Information recommendation method and device, electronic equipment and readable storage medium

Abstract

The application discloses an information recommendation method and device, electronic equipment and a readable storage medium, and belongs to the technical field of electronics. The method comprises the following steps: acquiring a recommendation content set, wherein the recommendation content set comprises a plurality of first recommendation contents; based on portrait feature information of a target user, user attribute information of the target user, scene information of a current scene in which the target user is located, and content information of a second recommendation content, a dense preference value of the target user to a first content type to which the second recommendation content belongs is calculated, the second recommendation content is at least one of the plurality of first recommendation contents, and the dense preference value is used for representing a dense degree of content corresponding to the first content type in the recommendation content set; and based on the dense preference value, the recommendation content in the recommendation content set is adjusted.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, specifically relating to an information recommendation method, apparatus, electronic device, and readable storage medium. Background Technology

[0002] With the development of internet technology, users are faced with information feeds on electronic devices that recommend different content to them within applications. These scenarios make users' lives more convenient and also allow them to access more knowledge and broaden their horizons.

[0003] However, when electronic device applications densely recommend the same type of content, it can cause user fatigue and greatly affect the effectiveness of the recommended content and the user experience.

[0004] Therefore, how to adjust the densely recommended content in the information feed is an urgent problem to be solved. Summary of the Invention

[0005] The purpose of this application is to provide an information recommendation method, apparatus, electronic device, and readable storage medium that can solve the problem of how to adjust densely recommended content in information flow recommendations.

[0006] In a first aspect, embodiments of this application provide an information recommendation method, which includes: obtaining a set of recommended content, the set of recommended content containing multiple first recommended content; calculating a density preference value of the target user for a first content type to which the second recommended content belongs, based on the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommended content, wherein the second recommended content is at least one of the multiple first recommended content, and the density preference value is used to characterize the density of the target user's preference for the content corresponding to the first content type in the set of recommended content; and adjusting the recommended content in the set of recommended content based on the density preference value.

[0007] Secondly, embodiments of this application provide an information recommendation device, which includes: an acquisition module and a processing module. The acquisition module is used to acquire a set of recommended content, which includes multiple first recommended content items. The processing module is used to calculate a density preference value of the target user for a first content type to which the second recommended content belongs, based on the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommended content. The second recommended content is at least one of the multiple first recommended content items. The density preference value is used to characterize the density of the target user's preference for the content corresponding to the first content type in the recommended content set. The processing module is also used to adjust the recommended content in the recommended content set based on the density preference value.

[0008] Thirdly, embodiments of this application provide an electronic device including a processor and a memory, wherein the memory stores programs or instructions executable on the processor, and the programs or instructions, when executed by the processor, implement the steps of the method described in the first aspect.

[0009] Fourthly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0010] Fifthly, embodiments of this application provide a chip, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the method as described in the first aspect.

[0011] In a sixth aspect, embodiments of this application provide a computer program product stored in a storage medium, which is executed by at least one processor to implement the method described in the first aspect.

[0012] In this embodiment, a recommended content set is obtained, which includes multiple first recommended content items. Based on the target user's profile features, user attribute information, scene information of the target user's current location, and content information of the second recommended content, a density preference value for the target user's first content type to which the second recommended content belongs is calculated. The second recommended content is at least one of the multiple first recommended content items. The density preference value is used to characterize the density of the target user's preference for the content corresponding to the first content type in the recommended content set. Based on the density preference value, the recommended content in the recommended content set is adjusted. Thus, by calculating the density preference values ​​for different content types in the recommended content set, and dynamically adjusting the recommended content according to the density preference values, when a certain content type in the recommended content set is highly recommended and the user's density preference value is low, the content of that content type is adjusted, improving the efficiency of recommended content and the user experience. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating an information recommendation method provided in an embodiment of this application;

[0014] Figure 2 This is a schematic diagram of an information recommendation method provided in an embodiment of this application;

[0015] Figure 3 This is a schematic diagram of the structure of an information recommendation device provided in an embodiment of this application;

[0016] Figure 4 This is one of the hardware structure diagrams of an electronic device provided in the embodiments of this application;

[0017] Figure 5 This is a second schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0019] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0020] The information recommendation method, apparatus, electronic device, and readable storage medium provided in this application will be described in detail below with reference to the accompanying drawings and through specific embodiments and application scenarios.

[0021] In news feed recommendation scenarios, electronic device systems adjust their recommendation strategies based on user behavior to understand user interests in a timely manner. These systems remember content clicked by users over a past period and continuously recommend related content to satisfy those interests. While content relevance reflects user interests to some extent, recommending content related to historical interests can meet current needs. However, excessive recommendation of related content can lead to user fatigue; even users who have long enjoyed the content are unlikely to continuously browse it extensively. The dense recommendation phenomenon in news feeds significantly impacts the system's recommendation effectiveness and user experience.

[0022] In existing technologies, the common approach to addressing the issue of dense recommendations in news feeds is to break up the content displayed in consecutive n views (i.e., within a single or multiple refreshes). For example, a single refresh should not contain content of the same category or with the same tags, thus avoiding the dense recommendation of too much similar content to the user. However, this method has drawbacks: it merely breaks up dense content and cannot dynamically adjust the density based on user behavior; furthermore, it lacks personalization, failing to differentiate between users' varying perceptions of density when using the same template (some users prefer dense content, while others prefer diversity); and moreover, users' density preferences differ across time and environment, and commonly used templates cannot be adjusted accordingly.

[0023] In the information recommendation method provided in this application embodiment, firstly, user profile feature information is meticulously constructed based on user behavior characteristics. The user's density preference for relevant content is calculated based on this user profile feature information, serving as the fundamental basis for dynamically adjusting the density of relevant content. Secondly, this application embodiment introduces user attribute features to more personalized calculate the density preference of different users for different content, solving the problem of templates lacking personalization. Furthermore, this application embodiment also introduces scenario feature information to distinguish the density preference of the same user in different times and environments, improving the user experience and solving the problem that templates cannot be adjusted with time and environment. Finally, all feature information is input into a density preference model, which calculates the user's density preference value for a certain content type. By suppressing content with low user density preference values ​​while increasing the exposure opportunities of other types of content, new user interests are explored, and the diversity of recommendations is improved.

[0024] This application provides an information recommendation method. Figure 1 A flowchart illustrating an information recommendation method provided in an embodiment of this application is shown, the method being executed by an electronic device. Figure 1 As shown, the information recommendation method provided in this application embodiment may include the following steps 201 to 204.

[0025] Step 201: Obtain the recommended content set.

[0026] In this embodiment of the application, the above-mentioned set of recommended content includes multiple first recommended content items.

[0027] In this embodiment of the application, the aforementioned recommended content set may be a set of product content recommended by e-commerce applications in electronic devices, or a set of news content recommended by leisure and entertainment applications.

[0028] In the embodiments of this application, the aforementioned multiple first recommended contents can be contents of different content types or contents of the same content type.

[0029] Step 202: Based on the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommended content, calculate the target user's dense preference value for the first content type to which the second recommended content in the recommended content set belongs.

[0030] In this embodiment of the application, the second recommended content is at least one of a plurality of first recommended content in the recommended content set.

[0031] In this embodiment of the application, the aforementioned density preference value is used to characterize the density of the content corresponding to the first content type to which the second recommended content in the recommended content set belongs to the target user in the recommended content set.

[0032] In this embodiment of the application, the aforementioned profile feature information of the target user can indicate the target user's preference for different content types.

[0033] In this embodiment of the application, the user attribute information of the target user may include the target user's International Mobile Equipment Identity (IMEI), the target user's age group, the target user's gender, and whether the target user is married.

[0034] In this embodiment of the application, the aforementioned user attributes may be attribute information uploaded by the user and obtained by the electronic device after authorization by the user.

[0035] For example, as shown in Table 1, the user attribute information of the target user can be represented by continuous numerical values ​​or discrete codes.

[0036]

[0037] Table 1

[0038] For example, by adding user attribute features, this application enables the dense preference model to distinguish different users based on these user attribute features, thereby overcoming the shortcomings of prediction using only user profile features. Furthermore, for cases where different users have the same user profile feature values, the calculation of dense preference values ​​for different users should be differentiated. For instance, in addition to the parameters of user profile features in the calculation formula of the dense preference model, this application also introduces parameters of user attribute features. This ensures that when different users have the same profile features, different user attribute features will yield different results. After inputting data of target users with the same profile features but different user attribute features into the dense preference model, the output prediction results show that the dense preference model can output different results for users with different user attribute features even when the profile features are the same. Therefore, adding user attribute features enables the dense preference model to distinguish the natural differences between users when predicting dense preferences, making the results more personalized.

[0039] In this embodiment of the application, the scene information of the target user's current location may include: current time, current location, whether it is a holiday, and network status, etc.

[0040] For example, as shown in Table 2, the scene information of the target user's current scene can be represented by continuous numerical values ​​or discrete codes.

[0041]

[0042] Table 2

[0043] In this embodiment of the application, the content information of the second recommended content may include the content title, content category, and content tags of the second recommended content.

[0044] For example, the title of the second recommended content could be "A commercial and residential land parcel in xx town of xx city is listed for sale with a starting price of xx billion yuan", the content category of the second recommended content could be "real estate - real estate market dynamics", and the content tag of the second recommended content could be "xx city".

[0045] For example, the same user may perceive intensive recommendations differently at different times. For instance, a user might prefer the system to push more related content during some time periods, while preferring a wider variety of content during others. Thus, introducing contextual information allows the model to learn the user's intensive preferences across different times and environments.

[0046] For example, the electronic device acquires the profile feature information of the target user, the user attribute information of the target user, the scene information of the current scene of the target user, and the content information of the second recommended content, inputs them into the dense preference model, and calculates the dense preference value of the target user for the first content type to which the second recommended content in the recommended content set belongs.

[0047] Optionally, in this embodiment of the application, before step 202, "calculating the dense preference value of the target user for the first content type to which the second recommended content in the recommended content set belongs, based on the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommended content," the information recommendation method provided in this embodiment of the application further includes the following step 301:

[0048] Step 301: Based on the behavioral characteristic information of the target user, determine the profile characteristic information of the target user.

[0049] For example, the behavioral characteristic information of the target user mentioned above includes: the user's behavior information on the displayed content and the content information of the displayed content.

[0050] For example, the user's behavior information regarding the displayed content may include: swiping away, clicking.

[0051] For example, the content information of the content shown above includes: content title, content category, and content tags.

[0052] For example, the user's behavioral characteristics information mentioned above can be offline data or real-time data. Offline data is delayed by several hours compared to the time when the user's behavior occurs, while real-time data is delayed by several seconds.

[0053]

[0054]

[0055] Table 3

[0056] For example, as shown in Table 3, at the current moment, the electronic device recommends and displays 3 pieces of content to the user: content categorized as "Sports_Basketball" is displayed but not clicked (i.e. swiped away) once, content categorized as "Real Estate_Housing Market Dynamics" is displayed but not clicked once, and content categorized as "Education_Industry" is displayed but not clicked once.

[0057] For example, the profile feature information of the target user mentioned above includes: first preference feature information and second preference feature information.

[0058] For example, the aforementioned first preference feature information is used to characterize the degree of intense preference of the target user for content corresponding to different content types.

[0059] For example, the aforementioned second preference feature information is used to characterize the degree of intensive neglect of content corresponding to different content types by the target user.

[0060] For example, based on the behavioral characteristics of the target user, a DNN model can be used to determine the profile characteristics of the target user.

[0061] In one example, in the information recommendation method provided in this application embodiment, the method of "determining the profile feature information of the target user based on the behavioral feature information of the target user" in step 301 includes the following steps 301a to 303d:

[0062] Step 301a: Determine the category profile features based on the target user's behavioral characteristics and the content category corresponding to the first content type in the recommended content set.

[0063] For example, the above-mentioned classification profile features include first classification preference feature information and second classification preference feature information.

[0064] For example, the first category preference feature information is used to characterize the target user's preference for content corresponding to different content categories.

[0065] For example, the second category preference feature information is used to characterize the degree to which the target user ignores the content corresponding to different content categories.

[0066] It is understandable that the aforementioned first category preference feature information and second category preference feature information are obtained based on the content classification information corresponding to the first content type.

[0067] For example, the first formula is used to calculate the first classification preference feature information of the content category corresponding to the first content type in the recommended content set.

[0068] For example, the first formula is:

[0069] Here, "click" represents the number of times a user clicks on content within that content category;

[0070] Show indicates the number of times a user swipes through content in this category;

[0071] W indicates that a preset number of swipes can offset one click.

[0072] For example, the second formula is used to calculate the second classification preference feature information of the content category corresponding to the first content type in the recommended content set.

[0073] For example, the second formula is:

[0074] Where k represents the total number of content items displayed;

[0075] cn i This represents the number of the most recently displayed (i) pieces of content that belong to this content category and have not been clicked by the user. For example, if 10 pieces of content were recently displayed, and 3 of them belong to this content category and have not been clicked by the user, then cn represents the number of such pieces. 10 =3.

[0076] It's important to note that when electronic devices intensively recommend content from a specific category, user reactions can be categorized into three types: 1. If the number of clicks is high, it indicates positive feedback from the user towards the intensive recommendations, meaning the first category preference feature value is relatively high; 2. If the number of clicks is far less than the number of swipes, the first category preference feature value is relatively low, indicating that the user is still interested in the content, but the excessive recommendation of this type of content by the electronic device will crowd out the exposure of other content, having a negative impact; 3. If the user cannot tolerate the overly intensive recommendations and continuously swipes past the content, the second category preference feature value is relatively high, indicating that the user's interest in this content category is rapidly decreasing or even intolerable at this moment. Continuing to push related content to the user will lead to user aversion. Therefore, category profile feature information can reflect the user's current psychological state in a timely manner based on user behavior.

[0077] Step 302b: Determine tag profile features based on the behavioral characteristics of the target user and the content tags corresponding to the first content type in the recommended content set.

[0078] For example, the aforementioned tag profile features include first tag preference feature information and second tag preference feature information.

[0079] For example, the first tag preference feature information is used to characterize the target user's preference for content corresponding to different content tags.

[0080] For example, the second tag preference feature information is used to characterize the degree to which the target user ignores the content corresponding to different content tags.

[0081] It is understandable that the aforementioned first tag preference feature information and second tag preference feature information are obtained based on the content tag information corresponding to the first content type.

[0082] For example, the first formula is used to calculate the first tag preference feature information of the content tags corresponding to the first content type in the recommended content set.

[0083] For example, the second formula is used to calculate the second tag preference feature information of the content tags corresponding to the first content type in the recommended content set.

[0084] It's important to note that content categorization is generally a coarse-grained attribute of content, offering limited ability to characterize it. To accurately depict user preferences at a finer granular level, tag-based profile features are constructed based on the target user's behavioral characteristics. These are calculated using the first and second formulas, the difference being that the content categorization information of the displayed content is replaced with content tags. Compared to categorized profile features, tag-based profile features can more precisely characterize user interests and preferences. However, content tags also present semantic conflicts. For example, the tag "Russia" for food content and the tag "Russia" for lifestyle content, while both related to Russia, contain vastly different content.

[0085] Step 302c: Based on the behavioral characteristics of the target user and the content category and content tag corresponding to the first content type in the recommended content set, determine the joint tag profile features.

[0086] For example, the content category and content tag in the content information of the above-displayed content are concatenated to obtain a joint tag, as shown in Table 4. The content category "Real Estate_Real Estate Market Dynamics" is concatenated with the content tag "xx City" to obtain the joint tag "Real Estate_Real Estate Market Dynamics_xx City".

[0087]

[0088] Table 4

[0089] For example, the joint label profile features include first joint preference feature information and second joint preference feature information.

[0090] For example, the first joint preference feature information is used to characterize the target user's preference for content corresponding to different joint tags.

[0091] For example, the second joint preference feature information is used to characterize the degree to which the target user ignores the content corresponding to different joint tags.

[0092] It is understandable that the first and second joint preference feature information mentioned above are obtained based on the content classification information and content tag information corresponding to the first content type.

[0093] For example, the first formula is used to calculate the first joint preference feature information of the joint tag corresponding to the first content type in the above-mentioned recommended content set.

[0094] For example, the second formula is used to calculate the second joint preference feature information of the joint tags corresponding to the first content type in the above-mentioned recommended content set.

[0095] It should be noted that, to address the issues of overly coarse granularity in category profile features and semantic conflicts in tag profile features, a joint tag is constructed based on the user behavior characteristics of the target user. By concatenating content categories and content tags, a more meaningful joint tag is generated. Referring to Table 3 and as shown in Table 4, from a category perspective, a user swipes over "Real Estate_Real Estate Market Dynamics" once, but because the category granularity is too coarse, it cannot be determined that the user is uninterested in all real estate-related content. From a tag perspective, a user swipes over "XX City" once, but because the tag granularity is too fine, it cannot be determined that the user is uninterested in content about "XX City". Therefore, by constructing a joint tag, we can clearly see that the user's interest in the real estate market of XX City has weakened after swiping over "Real Estate_Real Estate Market Dynamics_XX City" once. Finally, based on the joint tag data, the first and second formulas are used to construct the joint tag profile features.

[0096] Step 303e: Merge the first classification preference feature, the first label preference feature, and the first joint preference feature into first preference feature information; and merge the second classification preference feature, the second label preference feature, and the second joint preference feature into second preference feature information.

[0097] In this way, we can more accurately obtain users' preferences for different content in densely recommended content and create more accurate user profile characteristics.

[0098] Step 203: Adjust the recommended content in the recommended content set based on the target user's dense preference value for the first content type to which the second recommended content in the recommended content set belongs.

[0099] Optionally, in this embodiment of the application, the process of step 203, "adjusting the recommended content in the recommended content set based on the target user's dense preference value for the first content type to which the second recommended content in the recommended content set belongs," includes the following steps 203a and 203b:

[0100] Step 203a: If the above-mentioned dense preference value is less than a predetermined threshold, then the content belonging to the first content type in the above-mentioned recommended content set is deleted or downgraded, and content corresponding to the second content type, which is different from the first content type, is added to the recommended content set.

[0101] For example, the aforementioned predetermined threshold may be system-defined or user-set.

[0102] For example, based on the aforementioned density preference value, content of the first content type in the first recommended content is deleted with a probability p = 1 - density preference value. That is, the lower the density preference value, the greater the probability of deletion.

[0103] For example, when content with low content type density preference values ​​in the first recommended content is deleted, the resulting gap creates an opportunity for exposure. This gap is then filled with other types of content to further explore new user interests and improve the diversity of recommendations. For instance, after content related to "Entertainment_Humor" is deleted, new content such as "Society_Sudden Disasters" is used to fill the original position, thus exploring new user interests.

[0104] Step 203b: If the above-mentioned dense preference value is greater than the predetermined threshold, then retain the content belonging to the first content type in the recommended content set.

[0105] For example, the content of the first content type in the recommended content set is retained based on the aforementioned dense preference value.

[0106] For example, the adjusted set of recommended content is recommended to the target user.

[0107] In the information recommendation method provided in this application embodiment, a recommendation content set is obtained, which includes multiple first recommendation contents. Based on the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommendation contents, a density preference value for the target user's first content type to which the second recommendation contents belong is calculated. The second recommendation contents are at least one of the multiple first recommendation contents. The density preference value is used to characterize the density of the target user's preference for the content corresponding to the first content type in the recommendation content set. Based on the density preference value, the recommendation contents in the recommendation content set are adjusted. Thus, by calculating the density preference values ​​of different content types in the recommendation content set, the recommendation contents in the recommendation content set are dynamically adjusted according to the density preference values. This allows for adjustments to the content of a certain content type when it is highly recommended and the user's density preference value is low, thereby improving the efficiency of the recommendation content and the user experience.

[0108] Optionally, in this embodiment of the application, the process of step 202, "calculating the dense preference value of the target user for the first content type to which the second recommended content in the recommended content set belongs, based on the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommended content," includes the following steps 202a to 202e:

[0109] Step 202a: Input the target user's profile feature information, the target user's user attribute information, the target user's current scene information, and the content information of the second recommended content into the dense preference model.

[0110] For example, such as Figure 2 As shown, the dense preference model includes an embedding layer, a merging layer, a first fully connected layer, and a second fully connected layer. The merging layer contains a merging function (Concat function), and the second fully connected layer contains an activation function (simoid function).

[0111] Step 202b: Using the embedding layer in the above dense preference model, extract the user attribute feature information corresponding to the user attribute information of the target user, the scene feature information corresponding to the scene information, and the content feature information corresponding to the content information of the second recommended content.

[0112] For example, after inputting the user attribute information of the target user into the embedding layer of the dense preference model, it is converted into a feature vector that can represent the user attribute information of the target user, and the user attribute feature information of the target user is output.

[0113] For example, after the above scene information is input into the Embedding layer in the above dense preference model, it is converted into a feature vector that can represent the scene information, and the scene feature information is output.

[0114] For example, after inputting the content information of the second recommended content into the Embedding layer of the dense preference model, it is converted into a feature vector that can represent the content information of the second recommended content, and the content feature information is output.

[0115] Step 202c: Using the merging layer in the above dense preference model, the target user's profile feature information, the target user's user attribute feature information, the scene feature information of the target user's current scene, and the content feature information corresponding to the content information of the second recommended content are concatenated to obtain the target feature information.

[0116] For example, the profile feature information of the target user, the user attribute feature information of the target user, the scene feature information of the current scene of the target user, and the content feature information corresponding to the content information of the second recommended content are input into the merging layer (Concat layer). The Concat function is used to concatenate the various feature information and output the target feature information.

[0117] Step 202d: Using the first fully connected layer in the above dense preference model, the target feature information is transformed nonlinearly to obtain the transformed target feature information.

[0118] For example, after the above target feature information is input into the first fully connected layer, it undergoes nonlinear transformation to find the pattern of the target feature information and outputs more accurate transformed target feature information.

[0119] Step 202e: Using the second fully connected layer in the above dense preference model, the above dense preference value is calculated based on the transformed target feature information.

[0120] For example, the transformed target feature information is input into the second fully connected layer, and an activation function (simoid function) is used to convert the feature vector in the transformed target feature information into a numerical score of the final dense preference value.

[0121] In one example, in step 202c above, "using the merging layer in the dense preference model above, the target user's profile feature information, the target user's user attribute feature information, the scene feature information of the target user's current scene, and the content feature information corresponding to the content information of the second recommended content are concatenated to obtain the target feature information," the following step 202c1 is included:

[0122] Step 202c1: Using the merging layer in the above-mentioned dense preference model, the first preference feature information, the user attribute feature information of the target user, the scene feature information of the current scene of the target user, and the content feature information corresponding to the content information of the second recommended content are concatenated to obtain the target feature information.

[0123] Further optionally, in step 202e above, "using the second fully connected layer in the above dense preference model, and calculating the dense preference value based on the above transformed target feature information," the following step 202e1 is included:

[0124] Step 202e1: Using the second fully connected layer in the above dense preference model, based on the transformed target feature information and the second preference feature information, calculate the dense preference value of the target user for the first content type to which the second recommended content in the recommended content set belongs.

[0125] It should be noted that the second preference feature information is a strong signal expressing the user's continuous swiping behavior (i.e., the degree of neglect of content corresponding to different content types). In order for the model to directly remember this behavioral signal, the second preference feature information is used as the input to the shallow network when constructing the model, so that the second preference feature information reaches the back end of the model directly and avoids signal loss due to the large number of neural network layers. Other features are input to the deep network, so that the model can fully fit the user's behavioral patterns.

[0126] In this way, the problem of dense recommendations can be solved dynamically and in a personalized manner, increasing the diversity of recommendations, exploring new user interests, and improving the system's recommendation effectiveness and user experience.

[0127] Optionally, in this embodiment, cross-entropy is selected as the loss function, and a dense preference model is trained based on the training samples. The output of the dense preference model is the dense preference score of the user for candidate content, which can range from 0.0 to 1.0. The lower the dense preference value, the lower the user's dense preference for this type of content.

[0128] For example, combining Tables 1, 2, 3, and 4, as shown in Table 5, we can consider the case where user "800000000000001" has very low interest in "Real Estate_Market Dynamics" or "xx City" (at this time, the first preference feature information is extremely low and the second preference feature information is relatively large), and did not click on content "001" (which is related to the real estate market in xx City). This data can be considered a negative sample (user clicks are positive samples, and swiping away is a negative sample).

[0129]

[0130]

[0131] Table 5

[0132] In this way, by using different training samples as input to the dense preference model and performing calculations, user behavior patterns can be learned.

[0133] It should be noted that the information recommendation method provided in this application embodiment can be executed by an information recommendation device, an electronic device, or a functional module or entity within an electronic device. This application embodiment uses an information recommendation device executing the information recommendation method as an example to illustrate the information recommendation device provided in this application embodiment.

[0134] Figure 3 A schematic diagram of a possible structure of the information recommendation device involved in an embodiment of this application is shown. For example... Figure 3 As shown, the information recommendation device 700 may include: an acquisition module 701 and a processing module 702; the acquisition module 701 is used to acquire a set of recommended content, which includes multiple first recommended content; the processing module 702 is used to calculate the density preference value of the target user for the first content type to which the second recommended content belongs, based on the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommended content, wherein the second recommended content is at least one of the multiple first recommended content, and the density preference value is used to characterize the density of the target user's preference for the content corresponding to the first content type in the recommended content set; the processing module 702 is also used to adjust the recommended content in the recommended content set based on the density preference value.

[0135] Optionally, in this embodiment, the processing module 702 is specifically configured to: input the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommended content into the dense preference model; then, using the embedding layer in the dense preference model, extract the user attribute feature information corresponding to the target user's user attribute parameters, the scene feature information corresponding to the scene information, and the content feature information corresponding to the content information of the second recommended content; use the merging layer in the dense preference model to concatenate the profile feature information, the user attribute feature information, the scene feature information, and the content feature information to obtain target feature information; use the first fully connected layer in the dense preference model to perform a nonlinear transformation on the target feature information to obtain transformed target feature information; and use the second fully connected layer in the dense preference model to calculate the dense preference value based on the transformed target feature information.

[0136] Optionally, in this embodiment of the application, the processing module 702 is further configured to determine the profile feature information of the target user based on the behavioral feature information of the target user; wherein the profile feature information of the target user includes: first preference feature information and second preference feature information; the first preference feature information is used to characterize the degree of dense preference of the target user for content corresponding to different content types; the second preference feature information is used to characterize the degree of dense neglect of the target user for content corresponding to different content types.

[0137] Optionally, in this embodiment of the application, the processing module 702 is specifically used to: use the merging layer in the dense preference model to concatenate the first preference feature information, the user attribute feature information, the scene feature information and the content feature information to obtain target feature information; use the second fully connected layer in the dense preference model to calculate the dense preference value of the target user for the first content type to which the second recommended content belongs, based on the transformed target feature information and the second preference feature information.

[0138] Optionally, in this embodiment of the application, the processing module 702 is specifically used to: if the dense preference value is less than a predetermined threshold, delete or downgrade the content belonging to the first content type in the recommended content set, and add content corresponding to a second content type different from the first content type to the recommended content set; if the dense preference value is greater than the predetermined threshold, retain the content belonging to the first content type in the recommended content set.

[0139] In the information recommendation device provided in this application embodiment, the device acquires a set of recommended content, which includes multiple first recommended contents. Based on the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommended contents, the device calculates the target user's density preference value for the first content type to which the second recommended contents belong. The second recommended contents are at least one of the multiple first recommended contents. The density preference value is used to characterize the density of the target user's preference for the content corresponding to the first content type in the recommended content set. Based on the density preference value, the recommended contents in the recommended content set are adjusted. Thus, by calculating the density preference values ​​of different content types in the recommended content set, and dynamically adjusting the recommended contents in the recommended content set according to the density preference values, when a certain content type in the recommended content set is highly recommended and the user's density preference value is low, the content of that content type is adjusted, thereby improving the efficiency of recommended content and the user experience.

[0140] The information recommendation device in this application embodiment can be an electronic device or a component within an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal or other devices besides a terminal. For example, the electronic device can be a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc. It can also be a server, network attached storage (NAS), personal computer (PC), television set (TV), ATM, or self-service machine, etc. This application embodiment does not specifically limit the device.

[0141] The information recommendation device in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit it.

[0142] The information recommendation device provided in this application embodiment can achieve... Figures 1 to 2 The various processes implemented in the method implementation examples will not be described again here to avoid repetition.

[0143] Optionally, such as Figure 4 As shown, this application embodiment also provides an electronic device 800, including a processor 801 and a memory 802. The memory 802 stores a program or instructions that can run on the processor 801. When the program or instructions are executed by the processor 801, they implement the various steps of the above-described information recommendation method embodiment and can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0144] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0145] Figure 5 A schematic diagram of the hardware structure of an electronic device to implement an embodiment of this application.

[0146] The electronic device 100 includes, but is not limited to, components such as: radio frequency unit 101, network module 102, audio output unit 103, input unit 104, sensor 105, display unit 106, user input unit 107, interface unit 108, memory 109, and processor 110.

[0147] Those skilled in the art will understand that the electronic device 100 may also include a power supply (such as a battery) for supplying power to various components. The power supply may be logically connected to the processor 110 through a power management system, thereby enabling functions such as managing charging, discharging, and power consumption through the power management system. Figure 5 The electronic device structure shown does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements, which will not be elaborated here.

[0148] The processor 110 is configured to acquire a set of recommended content, which includes multiple first recommended content items; the processor 110 is also configured to calculate a density preference value of the target user for a first content type to which the second recommended content belongs, based on the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommended content, wherein the second recommended content is at least one of the multiple first recommended content items, and the density preference value is used to characterize the density of the target user's preference for the content corresponding to the first content type in the recommended content set; the processor 110 is also configured to adjust the recommended content in the recommended content set based on the density preference value.

[0149] Optionally, in this embodiment, the processor 110 is specifically configured to: input the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommended content into a dense preference model; then, using the embedding layer in the dense preference model, extract the user attribute feature information corresponding to the target user's user attribute parameters, the scene feature information corresponding to the scene information, and the content feature information corresponding to the content information of the second recommended content; use the merging layer in the dense preference model to concatenate the profile feature information, the user attribute feature information, the scene feature information, and the content feature information to obtain target feature information; use the first fully connected layer in the dense preference model to perform a nonlinear transformation on the target feature information to obtain transformed target feature information; and use the second fully connected layer in the dense preference model to calculate the dense preference value based on the transformed target feature information.

[0150] Optionally, in this embodiment of the application, the processor 110 is further configured to determine the profile feature information of the target user based on the behavioral feature information of the target user; wherein the profile feature information of the target user includes: first preference feature information and second preference feature information; the first preference feature information is used to characterize the degree of dense preference of the target user for content corresponding to different content types; the second preference feature information is used to characterize the degree of dense neglect of the target user for content corresponding to different content types.

[0151] Optionally, in this embodiment of the application, the processor 110 is specifically used to: use the merging layer in the dense preference model to concatenate the first preference feature information, the user attribute feature information, the scene feature information, and the content feature information to obtain target feature information; and use the second fully connected layer in the dense preference model to calculate the dense preference value of the target user for the first content type to which the second recommended content belongs, based on the transformed target feature information and the second preference feature information.

[0152] Optionally, in this embodiment of the application, the processor 110 is specifically configured to: if the dense preference value is less than a predetermined threshold, delete or downgrade the content belonging to the first content type in the recommended content set, and add content corresponding to a second content type different from the first content type to the recommended content set; if the dense preference value is greater than the predetermined threshold, retain the content belonging to the first content type in the recommended content set.

[0153] In the electronic device provided in this application embodiment, the device acquires a set of recommended content, which includes multiple first recommended content items. Based on the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommended content, the device calculates the target user's density preference value for the first content type to which the second recommended content belongs. The second recommended content is at least one of the multiple first recommended content items. The density preference value is used to characterize the density of the target user's preference for the content corresponding to the first content type in the recommended content set. Based on the density preference value, the recommended content in the recommended content set is adjusted. Thus, by calculating the density preference values ​​of different content types in the recommended content set, the recommended content in the recommended content set is dynamically adjusted according to the density preference values. This allows for adjustments to the content of a certain content type when it is highly recommended and the user's density preference value is low, thereby improving the efficiency of recommended content and the user experience.

[0154] It should be understood that, in this embodiment, the input unit 104 may include a graphics processing unit (GPU) 1041 and a microphone 1042. The GPU 1041 processes image data of still images or videos obtained by an image capture device (such as a camera) in video capture mode or image capture mode. The display unit 106 may include a display panel 1061, which may be configured in the form of a liquid crystal display, an organic light-emitting diode, or the like. The user input unit 107 includes at least one of a touch panel 1071 and other input devices 1072. The touch panel 1071 is also called a touch screen. The touch panel 1071 may include a touch detection device and a touch controller. Other input devices 1072 may include, but are not limited to, a physical keyboard, function keys (such as volume control buttons, power buttons, etc.), a trackball, a mouse, and a joystick, which will not be described in detail here.

[0155] The memory 109 can be used to store software programs and various data. The memory 109 may primarily include a first storage area for storing programs or instructions and a second storage area for storing data. The first storage area may store the operating system, application programs or instructions required for at least one function (such as sound playback, image playback, etc.). Furthermore, the memory 109 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DRRAM). The memory 109 in the embodiments of this application includes, but is not limited to, these and any other suitable types of memory.

[0156] Processor 110 may include one or more processing units; optionally, processor 110 integrates an application processor and a modem processor, wherein the application processor mainly handles operations involving the operating system, user interface, and applications, and the modem processor mainly handles wireless communication signals, such as a baseband processor. It is understood that the aforementioned modem processor may also not be integrated into processor 110.

[0157] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described information recommendation method embodiments and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0158] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0159] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described information recommendation method embodiment and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0160] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0161] This application provides a computer program product, which is stored in a storage medium and executed by at least one processor to implement the various processes of the above-described information recommendation method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0162] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0163] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0164] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. An information recommendation method, characterized in that, The method includes: Obtain a set of recommended content, which contains multiple first recommended content items; Based on the target user's profile features, the target user's user attributes, the scene information of the target user's current location, and the content information of the second recommended content, the density preference value of the target user for the first content type to which the second recommended content belongs is calculated. The second recommended content is at least one of the plurality of first recommended content. The density preference value is used to characterize the density of the target user's preference for the content corresponding to the first content type in the recommended content set. Based on the density preference value, adjust the recommended content in the recommended content set; The calculation of the dense preference value of the target user for the first content type to which the second recommended content belongs, based on the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommended content, includes: After inputting the target user's profile features, user attribute information, scene information of the target user's current location, and content information of the second recommended content into the dense preference model, Using the embedding layer in the dense preference model, user attribute feature information corresponding to the user attribute parameters of the target user, scene feature information corresponding to the scene information, and content feature information corresponding to the content information of the second recommended content are extracted. By employing the merging layer in the dense preference model, the profile feature information, the user attribute feature information, the scene feature information, and the content feature information are concatenated to obtain the target feature information; By using the first fully connected layer in the dense preference model, the target feature information is transformed nonlinearly to obtain the transformed target feature information; The dense preference value is calculated using the second fully connected layer in the dense preference model, based on the transformed target feature information. The step of adjusting the recommended content in the recommended content set based on the dense preference value includes: If the dense preference value is less than a predetermined threshold, the content belonging to the first content type in the recommended content set will be deleted or downgraded, and content corresponding to a second content type that is different from the first content type will be added to the recommended content set. If the density preference value is greater than the predetermined threshold, then the content belonging to the first content type in the recommended content set is retained.

2. The method according to claim 1, characterized in that, Before calculating the dense preference value of the target user for the first content type to which the second recommended content belongs, based on the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommended content, the method further includes: Based on the behavioral characteristic information of the target user, the profile characteristic information of the target user is determined; The profile feature information of the target user includes: first preference feature information and second preference feature information; The first preference feature information is used to characterize the degree of intense preference of the target user for content corresponding to different content types; The second preference feature information is used to characterize the degree to which the target user intensively ignores content corresponding to different content types.

3. The method according to claim 2, characterized in that, The method employs a merging layer in the dense preference model to concatenate the profile feature information, user attribute feature information, scene feature information, and content feature information to obtain target feature information, including: By employing the merging layer in the dense preference model, the first preference feature information, the user attribute feature information, the scene feature information, and the content feature information are concatenated to obtain the target feature information; The step of employing the second fully connected layer in the dense preference model to calculate the dense preference value based on the transformed target feature information includes: Using the second fully connected layer in the dense preference model, based on the transformed target feature information and the second preference feature information, the dense preference value of the target user for the first content type to which the second recommended content belongs is calculated.

4. An information recommendation device, characterized in that, The device includes: an acquisition module and a processing module; The acquisition module is used to acquire a set of recommended content, which includes multiple first recommended content items; The processing module is used to calculate the density preference value of the target user for the first content type to which the second recommended content belongs, based on the target user's profile feature information, the target user's user attribute information, the scene information of the target user's current scene, and the content information of the second recommended content. The second recommended content is at least one of the plurality of first recommended content. The density preference value is used to characterize the density of the target user's preference for the content corresponding to the first content type in the recommended content set. The processing module is also used to adjust the recommended content in the recommended content set based on the dense preference value; The processing module is specifically used for: After inputting the target user's profile features, user attribute information, scene information of the target user's current location, and content information of the second recommended content into the dense preference model, Using the embedding layer in the dense preference model, user attribute feature information corresponding to the user attribute parameters of the target user, scene feature information corresponding to the scene information, and content feature information corresponding to the content information of the second recommended content are extracted. By employing the merging layer in the dense preference model, the profile feature information, the user attribute feature information, the scene feature information, and the content feature information are concatenated to obtain the target feature information; By using the first fully connected layer in the dense preference model, the target feature information is transformed nonlinearly to obtain the transformed target feature information; The dense preference value is calculated using the second fully connected layer in the dense preference model, based on the transformed target feature information. The processing module is specifically used for: If the dense preference value is less than a predetermined threshold, the content belonging to the first content type in the recommended content set will be deleted or downgraded, and content corresponding to a second content type that is different from the first content type will be added to the recommended content set. If the density preference value is greater than the predetermined threshold, then the content belonging to the first content type in the recommended content set is retained.

5. The apparatus according to claim 4, characterized in that, The processing module is further configured to determine the profile feature information of the target user based on the behavioral feature information of the target user; The profile feature information of the target user includes: first preference feature information and second preference feature information; The first preference feature information is used to characterize the degree of intense preference of the target user for content corresponding to different content types; The second preference feature information is used to characterize the degree to which the target user intensively ignores content corresponding to different content types.

6. The apparatus according to claim 5, characterized in that, include: The processing module is specifically used for: By employing the merging layer in the dense preference model, the first preference feature information, the user attribute feature information, the scene feature information, and the content feature information are concatenated to obtain the target feature information; Using the second fully connected layer in the dense preference model, based on the transformed target feature information and the second preference feature information, the dense preference value of the target user for the first content type to which the second recommended content belongs is calculated.

7. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the information recommendation method as described in any one of claims 1 to 3.

8. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the information recommendation method as described in any one of claims 1 to 3.

Images