 10÷0.9 = 11.1. That is, when all content is generated based on user characteristics, the estimated number of historical data behaviors generated by the user is 50 times. When all content is generated based on template user characteristics, the estimated number of historical data behaviors generated is 11.1 times. This indicates that the user has a stronger preference for content with their own characteristics. The user's adjusted active count is considered less than or equal to the average daily active count, and personalized content optimization is performed.

[0037] In general, the present invention aims to design a personalized content recommendation method driven by deep learning. To address the problem that the real interests of low-activity users are difficult to accurately analyze due to data sparsity, the present invention uses the embedding technology in deep learning to map discretized historical behavior data into a continuous and dense low-dimensional vector space, which can better capture the potential semantic information and intrinsic correlation of user behavior and improve the accuracy of personalized recommendations. To address the problem of insufficient historical behavior data of inactive users, by establishing an active user attribute library and using Siamese networks or comparative learning to match user basic attributes, personalized recommendations can be generated with the help of behavioral data of similar active users, alleviating the problems caused by data sparsity. When the content feature similarity between inactive users and active users is high, recommendations can be generated directly with the help of template users. Otherwise, recommended content can be generated by mixing the inactive user's own features with the template user's features, taking into account the user's personal preferences and the experience of similar users, ensuring that the recommended content is more in line with the user's actual interests. By comparing adjusted active counts with average daily active counts and further evaluating the adjusted lift ratio, the effectiveness of recommendations can be monitored in real time. When data changes are insignificant or lift is insufficient, the weight of inactive users' characteristics in recommendations can be increased. This feedback mechanism ensures dynamic optimization of the recommendation strategy and system robustness. Furthermore, by dividing the generated content ratio and calculating the corresponding historical data behavior counts based on the estimated value, users' true response to their own characteristic content can be more intuitively judged. When the estimated data indicates that users respond more positively to their own characteristic content, the system further reduces its reliance on template user characteristics, achieving truly personalized recommendations. By measuring the adjusted lift ratio, the system not only addresses external fluctuations in user behavior or overall platform activity, but also leverages real-time behavioral data feedback to assess recommendation effectiveness, avoiding misinterpreting small lifts caused by external factors as effective recommendation strategies. By dynamically adjusting the content recommendation ratio based on users' historical data and real-time feedback, recommended content is ultimately more aligned with users' true interests, thereby increasing user clicks, browsing, and interaction, enhancing platform user stickiness and product activity.

[0038] Embodiments disclosed by the present invention. The processes described above with reference to the flowcharts can be implemented as computer software programs. Embodiments disclosed by the present invention include a computer program product, which includes a computer program carried on a computer-readable medium. The computer program contains program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part, and / or installed from a removable medium. When the computer program is executed by the central processing unit, the above functions defined in the methods of the present application are executed. It should be noted that the above computer-readable medium in the present application can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wire segments, a portable computer disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical fiber, a portable compact disk read-only memory, an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or combined with an instruction execution system, device, or component. In the present application, the computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries the computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, and this computer-readable medium can send, propagate, or transmit a program for use by or combined with an instruction execution system, device, or component. The program code contained on the computer-readable medium can be transmitted by any appropriate medium, including but not limited to: wireless segments, wire segments, optical cables, RF, etc., or any suitable combination of the above.

[0039] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions denoted in the blocks may occur in a different order than that denoted in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0040] Those skilled in the art should understand that the above description is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered within the protection scope of the present application.

Claims

1. A personalized content recommendation method driven by deep learning, characterized in that, The method includes the following steps: Collect the historical data behaviors of the user, obtain the number of historical data behaviors collected for this user, set a behavior volume threshold. When the number of historical data behaviors is less than the behavior volume threshold, mark it as an inactive user and execute a sparsity data processing strategy; The sparsity data processing strategy includes establishing an active user attribute library. The active user attribute library includes the historical data behaviors and basic attributes of active users, and obtain the basic attributes of inactive users. Use a Siamese network to compare and match the basic attribute features of active users with those of inactive users, and select the active user with the highest similarity to the basic attributes of the inactive user and mark it as a similar user; Extract the content features in the historical data behaviors already generated by the inactive user, compare them with the content features in the historical data behaviors already generated by all similar users, and mark the similar user with the most similar content features to the inactive user as a template user and obtain the highest feature similarity. Set a similarity threshold, compare the highest feature similarity with the similarity threshold, and make corresponding responses according to the comparison results; If the highest feature similarity is greater than or equal to the similarity threshold, temporarily regard this inactive user as a template user, and recommend personalized content according to the template user to this inactive user until the number of historical data behaviors generated by this inactive user reaches the behavior volume threshold; if the highest feature similarity is less than the similarity threshold, generate personalized content recommendations based on both the content features extracted from the historical data behaviors of the inactive user and the content features of the template user.

2. The personalized content recommendation method based on deep learning drive according to claim 1, wherein When the highest feature similarity is less than the similarity threshold, obtain the data volume ratio by dividing the number of historical data behaviors of the inactive user by the behavior volume threshold, and when making personalized content recommendations for the inactive user, recommend content that conforms to the user's own content features according to the ratio of the data volume ratio to this user, and the remaining personalized content is recommended according to the content features of the template user.

3. The personalized content recommendation method based on deep learning drive according to claim 2, wherein When the highest feature similarity is greater than or equal to the similarity threshold, obtain the average daily active times of the inactive user based on the number of historical data behaviors of the inactive user, and obtain the adjusted active times within a preset time after executing the sparsity data processing strategy. Compare the adjusted active times with the average daily active times, and make corresponding responses according to the comparison results.

4. A personalized content recommendation method based on deep learning drive according to claim 3, characterized in that The specific way to obtain the average daily active times is to obtain the number of days from the day when the inactive user established the account to the present, and divide the number of historical data behaviors of the inactive user by the number of days from the day when the account was established to the present to obtain the average daily active times; The specific way to obtain the adjusted active times is to obtain the number of historical data behaviors generated by the inactive user within the preset time after executing the sparsity data processing strategy, and divide the number of historical data behaviors by the preset time to obtain the adjusted active times.

5. A personalized content recommendation method driven by deep learning according to claim 4, characterized in that, When the adjusted active times are greater than the average daily active times, continue to generate personalized content for this user according to the content features of the template user without performing additional operations; When the adjusted active times are less than or equal to the daily average active times, personalized content optimization is performed, increasing the proportion of the content features of inactive users themselves in personalized recommendations and reducing the dependence on the content features of template users.

6. A personalized content recommendation method based on deep learning drive according to claim 5, characterized in that, Regarding the specific way of performing personalized content optimization, half of the generated content volume is generated according to the content features of the inactive user, and the other half is generated according to the content features of the template user, or the highest feature similarity of the user is regarded as less than the similarity threshold.

7. A personalized content recommendation method driven by deep learning according to claim 6, characterized in that, When the adjusted active times are greater than the daily average active times, an adjusted increase ratio is obtained. The specific way to obtain the adjusted increase ratio is to subtract the daily average active times from the adjusted active times to get the daily extra times, and then divide the daily extra times by the daily average active times to obtain the adjusted increase ratio. A threshold for the increase amount is set, and the adjusted increase ratio is compared with the threshold for the increase amount, and corresponding actions are taken according to the comparison result.

8. A personalized content recommendation method driven by deep learning according to claim 7, characterized in that If the adjusted increase ratio is greater than or equal to the threshold for the increase amount, continue to generate personalized content for the user according to the content features of the template user without performing additional operations; if the adjusted increase ratio is less than the threshold for the increase amount, further user behavior judgment should be performed.

9. A personalized content recommendation method driven by deep learning according to claim 8, characterized in that, Performing further user behavior judgment specifically includes: generating 10% of the content volume in the generated content according to the content features of the user, marking this part of the content volume as user feature content, and increasing the priority of the user feature content so that the user can be preferentially recommended to the user feature content when using the product. If the predicted value of the historical data behavior times when browsing the user feature content is greater than the predicted value of the historical data behavior times when browsing the rest of the generated content, then the adjusted active times of the user should be regarded as less than or equal to the daily average active times, and personalized content optimization is performed to reduce the dependence on the content features of template users; if the predicted value of the historical data behavior times when browsing the user feature content is less than or equal to the predicted value of the historical data behavior times when browsing the rest of the generated content, then continue to generate personalized content for the user according to the content features of the template user without performing additional operations.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement a personalized content recommendation method according to any one of claims 1-9 above.

Images