Content pushing method based on artificial intelligence

Abstract

The invention relates to the technical field of big data analysis, in particular to a content pushing method based on artificial intelligence, and the method comprises the steps: S1, constructing an alternative content set for a to-be-pushed object; s2, screening from the alternative content set to obtain a plurality of sample contents; s3, sequentially pushing the sample contents to the to-be-pushed object so as to collect a browsing behavior of the to-be-pushed object; and S3, updating the user portrait of the to-be-pushed object according to the browsing behavior. In order to solve the problem of inaccurate pushing in an initial state in a user portrait pushing process in the prior art, in the embodiment, for a to-be-pushed object, firstly, an alternative sample set with a large amount of content is constructed, and sample content is obtained through screening to serve as test data to be provided for a user; browsing behaviors of the user for different types of videos are collected, and the user portrait of the to-be-pushed object is updated according to the browsing behaviors, so that rapid establishment of the initial portrait is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of big data analysis, and in particular to a content push method based on artificial intelligence. Background Art

[0002] Content push refers to the act of selecting and pushing existing media resources to specific targets on a media platform based on information such as user portraits and recommendation algorithms. Through appropriate content push algorithms, the user's access experience and platform interaction can be effectively improved, thereby bringing about certain commercial conversions.

[0003] In the prior art, there are technical solutions for pushing content to users. For example, Chinese patent CN202010195145.7 discloses a content push method and device. The content push method includes: obtaining a content feature vector corresponding to the target push content, a device feature vector corresponding to the client of the target push content, and a user feature vector of the user currently logged in to the client; inputting the content feature vector, the device feature vector, and the user feature vector into a push mode prediction model; using the push mode prediction model to process the content feature vector, the device feature vector, and the user feature vector in multiple fully connected networks within the model to obtain a message push mode label; based on the message push mode corresponding to the message push mode label, the target push content is pushed to the client. The technical solution of the embodiment of the present application can improve the accuracy of the selected message push mode for pushing content.

[0004] However, during the actual implementation process, the inventors found that the user portrait targeted by this type of technical solution itself relies on the user's existing historical data to build the portrait label, and cannot be effectively pushed when the data is less. Summary of the invention

[0005] In view of the above problems existing in the prior art, a content push method based on artificial intelligence is now provided.

[0006] The specific technical solutions are as follows:

[0007] A content push method based on artificial intelligence, comprising:

[0008] Step S1: construct a set of candidate content for the object to be pushed;

[0009] Step S2: Filtering and obtaining a plurality of sample contents from the candidate content set;

[0010] Step S3: Pushing the sample content to the objects to be pushed in sequence to collect browsing behaviors of the objects to be pushed;

[0011] Step S3: updating the user portrait of the object to be pushed according to the browsing behavior;

[0012] The user portrait is used to match push content.

[0013] On the other hand, the step S1 comprises:

[0014] Step S11: Obtain the user portrait of the object to be pushed and the sample database of the sample content:

[0015] Step S12: extracting a portrait tag from the user portrait, and searching the sample database for the sample content according to the portrait tag;

[0016] Step S13: Generate the candidate content set according to the sample content.

[0017] On the other hand, the step S12 comprises:

[0018] Step S121: extracting the portrait tag from the user portrait, and searching the sample database for corresponding pre-screening content according to the portrait tag;

[0019] Step S122: extracting historical access data for the pre-screened content, and performing weighted calculation according to the historical access data to obtain a historical recommendation score;

[0020] Step S123: sorting the pre-screened contents according to the historical recommendation scores to obtain pre-recommended contents corresponding to the portrait tags;

[0021] Step S124: generating the candidate content set according to the pre-recommended content.

[0022] On the other hand, in step S122, the historical access data includes: page views, average page view duration, interaction volume, and user coverage ratio.

[0023] On the other hand, the step S2 comprises:

[0024] Step S21: constructing a sample content vector for the sample content;

[0025] Step S22: clustering the sample content vectors to obtain multiple content clusters;

[0026] Step S23: for each of the content clusters, the sample contents are screened and outputted respectively.

[0027] On the other hand, the step S23 comprises:

[0028] Step S231: for each content cluster, respectively calculate and obtain the cluster center;

[0029] Step S232: respectively calculating the cluster center distance of each sample content vector relative to the cluster center;

[0030] Step S233: Sort the sample content vectors according to the cluster center distances to filter a predetermined number of sample content outputs.

[0031] On the other hand, the step S3 comprises:

[0032] Step S31: arranging a push sequence for the sample contents according to the sample content tags, and marking the sample content tag groups;

[0033] Step S32: pushing the sample contents in sequence according to the pushing sequence, and collecting the browsing behavior;

[0034] Step S33: adjusting the sample content in the push sequence according to the browsing behavior, and then returning to step S32 until the push is completed.

[0035] On the other hand, the step S33 comprises:

[0036] When the object to be pushed is marked as not interested in the sample content, the sample content corresponding to the sample content tag group is deleted from the push sequence.

[0037] The above technical solution has the following advantages or beneficial effects:

[0038] In order to solve the problem of inaccurate user portrait push in the initial state in the prior art, in this embodiment, a candidate sample set with a large amount of content is first constructed for the object to be pushed, and the sample content is screened and provided to the user as test data. Then, the user's browsing behavior for different types of videos is collected, and the user portrait of the object to be pushed is updated according to the browsing behavior, thereby realizing the rapid establishment of the initial portrait. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The embodiments of the present invention will be described more fully with reference to the attached drawings, which are provided for illustration and description only and are not intended to limit the scope of the present invention.

[0040] Figure 1 is an overall schematic diagram of an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram of step S1 in an embodiment of the present invention;

[0042] Figure 3 This is a schematic diagram of step S12 in an embodiment of the present invention;

[0043] Figure 4This is a schematic diagram of step S2 in an embodiment of the present invention;

[0044] Figure 5 This is a schematic diagram of step S23 in an embodiment of the present invention;

[0045] Figure 6 Schematic diagram of step S3 in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0048] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0049] The present invention comprises:

[0050] A content push method based on artificial intelligence, such as Figure 1 As shown, including:

[0051] Step S1: construct a set of candidate content for the object to be pushed;

[0052] Step S2: Filtering a plurality of sample contents from the candidate content set;

[0053] Step S3: Pushing the sample content to the object to be pushed in sequence to collect the browsing behavior of the object to be pushed;

[0054] Step S4: updating the user portrait of the object to be pushed according to the browsing behavior;

[0055] User portraits are used to match push content.

[0056] Specifically, in order to address the problem of inaccurate push of user portraits in the initial state in the prior art, in this embodiment, a candidate sample set with a large amount of content is first constructed for the object to be pushed, and the sample content is screened and provided to the user as test data, and then the user's browsing behavior for different types of videos is collected, and the user portrait of the object to be pushed is updated according to the browsing behavior, thereby achieving the rapid establishment of the initial portrait.

[0057] In one embodiment, Figure 2As shown, step S1 includes:

[0058] Step S11: Obtain a sample database of user portraits and sample contents of the object to be pushed:

[0059] Step S12: extracting a portrait tag from the user portrait, and searching for sample content from a sample database according to the portrait tag;

[0060] Step S13: Generate a candidate content set according to the sample content.

[0061] Specifically, in order to achieve better recognition efficiency, in this embodiment, the user portrait in the initial state is first extracted for the object to be pushed. In this embodiment, when the user registers the corresponding account, the corresponding initial label is constructed through the user's attention information and registration related information, and then the user portrait is initially assigned according to the initial label, such as registration location, gender, interest area, etc. Then, in the sample database, search according to the initial value of the portrait label. Among them, the sample content refers to a large amount of media data already in the media platform, and each media data is assigned a corresponding type label when it is uploaded. By searching and comparing the portrait label with the type label, relatively matching sample content can be found, and then a set of alternative content outputs can be generated.

[0062] In one embodiment, Figure 3 As shown, step S12 includes:

[0063] Step S121: extracting a portrait tag from the user portrait, and searching the sample database for corresponding pre-screened content according to the portrait tag;

[0064] Step S122: extracting historical access data for the pre-screened content, and performing weighted calculation according to the historical access data to obtain a historical recommendation score;

[0065] Step S123: sorting the pre-screened content according to the historical recommendation scores to obtain sample content corresponding to the portrait tag;

[0066] Specifically, in order to achieve a better effect of generating a set of alternative content and facilitate updating of user portraits through rapid comparison, in this embodiment, after extracting the initial portrait tags from the user portraits, media content with corresponding tags are searched in the sample database according to the portrait tags as pre-screened content.

[0067] Subsequently, for the pre-screened content under each extracted tag, the corresponding historical access data is extracted, including the number of views, average browsing time, interaction volume, and user coverage ratio.

[0068] Among them, the number of views is a numerical value, and the average browsing time is converted into a ratio relative to the total browsing time of each media content through a normalization algorithm. For graphic content, the total browsing time can be estimated based on the amount of text and images. The amount of interaction is a numerical value, including user comments, likes, favorites, shares, etc., which can be designed as a weighted value according to a specific formula.

[0069] Then, a weighted calculation is performed according to the historical browsing data to obtain a historical recommendation score, which is used to filter out more popular media content. The pre-filtered content is then sorted according to the historical recommendation score to obtain the top sample content corresponding to the portrait tag for output. By filtering out more popular media content, the possibility of subsequent user interaction with the sample content is increased, thereby completing the update of the user portrait more quickly.

[0070] In one embodiment, Figure 4 As shown, step S2 includes:

[0071] Step S21: constructing a sample content vector for the sample content;

[0072] Step S22: clustering the sample content vectors to obtain multiple content clusters;

[0073] Step S23: For each content cluster, sample contents are screened and outputted respectively.

[0074] Specifically, in order to improve the updating efficiency of the user portrait, in this embodiment, the content of the sample content is selected for identification, so as to extract representative sample content for output, so as to collect more user behaviors.

[0075] Among them, a semantic recognition model was used to identify each sample content, and some semantic features were extracted. The sample content vector was constructed according to the semantic features and data of other dimensions, such as video length, classification and other information.

[0076] Then, clustering is performed according to the sample content vectors to obtain multiple content clusters, each of which represents a different type of content set.

[0077] The clustering process includes, for the sample content vectors, respectively calculating the vector distance of each sample content vector relative to other sample content vectors;

[0078] Screening is performed according to the vector distance and a pre-calibrated clustering threshold, thereby constructing a plurality of sample range groups corresponding to each sample content vector;

[0079] For each sample range group, the number of occurrences of the sample content vector is counted respectively, and the sample content vector with the largest number of occurrences is taken as the center point. Sample content vectors within a certain range from the center point are screened according to the second screening distance as the clustering result in the content cluster.

[0080] For each content cluster, representative sample contents are extracted and outputted respectively, thus achieving effective screening of representative sample contents.

[0081] In one embodiment, Figure 5 As shown, step S23 includes:

[0082] Step S231: for each content cluster, respectively calculate and obtain the cluster center;

[0083] Step S232: Calculate the cluster center distance of each sample content vector relative to the cluster center;

[0084] Step S233: sorting the sample content vectors according to the cluster center distance to select a predetermined number of sample content outputs.

[0085] Specifically, in order to extract sample content with more significant features, in this embodiment, after clustering is completed to obtain content clusters, sample content vectors are extracted for each content cluster, and then the mean of the sample content vectors is calculated to obtain the cluster center.

[0086] After obtaining the cluster center, the cluster center distance of each sample content vector relative to the cluster center is calculated respectively, and then the sample content vectors are sorted according to the cluster center distance, so as to determine the distance of each sample content vector relative to the cluster center.

[0087] On this basis, a predetermined number of sample content vectors closest to the cluster center are selected for output to obtain sample content with more significant features, thereby facilitating improving the user interaction rate.

[0088] In one embodiment, Figure 6 As shown, step S3 includes:

[0089] Step S31: arranging a push sequence for sample contents according to sample content tags, and marking sample content tag groups;

[0090] Step S32: pushing sample contents in sequence according to the push sequence and collecting browsing behaviors;

[0091] Step S33: adjust the sample content in the push sequence according to the browsing behavior, and then return to step S32 until the push is completed.

[0092] Specifically, in order to achieve a better push effect, in this embodiment, after the sample content is extracted, the sample content is firstly grouped according to the sample content tag tags.

[0093] This step may also be performed according to content clusters, that is, the sample contents extracted from the same content cluster are marked as the same sample content label group.

[0094] Then, in the push sequence, the sample contents are arranged in random order, and then pushed to the objects to be pushed one by one, and the user's browsing behavior is collected, such as effective browsing, likes, collections, sharing, and no interest, etc. Each behavior has a corresponding score, and the relevant score of each group is calculated by counting the browsing behaviors, and the number of sample contents of each group in the push sequence is adjusted in real time according to the score.

[0095] For example, in one embodiment, when the object to be pushed is marked as not interested in the sample content, the sample content corresponding to the sample content tag group is deleted from the push sequence.

[0096] Those skilled in the art will appreciate that various aspects of the present invention, or possible implementations of various aspects, may be specifically implemented as systems, methods, or computer program products. Therefore, various aspects of the present invention, or possible implementations of various aspects, may take the form of complete hardware embodiments, complete software embodiments (including firmware, resident software, etc.), or embodiments of combined software and hardware aspects, all collectively referred to herein as "circuits," "modules," or "systems." In addition, various aspects of the present invention, or possible implementations of various aspects, may take the form of computer program products, which refer to computer instructions stored in a memory.

[0097] The memory may be a computer-readable signal medium or a computer-readable storage medium. Computer-readable storage media include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or apparatuses, or any suitable combination of the foregoing, such as random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable read-only memory (CD-ROM).

[0098] The processor in the computer reads the computer instructions stored in the memory, so that the processor can execute the functional actions specified in each step or the combination of steps in the flowchart; and generate a device for implementing the functional actions specified in each block or the combination of blocks in the block diagram.

[0099] It should be understood that the processor in the computer can be understood as one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components implemented to execute the aforementioned computer instructions.

[0100] Computer instructions can be executed completely on the user's local computer, partially on the user's local computer, as a separate software package, partially on the user's local computer and partially on a remote computer, or completely on a remote computer or server. It should also be noted that in some alternative embodiments, the functions noted in each step in the flow chart or each block in the block diagram may not occur in the order noted in the figure. For example, depending on the functions involved, two steps or two blocks shown in succession may actually be executed roughly simultaneously, or these blocks may sometimes be executed in reverse order.

[0101] The above are only preferred embodiments of the present invention, and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the description and illustrations of the present invention should be included in the protection scope of the present invention.

Claims

1. A content push method based on artificial intelligence, characterized in that: include: Step S1: construct a set of candidate content for the object to be pushed; Step S2: Filtering and obtaining a plurality of sample contents from the candidate content set; Step S3: Pushing the sample content to the objects to be pushed in sequence to collect browsing behaviors of the objects to be pushed; Step S4: updating the user portrait of the object to be pushed according to the browsing behavior; The user portrait is used to match push content.

2. The content push method according to claim 1, characterized in that: The step S1 comprises: Step S11: Obtain the user portrait of the object to be pushed and the sample database of the sample content: Step S12: extracting a portrait tag from the user portrait, and searching the sample database for the sample content according to the portrait tag; Step S13: Generate the candidate content set according to the sample content.

3. The content push method according to claim 2, characterized in that: The step S12 comprises: Step S121: extracting the portrait tag from the user portrait, and searching the sample database for corresponding pre-screening content according to the portrait tag; Step S122: extracting historical access data for the pre-screened content, and performing weighted calculation according to the historical access data to obtain a historical recommendation score; Step S123: Sort the pre-screened content according to the historical recommendation scores to obtain the sample content corresponding to the portrait label.

4. The content push method according to claim 3, characterized in that: In step S122, the historical access data includes: page views, average page view duration, interaction volume, and user coverage ratio.

5. The content push method according to claim 1, characterized in that: The step S2 comprises: Step S21: constructing a sample content vector for the sample content; Step S22: clustering the sample content vectors to obtain multiple content clusters; Step S23: for each of the content clusters, the sample contents are screened and outputted respectively.

6. The content push method according to claim 5, characterized in that: The step S23 comprises: Step S231: for each content cluster, respectively calculate and obtain the cluster center; Step S232: respectively calculating the cluster center distance of each sample content vector relative to the cluster center; Step S233: Sort the sample content vectors according to the cluster center distances to filter a predetermined number of sample content outputs.

7. The content push method according to claim 1, characterized in that: The step S3 comprises: Step S31: arranging a push sequence for the sample contents according to the sample content tags, and marking the sample content tag groups; Step S32: pushing the sample contents in sequence according to the pushing sequence, and collecting the browsing behavior; Step S33: adjusting the sample content in the push sequence according to the browsing behavior, and then returning to step S32 until the push is completed.

8. The content push method according to claim 7, characterized in that: The step S33 comprises: When the object to be pushed is marked as not interested in the sample content, the sample content corresponding to the sample content tag group is deleted from the push sequence.

Images