Precise pushing method and system based on user portrait

Through the hidden Markov model and improved cosine similarity calculation, user portraits are generated and push content is optimized, the problem of inaccurate push in the existing technology is solved, and user viscosity and business traffic are improved.

CN118886980AActive Publication Date: 2024-11-01SHENZHEN SHUYAO TECH CO LTD

Patent Information

Application Number
CN202410885874.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-03
Publication Date
2024-11-01
Estimated Expiration
2044-07-03

AI Technical Summary

Technical Problem

The existing push methods based on user profiles only rely on the same category tags of historical data, and cannot fully capture user preferences, resulting in inaccurate push in the absence of the same category tags, resulting in reduced user stickiness and business impact.

Method used

The hidden Markov model is used to generate state transition probability and user preference labels based on behavioral data, and to calculate category tag similarity values ​​in combination with improved cosine similarity, generate user portraits with push time periods, and optimize push content by correcting state transition probability.

Benefits of technology

Effectively expand the scope of application and accuracy of push data, improve user viscosity and business traffic, ensure that push content conforms to user behavior, and reduce dependence on data category labels.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118886980A_ABST
    Figure CN118886980A_ABST
Patent Text Reader

Abstract

The invention provides a precise pushing method and system based on a user portrait, and relates to the technical field of data processing, and the method comprises the steps: obtaining a category label of each data content in data in a library; acquiring behavior data of the user on the data content corresponding to each category label; in combination with the behavior data and a hidden Markov model, generating a state transition probability and a user preference tag; generating a user portrait with a push time period; obtaining a category label of the current access data content of the user; calculating a similarity value between each category label and the category label of the data content currently accessed by the user; generating push data content of the next moment in combination with the user portrait and the similarity value; recording the staying duration of the user under the condition of pushing the data content; setting a correction factor to correct the state transition probability corresponding to the category label of the data content currently accessed by the user; and taking the corrected state transition probability as a new state transition probability, and repeatedly executing pushing. The pushing accuracy is improved, and the user stickiness is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a user portrait-based precise push method and system. Background Art

[0002] A user portrait is a detailed description based on the user's behavioral data. It is used to characterize the user's characteristics and interests, enabling companies to understand and predict the user's needs and behaviors. By analyzing the collected data, user portraits help improve the accuracy and efficiency of personalized services.

[0003] In an era of information overload, users often cannot effectively filter out content that is truly valuable to them. Precision push can provide customized information and services by using user portraits to identify users' specific needs and preferences. For example, e-commerce platforms can recommend products based on users' purchase history and browsing habits to increase sales opportunities, while media services can push news or videos that users are interested in to increase user engagement and satisfaction. In addition, precision push also helps companies optimize advertising results by showing relevant ads to specific target groups, increasing conversion rates and reducing resource waste. In short, precision push can not only strengthen user stickiness and enhance user experience, but also bring economic benefits and brand loyalty to companies, and can greatly improve user experience and corporate efficiency.

[0004] However, although most of the existing data push methods also adopt a push method based on user portraits, they are often only recommendations based on the same category tags of historical data. To be precise, it is a retrieval-based push method that does not fully capture user preferences. When the same category tags do not appear in the data in the library, effective data push cannot be performed. This method is heavily dependent on the category tags of the data, and it is very easy to blindly push, resulting in reduced user stickiness and affecting business operations. Summary of the invention

[0005] In order to solve the technical problem that the existing technology only recommends the same category tags based on historical data, which is more precisely a retrieval-based push method that does not fully capture user preferences, and cannot perform effective data push when the same category tags do not appear in the data in the library, and this method is heavily dependent on the category tags of the data, and is prone to blind push, resulting in reduced user stickiness and affecting business operations, the present invention provides a precise push method and system based on user portraits.

[0006] The technical solution provided by the embodiment of the present invention is as follows:

[0007] First aspect

[0008] An embodiment of the present invention provides a user portrait-based accurate push method, including:

[0009] S1: Get the category labels of each data content in the database;

[0010] S2: Obtaining the user's behavior data on the data content corresponding to each category label, wherein the behavior data includes browsing data content category, browsing time, browsing interval, stay time and click frequency;

[0011] S3: Combining the behavior data and the hidden Markov model, generating a state transition probability and a user preference label, wherein the state transition probability is the probability that the user directly browses the data content of the second category label when browsing the data content of the first category label;

[0012] S4: Generate user profiles with push time periods based on behavior data and user preference tags;

[0013] S5: Obtain the category label of the data content currently accessed by the user;

[0014] S6: Calculate the similarity between each category label and the category label of the data content currently accessed by the user using the improved cosine similarity with element sparsity;

[0015] S7: Generate the push data content at the next moment based on the category label of the data content currently accessed by the user, combined with the user portrait and similarity value;

[0016] S8: Record the length of time the user stays under the pushed data content;

[0017] S9: When the stay time is less than the preset stay time, a correction factor for the difference between the stay time and the preset stay time is set to correct the state transition probability corresponding to the category label of the data content currently accessed by the user;

[0018] S10: Use the corrected state transition probability as the new state transition probability and return to step S5.

[0019] Second aspect

[0020] An embodiment of the present invention provides a user portrait-based accurate push system, including:

[0021] processor;

[0022] A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the precise push method based on user portrait as described in the first aspect is implemented.

[0023] The third aspect

[0024] An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the precise push method based on user portrait as described in the first aspect is implemented.

[0025] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0026] In the present invention, firstly, the category labels of each data content in the data library and the historical behavior data of the user are obtained, and then the hidden Markov model is introduced to perform data trend analysis on the user's historical behavior data based on the category labels, so as to capture the user's preference and state transition probability, and can fully mine the trend and probability of the user transferring to different categories of data under the current access data state based on the behavior data, and generate a user portrait with a push time period, and then combine the improved cosine similarity with element sparsity to calculate the similarity value between each category label and the category label of the data content currently accessed by the user, even if there is no category label that the user wants to access, it is still possible to push content that is extremely in line with the user's expectations based on the similarity value. This data-driven approach reduces the problem of heavy reliance on data category labels under data-driven, and can effectively expand the push data according to user behavior. While ensuring that the push is in line with user behavior, it greatly improves the scope of application of the push method, the push accuracy and the rationality of the push time, improves user stickiness, and increases business traffic. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0028] Figure 1 A flowchart of a user portrait-based accurate push method provided by an embodiment of the present invention;

[0029] Figure 2 A schematic diagram of the structure of a user portrait-based precision push system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0030] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0031] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0032] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0033] In the embodiments of the present invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are consistent.

[0034] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0035] Reference Manual Attached Figure 1 , which shows a flow chart of a precise push method based on user portrait provided in an embodiment of the present invention.

[0036] The embodiment of the present invention provides a method for accurate push based on user portrait, which can be implemented by an accurate push device based on user portrait, which can be a terminal or a server. The processing flow of the method for accurate push based on user portrait can include the following steps:

[0037] S1: Get the category labels of each data content in the database.

[0038] Among them, the data in the library, that is, the data in the database, is the accessible data content provided to customers, which may include product information, video information, audio information, text information, image information, etc. The multiple categories in the multi-category data content are actually equivalent to the category labels of the data content. For the data in the library, each data content has a corresponding category label synchronously. The category label can be marked by the data content provider or generated based on the existing algorithm. For the data content without category labels in the library data, the product information and text information can be directly obtained by natural language processing technology. Video data can use computer vision technology such as CNN to analyze video frames to obtain video data content. Audio data can use MFCCs to extract data content. Image information can be extracted by image recognition model. For non-text data content, the extracted data content is finally extracted by natural language processing technology. On the basis of obtaining the category label, the category label can be continuously classified by support vector machine or k-means clustering algorithm until the description label of each data content in the library data is different. In short, the greater the depth of classification, the higher the accuracy of subsequent push.

[0039] S2: Obtain the user's behavior data on the data content corresponding to each category label.

[0040] Among them, behavioral data includes browsing data content category, browsing time, browsing interval, stay duration and click frequency.

[0041] Among them, the browsing data content category refers to the type of content that users are exposed to during the browsing process, that is, the data content corresponding to the category label. Knowing which categories of content users tend to browse can help the system more accurately predict and recommend other content that users may be interested in. Browsing time refers to the specific time point when users conduct browsing activities, including date and time. This helps to analyze the user's activity pattern. For example, some users may be more inclined to browse content in the morning or evening. User behavior patterns may be different at different times, which is extremely important for optimizing push timing. Browsing interval refers to the time between two consecutive browsings of content with the same category label by a user. The browsing interval can reveal the stickiness of the content with the category label to the user. The dwell time refers to the time a user spends on a specific content in a single browsing, including looping videos or looping through photo collections. A longer dwell time usually indicates that the user has a higher interest or satisfaction in the content, and is an important indicator for evaluating content quality and user interest. The click frequency of the same content refers to the number of times a user clicks on a specific content of the same type. This indicator reflects the user's interest or interaction in this type of content.

[0042] It should be noted that behavioral data is the basis for building accurate user portraits, which can help understand users' interests and behavioral habits. Collecting these detailed data enables the system to achieve higher personalized recommendation accuracy, thereby improving user satisfaction and engagement, optimizing user experience, and increasing potential benefits. By analyzing these behavioral data, the system can identify users' likes and dislikes, and adjust the content recommendation algorithm to ensure that the pushed content is more in line with users' actual needs.

[0043] S3: Combine behavioral data and hidden Markov models to generate state transition probabilities and user preference labels.

[0044] The state transition probability is the probability that the user directly browses the data content of the second category label when browsing the data content of the first category label.

[0045] Among them, the hidden Markov model is a statistical model that assumes that the state of the system is "hidden". That is, the system state is not directly visible, but can be inferred through observable variables (in our scenario, these variables are user behavior data), and then describe a series of states that are not directly observable (hidden) in a system, and the change process of these states has Markov properties. In step S3, the hidden Markov model analyzes the user's behavior data, generates state transition probabilities and user preference labels, simulates the user's behavior pattern from one content category to another, and thus predicts the user's behavior trend, effectively processes time series data, captures the dynamic changes of user behavior, and helps the push system more accurately predict and adapt to user needs, and achieve more accurate content recommendations and personalized experience.

[0046] In a possible implementation, S3 specifically includes:

[0047] S301: Define the state element and observation element of the hidden Markov model, wherein the state element is the category label corresponding to the data content browsed by the user, and the observation element is the behavior data.

[0048] S302: Establish an observation sequence based on the behavior data.

[0049] S303: Initialize the algorithm parameters of the hidden Markov model according to the observation sequence, wherein the algorithm parameters include a state transition probability matrix, an observation probability matrix and an initial state probability:

[0050] The state transition probability matrix contains the probability of transitioning from one state to another. The observation probability matrix refers to the probability of observing a specific observation result given a specific state. The initial state distribution refers to the probability that the user may be in each state at the beginning.

[0051] The specific elements in the state transition probability matrix are:

[0052]

[0053] Among them, a ij represents the state transition probability from category label i to category label j, q t represents the state element at time t, i.e., the category label, T represents the total number of discrete moments in the observation sequence, P(q t =i,q t+1 =j|O,λ) represents the joint probability that a class is in class label i at time t and is transferred to class label j at time t+1, given the observation sequence O and the model parameter λ. t =i|O,λ) represents the probability of being in category label i at time t.

[0054] Among them, the category label actually refers to the hidden state.

[0055] The specific elements in the observation probability matrix are:

[0056]

[0057] Among them, b j (k) represents the observation probability of k when the class label is j, o t represents the observed value at time t in the observation sequence, P(q t =j,o t = k|λ) represents the joint probability of being in category label j and observing k at time t given the model parameter λ, P(q t =j|O,λ) means that the state element q at time t is given the observation sequence O and the model parameter λ. t is the overall occurrence probability of category label j, that is, the initial state probability.

[0058] Among them, observation probability is the core component of the hidden Markov model, which allows the model to interpret observed data based on assumptions about the current state. This is particularly important for behavioral data analysis because it can help the model accurately predict the external manifestations that may be observed under a specific internal state, and then be used for state inference and prediction of future behavior.

[0059] S304: Using the behavior data as input data of the hidden Markov model, training the hidden Markov model, and updating the state transition probability.

[0060] In a possible implementation, the update formula of the state transition probability is specifically:

[0061]

[0062] Among them, a' ij represents the updated state transition probability, α t (i) represents the forward probability at time t, that is, the probability of the class label being i in the observation sequence from the beginning of the sequence to time t, b j (o t+1 ) means that o is observed when the class label is j t+1 The observation probability, β t+1 (j) represents the backward probability at time t+1, that is, the probability of the category label being j at time t+1 in the observation sequence O from time t+1 to the end of the sequence. P(O|λ) represents the probability of occurrence of the observation sequence O under given model parameters.

[0063] It should be noted that in the process of updating the state transition probability in the hidden Markov model, the forward probability, backward probability, observation probability and the probability of the entire observation sequence are combined to re-estimate the transition probability from one state to another. The parameters of the model can be dynamically adjusted according to the actual observed data, so that the model can more accurately reflect the true pattern of user behavior, improve the accuracy of the prediction and the model's ability to adapt to new data. This ability to continuously learn and adapt based on actual data is the key to maintaining high efficiency of precise recommendation systems in complex and changing user behavior patterns.

[0064] S305: Given an observation sequence, a hidden state sequence based on the behavior data is generated through a trained hidden Markov model, wherein a state element of the hidden state sequence is a user preference label.

[0065] Specifically, first, the state elements (corresponding to the content category labels browsed by the user) and observation elements (user behavior data) of the model are defined in S301. Next, the observation sequence is constructed based on the behavior data in S302, which provides the model with the input of time series data. In S303, the core parameters of the model are initialized: the state transition probability matrix, the observation probability matrix, and the initial state probability. These parameters determine how the model predicts the transition between states and responds to the observed behavior. Through the training process of S304, the behavior data is used to adjust the state transition probability of the model to ensure that the model can accurately map the actual pattern of user behavior. Finally, in S305, the model generates a hidden state sequence, which represents the sequence of user preference labels inferred from the observed behavior data.

[0066] This process can deeply understand and mine the user's behavior dynamics and predict their future behavior trends. The use of hidden Markov models enables the system to effectively capture the conversion rules between user behaviors, even if these behaviors appear random on the surface. In addition, the dynamic adjustment of the model and the update of state transition probabilities ensure that the push system can adapt to changes in user behavior over time, maintain the accuracy and personalization of recommendations, and thus greatly enhance user experience and satisfaction.

[0067] S4: Generate user profiles with push time periods based on behavioral data and user preference tags.

[0068] It should be noted that using the user preference tags and behavior data generated in the previous steps to create a user profile with a push time period not only reflects the user's content preference, but also refines it to the optimal content push time, so that content push can be carried out during the time period when the user is most likely to be active. The advantage of this method is that it can significantly improve user interaction rate and satisfaction with content consumption, because the content push is highly consistent with the user's actual online time and preference for receiving information, thereby improving the effectiveness of the pushed content and the user's overall experience. By accurately grasping the push timing, it is also possible to optimize resource utilization efficiency, reduce invalid pushes, and improve the return on marketing activities.

[0069] In a possible implementation, S4 specifically includes:

[0070] S401: Calculate the user activity index in each time period based on the behavior data:

[0071]

[0072] Among them, A(h) represents the user activity index in the time period h of the browsing time, N h represents the number of user activities in time period h collected based on browsing intervals, d i represents the duration of the i-th user activity, D represents the maximum duration of all user activities, k i represents the click frequency in the i-th user activity, and log represents logarithmic operation.

[0073] The time period can be in hours.

[0074] It should be noted that the logarithmic transformation of click frequency can help smooth out extreme values, ensuring that a single high-interaction activity does not have too large an impact on the overall activity index, while maintaining an appropriate influence of low-level interactions.

[0075] S402: The time period corresponding to the maximum user activity index is used as the push time period:

[0076] T* = argmax h A(h)

[0077] Among them, T * Indicates the push time period.

[0078] S403: Each state element of the hidden Markov model under the observation sequence is used as the basic state of the user portrait, and each state element in the hidden state sequence, namely the user preference label, is used as the preference state of the user portrait. The connection between the basic state and the preference state is the state transition probability between the two, and the push time period is added to the user portrait to complete the creation of the user portrait.

[0079] Specifically, the user activity index in each time period is calculated by analyzing user behavior data, taking into account the frequency, duration and click frequency of user activities. The click frequency is logarithmically transformed to balance the influence of activities and avoid excessive impact of extreme values ​​on the overall activity index. Then, the time period with the highest activity index is selected as the best push time, and the user behavior state (such as preferred content type) obtained by hidden Markov model analysis is integrated into the user portrait, while linking the transition probability of the basic state and the preferred state, and adding the determined optimal push time period to the user portrait. It can accurately capture and utilize the user's activity in a specific time period, thereby optimizing the timeliness and relevance of content push. By directly reflecting the user's behavior patterns and preferences in the user portrait and combining them with real-time activity data, the push system can more intelligently decide when to display content to the user, greatly improving user participation and satisfaction, which not only enhances the user experience, but also optimizes resource allocation and improves operational efficiency by increasing the viewing rate and interaction rate of content.

[0080] S5: Obtain the category label of the data content currently accessed by the user.

[0081] S6: Calculate the similarity between each category label and the category label of the data content currently accessed by the user using the improved cosine similarity with element sparsity.

[0082] It should be noted that category labels are all text data, and text data is a sparse data, and most of the elements of the vector may be zero. In this case, even if two documents or labels have similar trends in several non-zero dimensions, their cosine similarity may be pulled down due to the presence of a large number of zeros, which cannot truly reflect their similarity. By introducing element sparsity in cosine similarity, the sparsity problem can be considered in the process of calculating the similarity of category labels, and a more accurate similarity value result can be obtained, thereby improving the accuracy of subsequent content push based on similarity. The similarity between the category label of the content currently accessed by the user and other category labels in the database is calculated by considering the impact of data sparsity. It can more accurately handle the sparse characteristics of text data, thereby improving the accuracy of similarity calculation, which is particularly important for recommendation systems because it can ensure that even in the case of sparse data, the true similarity between content can be effectively identified, thereby improving the relevance of recommendations and user satisfaction.

[0083] In a possible implementation, S6 specifically includes:

[0084] S601: vectorize each category label to obtain a vectorized category label.

[0085] S602: Introduce an element sparsity coefficient that is inversely proportional to the number of non-zero vectorized class labels:

[0086]

[0087] Among them, s represents the element sparsity coefficient, m represents the number of non-zero elements in the vectorized category label, and M represents the total number of elements in the vectorized category label.

[0088] S603: Calculate the similarity between any two category labels by combining the element sparsity coefficients of each category label:

[0089]

[0090] in, Represent the vectorized category labels of category label u and category label v respectively, Represents the similarity value between category label u and category label v, s u and v Represents the vectorized category labels and vectorized class labels The element sparsity coefficient of .

[0091] Among them, the square root is added in the calculation process of the similarity value, mainly to alleviate the excessive adjustment of the sparsity factor on the similarity value. Through the use of square roots, the adjustment effect of the sparsity factor can be made more gentle, avoiding the abnormal decrease in the similarity value caused by the two vectors being very sparse, thereby maintaining the accuracy and stability of the similarity measurement. Such processing not only makes the results more in line with the actual situation, but also enhances the applicability and explanatory power of the algorithm.

[0092] Specifically, by introducing the element sparsity coefficient, the common problems in processing sparse vector data are effectively solved. The sparsity coefficient increases the importance of non-zero elements when calculating similarity, so that category labels that are very sparse in vector representation can be more accurately evaluated and distinguished. This method is particularly suitable for processing text data, which can reduce the noise impact caused by the sparsity of high-dimensional data and improve the accuracy and efficiency of similarity calculation. In addition, considering the sparsity coefficient, this method can better capture the subtle differences between different category labels in large-scale data sets, which helps to improve the correlation strength between content pushed based on similarity values.

[0093] S7: Generate the push data content for the next moment based on the category label of the data content currently accessed by the user, combined with the user portrait and similarity value.

[0094] It can be understood that the system generates the push data content for the next moment based on the content category label currently visited by the user, combined with the constructed user portrait and the similarity calculated through step S6, which can adapt to the user's latest interests in real time and optimize content push by matching user portraits and content similarities, making recommended content more relevant and attractive to users. This not only improves user engagement and satisfaction, but also helps to enhance user loyalty and improve the overall user experience by accurately matching user needs and preferences.

[0095] In a possible implementation, S7 specifically includes:

[0096] S701: Generate an initial category label corresponding to the category label of the data content currently accessed by the user based on the state transition probability and the user portrait, wherein the initial category label is the category label with the highest state transition probability under the category label of the data content currently accessed by the user.

[0097] It should be noted that those skilled in the art can set the size of the preset state transition probability according to actual needs, and the present invention does not limit this herein.

[0098] S702: Calculate the category label with the largest similarity value with the initial category label as the expected category label.

[0099] S703: Output the data content corresponding to the expected category label as the push data content.

[0100] Specifically, first, the most likely initial category label to be transferred to is determined based on the category label of the content currently accessed by the user and the state transition probability. This step takes into account the user's historical behavior pattern to predict the next category that the user is most likely to be interested in. Then the similarity between this initial category label and other category labels is calculated, and the category label with the highest similarity is selected as the expected category label. Finally, the system pushes the content corresponding to this expected category label to the user. It not only predicts and adapts to the user's behavior based on the user's historical interaction data to avoid the situation where the pushed content is exactly the same as the browsed content, but also enhances the relevance and accuracy of content recommendation through similarity analysis. This method can dynamically expand the user's content consumption range, not only limited to the content that the user has shown direct interest in, but also explore and recommend new and relevant content, thereby enhancing the user experience, improving user satisfaction and engagement. In addition, the flexibility and adaptability of this strategy enables it to cope with changing user preferences and a wide range of content types, enhancing the platform's attractiveness and user stickiness.

[0101] It should be noted that this process not only makes full use of user browsing data, but also mines out other expected category labels with higher similarity values ​​to user browsing data based on similarity values ​​while excluding the influence of sparseness. These other expected category labels can be existing data in the library or newly added. Finally, the expected category labels are obtained based on the similarity values, and then the corresponding data content is pushed accurately based on the expected category labels. In addition, the push type labels can be effectively expanded, and are not limited to the type labels of possible transfers recorded by the state transition probability, because it is impossible for all type labels to be the same. This method effectively increases the applicable scenarios and scope of the solution.

[0102] S8: Record the length of time the user stays under the pushed data content.

[0103] S9: When the stay time is less than the preset stay time, a correction factor for the difference between the stay time and the preset stay time is set to correct the state transition probability corresponding to the category label of the data content currently accessed by the user.

[0104] It should be noted that when the user's stay time is less than the preset stay time, the difference is used as a correction factor to adjust the state transition probability. By correcting these probabilities, the system can more accurately reflect the user's actual behavior and preference changes, and use the stay time as the driving data for the correction factor to continuously increase the user's stay time on the pushed content, forming a virtuous circle. Through dynamic adjustment, the pushed content is more in line with the user's current interests and reactions, thereby improving user engagement and content click-through rate. In addition, timely adjustment of push strategies can also help avoid user fatigue and improve user experience, making the content push system more efficient and accurate.

[0105] It should be noted that those skilled in the art can set the preset stay time according to actual needs, and the present invention is not limited thereto.

[0106] In a possible implementation, S9 specifically includes:

[0107] S901: Setting a correction factor based on the difference between the stay duration and the preset stay duration. The correction factor is calculated as follows:

[0108]

[0109] Among them, γ(τ) represents the correction factor for the residence time τ, λ represents the decay rate, and τ target represents the preset stay duration, μ represents the average stay duration under the total number of stay duration records, and e represents the base of the exponential function.

[0110] Among them, the decay rate is directly related to the average length of stay of users, which is an intuitive engagement indicator. A shorter average length of stay means a higher decay rate, reflecting the user's rapid consumption of content or low engagement. A longer length of stay indicates a lower decay rate, showing the user's high level of investment and satisfaction with the content.

[0111] It should be noted that the correction factor increases as the stay time approaches the preset time. When the actual stay time is less than the preset time, this value will decrease, thereby reducing the state transition probability, and then making the stay time continue to tend towards the preset stay time, more accurately capturing the user's browsing behavior.

[0112] S902: Correct the state transition probability corresponding to the category label of the data content currently accessed by the user according to the correction factor:

[0113] A ij =γ(τ)×a ij

[0114] Among them, A ij represents the corrected state transition probability.

[0115] S10: Use the corrected state transition probability as the new state transition probability and return to step S5.

[0116] It is understandable that the state transition matrix of the entire model is updated using the previously corrected state transition probability. This update process ensures that the model reflects the latest user behaviors and preferences, making future content recommendations more accurate and personalized, and enabling the push method to continuously optimize itself and adjust its behavior through real-time feedback loops, thereby improving the relevance of pushed content and user satisfaction, enhancing user experience, and ultimately increasing long-term user engagement. This dynamic update mechanism provides support for ensuring the long-term effectiveness and adaptability of the push method, continuously dynamically updating and optimizing push content to achieve accurate content push.

[0117] In a possible implementation manner, after S10, the method further includes:

[0118] Update the data in the library at preset intervals.

[0119] It should be noted that the purpose of updating the data in the database at preset intervals is to ensure that the content library remains up to date, thereby increasing the relevance and attractiveness of content push. By regularly updating the data in the library, the system can introduce new content or update existing content to ensure that the information users are exposed to is the latest and most relevant, which not only helps maintain user interest and engagement, but also improves the competitiveness of the platform and ensures the freshness and diversity of content.

[0120] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0121] In the present invention, firstly, the category labels of each data content in the data library and the historical behavior data of the user are obtained, and then the hidden Markov model is introduced to perform data trend analysis on the user's historical behavior data based on the category labels, so as to capture the user's preference and state transition probability, and can fully mine the trend and probability of the user transferring to different categories of data under the current access data state based on the behavior data, and generate a user portrait with a push time period, and then combine the improved cosine similarity with element sparsity to calculate the similarity value between each category label and the category label of the data content currently accessed by the user, even if there is no category label that the user wants to access, it is still possible to push content that is extremely in line with the user's expectations based on the similarity value. This data-driven approach reduces the problem of heavy reliance on data category labels under data-driven, and can effectively expand the push data according to user behavior. While ensuring that the push is in line with user behavior, it greatly improves the scope of application of the push method, the push accuracy and the rationality of the push time, improves user stickiness, and increases business traffic.

[0122] Reference Manual Attached Figure 2, showing a structural diagram of a user portrait-based precise push system provided by the present invention.

[0123] The present invention further provides a user portrait-based accurate push system 20, which is applied to the above-mentioned user portrait-based accurate push method, including:

[0124] Processor 201.

[0125] The memory 202 stores computer-readable instructions. When the computer-readable instructions are executed by the processor 201, a precise push method based on user portrait as in the method embodiment is implemented.

[0126] The user portrait-based precise push system 20 provided by the present invention can execute the above-mentioned user portrait-based precise push method and achieve the same or similar technical effects. To avoid repetition, the present invention will not go into details.

[0127] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0128] In the present invention, firstly, the category labels of each data content in the data library and the historical behavior data of the user are obtained, and then the hidden Markov model is introduced to perform data trend analysis on the user's historical behavior data based on the category labels, so as to capture the user's preference and state transition probability, and can fully mine the trend and probability of the user transferring to different categories of data under the current access data state based on the behavior data, and generate a user portrait with a push time period, and then combine the improved cosine similarity with element sparsity to calculate the similarity value between each category label and the category label of the data content currently accessed by the user, even if there is no category label that the user wants to access, it is still possible to push content that is extremely in line with the user's expectations based on the similarity value. This data-driven approach reduces the problem of heavy reliance on data category labels under data-driven, and can effectively expand the push data according to user behavior. While ensuring that the push is in line with user behavior, it greatly improves the scope of application of the push method, the push accuracy and the rationality of the push time, improves user stickiness, and increases business traffic.

[0129] It should be understood that the processor in the embodiment of the present invention may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0130] It should also be understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus random access memory (DR RAM).

[0131] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0132] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0133] In the present invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0134] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0135] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0136] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0137] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0138] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0139] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0140] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc., various media that can store program codes.

[0141] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, the precise push method based on user portrait as described in the method embodiment is implemented.

[0142] A computer-readable storage medium provided by the present invention can implement the steps and effects of the precise push method based on user portrait in the above method embodiment. To avoid repetition, the present invention will not go into details.

[0143] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0144] In the present invention, firstly, the category labels of each data content in the data library and the historical behavior data of the user are obtained, and then the hidden Markov model is introduced to perform data trend analysis on the user's historical behavior data based on the category labels, so as to capture the user's preference and state transition probability, and can fully mine the trend and probability of the user transferring to different categories of data under the current access data state based on the behavior data, and generate a user portrait with a push time period, and then combine the improved cosine similarity with element sparsity to calculate the similarity value between each category label and the category label of the data content currently accessed by the user, even if there is no category label that the user wants to access, it is still possible to push content that is extremely in line with the user's expectations based on the similarity value. This data-driven approach reduces the problem of heavy reliance on data category labels under data-driven, and can effectively expand the push data according to user behavior. While ensuring that the push is in line with user behavior, it greatly improves the scope of application of the push method, the push accuracy and the rationality of the push time, improves user stickiness, and increases business traffic.

[0145] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0146] There are a few points to note:

[0147] (1) The drawings of the embodiments of the present invention only relate to the structures related to the embodiments of the present invention, and other structures may refer to the general design.

[0148] (2) For the sake of clarity, in the drawings used to describe the embodiments of the present invention, the thickness of the layers or regions is exaggerated or reduced, that is, these drawings are not drawn according to the actual scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or there may be intermediate elements.

[0149] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to obtain new embodiments.

[0150] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A precise push method based on user portrait, characterized in that: include: S1: Get the category labels of each data content in the database; S2: Obtaining the user's behavior data on the data content corresponding to each category label, wherein the behavior data includes browsing data content category, browsing time, browsing interval, stay time and click frequency; S3: combining the behavior data and the hidden Markov model to generate a state transition probability and a user preference label, wherein the state transition probability is the probability that the user directly browses the data content of the second category label when browsing the data content of the first category label; S4: Generate a user portrait with a push time period based on the behavior data and the user preference tag; S5: Obtain the category label of the data content currently accessed by the user; S6: Calculate the similarity value between each category label and the category label of the data content currently accessed by the user by using the improved cosine similarity with element sparsity; S7: Generate push data content at the next moment according to the category label of the data content currently accessed by the user, combined with the user portrait and the similarity value; S8: Recording the length of time the user stays under the pushed data content; S9: When the stay time is less than the preset stay time, a correction factor for the difference between the stay time and the preset stay time is set to correct the state transition probability corresponding to the category label of the data content currently accessed by the user; S10: Use the corrected state transition probability as the new state transition probability and return to step S5.

2. The accurate push method based on user portrait according to claim 1 is characterized in that: The S3 specifically includes: S301: defining the state element and observation element of the hidden Markov model, wherein the state element is the category label corresponding to the data content browsed by the user, and the observation element is the behavior data; S302: Establishing an observation sequence based on the behavior data; S303: Initializing algorithm parameters of the hidden Markov model according to the observation sequence, wherein the algorithm parameters include a state transition probability matrix, an observation probability matrix and an initial state probability: Each element in the state transition probability matrix is ​​specifically: Among them, a ij represents the state transition probability from category label i to category label j, q t represents the state element at time t, i.e., the category label, T represents the total number of discrete moments in the observation sequence, P(q t =i,q t+1 =j|O,λ) represents the joint probability that a class is in class label i at time t and is transferred to class label j at time t+1, given the observation sequence O and the model parameter λ. t =i|O,λ) represents the probability of being in category label i at time t; Each element in the observation probability matrix is ​​specifically: Among them, b j (k) represents the observation probability of k when the class label is j, o t represents the observed value at time t in the observation sequence, P(q t =j,o t = k|λ) represents the joint probability of being in category label j and observing k at time t given the model parameter λ, P(q t =j|O,λ) means that the state element q at time t is given the observation sequence O and the model parameter λ. t is the overall occurrence probability of category label j, i.e., the initial state probability; S304: using the behavior data as input data of the hidden Markov model, training the hidden Markov model, and updating the state transition probability; S305: Given the observation sequence, generate a hidden state sequence based on the behavior data through a trained hidden Markov model, wherein the state element of the hidden state sequence is the user preference label.

3. The accurate push method based on user portrait according to claim 2 is characterized in that: The update formula of the state transition probability is specifically: Among them, a' ij represents the updated state transition probability, α t (i) represents the forward probability at time t, that is, the probability of the class label being i in the observation sequence from the beginning of the sequence to time t, b j (o t+1 ) means that o is observed when the class label is j t+1 The observation probability, β t+1 (j) represents the backward probability at time t+1, that is, the probability of the category label being j at time t+1 in the observation sequence O from time t+1 to the end of the sequence. P(O|λ) represents the probability of occurrence of the observation sequence O under given model parameters.

4. The accurate push method based on user portrait according to claim 2 is characterized in that: The S4 specifically includes: S401: Calculate the user activity index in each time period based on the behavior data: Where A(h) represents the user activity index in the time period h of the browsing moment, N h represents the number of user activities collected in time period h according to the browsing interval, d i represents the duration of the i-th user activity, D represents the maximum duration of all user activities, k i represents the click frequency in the i-th user activity, and log represents logarithmic operation; S402: The time period corresponding to the maximum user activity index is used as the push time period: T * =argmax h A(h) Among them, T * Indicates the push time period; S403: Each state element of the hidden Markov model under the observation sequence is used as the basic state of the user portrait, each state element in the hidden state sequence, namely the user preference label, is used as the preference state of the user portrait, the line between the basic state and the preference state is the state transition probability between the two, and the push time period is added to the user portrait to complete the creation of the user portrait.

5. The accurate push method based on user portrait according to claim 1 is characterized in that: The S6 specifically includes: S601: vectorize each category label to obtain a vectorized category label; S602: Introduce an element sparsity coefficient that is inversely proportional to the number of non-zero vectorized class labels: Wherein, s represents the element sparsity coefficient, m represents the number of non-zero elements in the vectorized category label, and M represents the total number of elements in the vectorized category label; S603: Calculate the similarity between any two category labels by combining the element sparsity coefficients of each category label: in, Represent the vectorized category labels of category label u and category label v respectively, Represents the similarity value between category label u and category label v, s u and v Represents the vectorized category labels and vectorized class labels The element sparsity coefficient of .

6. The accurate push method based on user portrait according to claim 1 is characterized in that: The S7 specifically includes: S701: Generate an initial category label corresponding to the category label of the data content currently accessed by the user according to the state transition probability and the user portrait, wherein the initial category label is the category label with the highest state transition probability when the category label of the data content currently accessed by the user is the category label; S702: Calculate the category label with the largest similarity value with the initial category label as the expected category label; S703: Output the data content corresponding to the expected category label as the pushed data content.

7. The accurate push method based on user portrait according to claim 1 is characterized in that: The S9 specifically includes: S901: Setting the correction factor based on the difference between the stay duration and the preset stay duration, wherein the correction factor is calculated as follows: Among them, γ(τ) represents the correction factor for the residence time τ, λ represents the decay rate, and τ target represents the preset stay time, μ represents the average stay time under the total number of stay time records, and e represents the base of the exponential function; S902: Correct the state transition probability corresponding to the category label of the data content currently accessed by the user according to the correction factor: A ij =γ(τ)×a ij Among them, A ij represents the corrected state transition probability.

8. The accurate push method based on user portrait according to claim 1 is characterized in that: After S10, the method further includes: The data in the library is updated at preset time intervals.

9. A precise push system based on user portrait, characterized in that: include: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the precise push method based on user portrait as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, a precise push method based on user portrait as described in any one of claims 1 to 8 is implemented.

Citation Information

Patent Citations

  • Data processing method and device for recommendation system and electronic device

    CN110555160A

  • Label-based short video recommendation method, system, equipment and medium

    CN114265953A

  • System for testing accuracy of big data recommendation algorithm

    CN114297518A

  • Message pushing method based on cosine similarity and related device

    CN114780848A

  • Video recommendation method and device

    CN116801056A

Cited By

  • User portrait label generation method and system based on time-aligned transfer model

    CN119989032A

  • User Portrait Label Generation Method and System Based on a Homogeneous Transition Model

    CN119989032B

  • Dynamic design strategy generation method, device and equipment and readable storage medium

    CN120822274A