Large model recommendation method and system based on individual user side behaviors

Abstract

The invention relates to the field of data processing, in particular to a large model recommendation method and system based on personal user side behaviors. The method comprises the steps of obtaining user behavior data of a target user; inputting the user behavior data into the active analysis model for user portraying to obtain behavior preference information of the target user; on the basis of the behavior preference information and the current geographic position of the target user, behavior track information of the target user in a future preset duration is predicted through a behavior construction model; the behavior track information comprises a behavior node of which the prediction execution probability reaches a preset threshold in a future preset duration of the target user; and recommending a comprehensive service unit matched with the current geographic position to the target user based on the behavior track information. The user behavior and the environment where the user is located are comprehensively analyzed from different angles through the active analysis model and the behavior construction model, the information recommendation accuracy and recommendation efficiency are improved, various comprehensive service information is provided for the user, and the user experience is improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a large model recommendation method and system based on individual user terminal behavior. Background Art

[0002] User behavior covers a variety of online and offline activities, including shopping, socializing, entertainment, learning, work, etc. Each behavior has multiple forms and scenarios. For example, in terms of shopping behavior, users may shop in physical stores or through different e-commerce platforms. The categories, frequency, and time of shopping vary greatly, making it difficult to accurately describe and predict with a universal model.

[0003] In related technologies, user behavior covers a variety of online and offline activities, including shopping, socializing, entertainment, learning, work, etc., and each behavior has multiple forms and scenarios. For example, in terms of shopping behavior, users may shop in physical stores or through different e-commerce platforms. The categories, frequency, time, etc. of shopping vary greatly, which makes it difficult to accurately describe and predict with a universal model. In addition, with the development of the Internet and mobile devices, user behavior changes rapidly, requiring models to be able to process and analyze data in real time and update predictions of user behavior in a timely manner. However, some existing models may face problems such as insufficient computing resources and slow processing speed when processing large-scale real-time data, and cannot meet the needs of real-time prediction.

[0004] Therefore, it is urgent to design a new technical solution to solve at least one of the above technical problems. Summary of the invention

[0005] In view of the technical problems existing in the prior art, the present invention provides a large model recommendation method and system based on individual user-side behavior, which is used to solve at least one of the above-mentioned technical problems such as data description differences, insufficient computing resources, and slow processing speed.

[0006] In a first aspect, an embodiment of the present application provides a large model recommendation method based on individual user terminal behavior, including:

[0007] Obtaining user behavior data of the target user; the user behavior data at least includes: online behavior data and offline behavior data;

[0008] The user behavior data is input into an active analysis model to perform user profiling, thereby obtaining the behavior preference information of the target user; the active analysis model is used to mine and analyze the behavior preference features in the user behavior data from multiple dimensions;

[0009] Based on the behavior preference information and the current geographic location of the target user, predict the behavior trajectory information of the target user within a preset time period in the future through a behavior construction model; the behavior trajectory information includes the behavior nodes of the target user whose predicted execution probability reaches a preset threshold within the preset time period in the future;

[0010] Based on the behavior trajectory information, a comprehensive service unit matching the current geographical location is recommended to the target user; the comprehensive service unit includes a service unit matching the behavior node in the behavior trajectory information, and the service unit is at least one of an offline product, an offline store, an offline service business, an online service business, and an online product.

[0011] In a second aspect, an embodiment of the present application provides a large model recommendation system based on individual user terminal behavior, the system comprising the following units:

[0012] An acquisition unit, configured to acquire user behavior data of a target user; the user behavior data at least includes: online behavior data and offline behavior data;

[0013] An analysis unit is used to input the user behavior data into an active analysis model to perform user profiling and obtain the behavior preference information of the target user; the active analysis model is used to mine and analyze the behavior preference features in the user behavior data from multiple dimensions;

[0014] A prediction unit, configured to predict the behavior trajectory information of the target user within a preset time period in the future through a behavior construction model based on the behavior preference information and the current geographical location of the target user; the behavior trajectory information includes the behavior nodes of which the predicted execution probability of the target user within the preset time period in the future reaches a preset threshold;

[0015] A recommendation unit is used to recommend a comprehensive service unit that matches the current geographical location to the target user based on the behavior trajectory information; the comprehensive service unit includes a service unit that matches the behavior node in the behavior trajectory information, and the service unit is at least one of an offline product, an offline store, an offline service business, an online service business, and an online product.

[0016] In a third aspect, an embodiment of the present application provides an electronic device, the electronic device comprising:

[0017] at least one processor, memory, and input-output unit;

[0018] The memory is used to store a computer program, and the processor is used to call the computer program stored in the memory to execute the large model recommendation method based on personal user terminal behavior of the first aspect.

[0019] In a fourth aspect, a computer-readable storage medium is provided, which includes instructions, and when the instructions are executed on a computer, the computer executes the large model recommendation method based on individual user-side behavior of the first aspect.

[0020] The beneficial effect of the present invention is that it provides a large model recommendation method and system based on individual user-side behavior. In this technical solution, the user behavior data of the target user is obtained; the user behavior data at least includes: online behavior data and offline behavior data. Then, the user behavior data is input into the active analysis model for user profiling to obtain the behavior preference information of the target user; the active analysis model is used to mine and analyze the behavior preference characteristics in the user behavior data from multiple dimensions. Then, based on the behavior preference information and the current geographical location of the target user, the behavior trajectory information of the target user within a preset time period in the future is predicted through the behavior construction model; the behavior trajectory information includes the behavior nodes of the target user whose predicted execution probability reaches a preset threshold within the preset time period in the future. Finally, based on the behavior trajectory information, a comprehensive service unit matching the current geographical location is recommended to the target user; the comprehensive service unit includes a service unit matching the behavior node in the behavior trajectory information, and the service unit is at least one of an offline product, an offline store, an offline service business, an online service business, and an online product.

[0021] The embodiment of the present application can obtain comprehensive and detailed insights into the user's real behavior patterns and preferences by acquiring user behavior data covering both online and offline, laying a solid foundation for subsequent precise analysis. The active analysis model deeply mines and analyzes behavioral preference characteristics from multiple dimensions, can accurately construct user portraits, outline user behavior preference information, and allow the system to have a more thorough understanding of the user. Based on the behavioral preference information and the user's current geographic location, the behavioral trajectory information within a preset time period in the future is predicted with the help of the behavior construction model, which can proactively predict the user's possible behavior, and filter out high-probability behavior nodes by setting a preset threshold for the prediction execution probability, making the prediction more reliable and practical. Finally, based on the behavioral trajectory information, a comprehensive service unit matching the current geographic location is recommended, which includes various service units matching the behavior nodes, achieving a high degree of fit between the service recommendation and the user's immediate needs and the scene in which they are located, greatly improving the accuracy and relevance of the recommendation, providing users with services that are more in line with actual needs, enhancing the user experience, and also helping to improve the operating efficiency and resource utilization of service providers. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flowchart of a large model recommendation method based on individual user terminal behavior in an embodiment of the present application;

[0023] Figure 2It is a structural diagram of a large model recommendation system based on individual user terminal behavior in an embodiment of the present application;

[0024] Figure 3 is a schematic diagram of the structure of an electronic device according to an embodiment of the present application;

[0025] Figure 4 It is a structural schematic diagram of a medium device according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0027] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, "plurality" means two or more, unless otherwise clearly and specifically defined.

[0028] In the description of the present application, the term "for example" is used to mean "used as an example, illustration or description". Any embodiment described as "for example" in the present application is not necessarily to be construed as being more preferred or advantageous than other embodiments. In order to enable any technician in the field to implement and use the present invention, the following description is given. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid unnecessary details that make the description of the present invention obscure. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present application.

[0029] The embodiment of the present application provides a large model recommendation method and system based on individual user-side behavior. In this technical solution, the user behavior data of the target user is obtained; the user behavior data at least includes: online behavior data and offline behavior data. Then, the user behavior data is input into the active analysis model for user profiling to obtain the behavior preference information of the target user; the active analysis model is used to mine and analyze the behavior preference characteristics in the user behavior data from multiple dimensions. Then, based on the behavior preference information and the current geographical location of the target user, the behavior trajectory information of the target user within a preset time period in the future is predicted through the behavior construction model; the behavior trajectory information includes the behavior nodes of the target user whose predicted execution probability reaches a preset threshold within the preset time period in the future. Finally, based on the behavior trajectory information, a comprehensive service unit matching the current geographical location is recommended to the target user; the comprehensive service unit includes a service unit matching the behavior node in the behavior trajectory information, and the service unit is at least one of an offline product, an offline store, an offline service business, an online service business, and an online product.

[0030] The embodiment of the present application can obtain comprehensive and detailed insights into the user's real behavior patterns and preferences by acquiring user behavior data covering both online and offline, laying a solid foundation for subsequent precise analysis. The active analysis model deeply mines and analyzes behavioral preference characteristics from multiple dimensions, can accurately construct user portraits, outline user behavior preference information, and allow the system to have a more thorough understanding of the user. Based on the behavioral preference information and the user's current geographic location, the behavioral trajectory information within a preset time period in the future is predicted with the help of the behavior construction model, which can proactively predict the user's possible behavior, and filter out high-probability behavior nodes by setting a preset threshold for the prediction execution probability, making the prediction more reliable and practical. Finally, based on the behavioral trajectory information, a comprehensive service unit matching the current geographic location is recommended, which includes various service units matching the behavior nodes, achieving a high degree of fit between the service recommendation and the user's immediate needs and the scene in which they are located, greatly improving the accuracy and relevance of the recommendation, providing users with services that are more in line with actual needs, enhancing the user experience, and also helping to improve the operating efficiency and resource utilization of service providers.

[0031] It is understandable that the above-mentioned large model recommendation method and system based on individual user terminal behavior solves the problems existing in the related art in the following ways:

[0032] In order to solve the problem that the diversity of user behavior is difficult to describe and predict in a universal way, the embodiments of the present application obtain the online and offline behavior data of the target user, which comprehensively covers the user's shopping, social, entertainment, learning, work and other activity data. For example, not only online data such as shopping records on e-commerce platforms are collected, but also offline data such as consumption records in physical stores are collected. This multi-source data integration method avoids the one-sidedness of information caused by relying solely on a single data source, thereby gaining a more comprehensive understanding of the overall picture of user behavior and providing a rich data foundation for accurately describing and predicting user behavior.

[0033] In addition, active analysis models are used to mine and analyze the behavioral preference characteristics in user behavior data from multiple dimensions. For shopping behavior, it will be analyzed from multiple dimensions such as shopping platform, category, frequency, time, etc. For example, by analyzing the shopping frequency of different e-commerce platforms and the differences in shopping categories in different time periods, a detailed user shopping behavior preference portrait is constructed. This multi-dimensional analysis breaks the limitations of a single general model and can more accurately portray the characteristics of users in different behavioral forms and scenarios, thereby improving the accuracy of predictions.

[0034] In order to solve the computing resource and speed problems faced by real-time prediction, the large model recommendation system based on the behavior of individual user terminals in the embodiment of the present application can better cope with large-scale real-time data with the help of the powerful computing and processing capabilities of the large model. The large model has a large number of parameters and a strong learning ability, which can efficiently process and analyze a large amount of influx of real-time data. Compared with traditional models, it has significant advantages in processing scale and speed. For example, when faced with a large number of online and offline behavior data generated by users in real time, the large model can extract features and perform analysis faster to provide support for real-time prediction.

[0035] In the embodiment of the present application, when designing the behavior construction model, the architecture and algorithm are optimized according to the real-time requirements. For example, a stream computing architecture suitable for processing real-time data is adopted, combined with a lightweight but efficient algorithm, which can quickly process data under limited computing resources. When predicting the user's future behavior trajectory information within a preset time, the algorithm is optimized to reduce the computational complexity and improve the processing speed, so that the model can update the prediction of user behavior in a timely manner to meet the real-time requirements.

[0036] The large model recommendation scheme based on personal user-side behavior provided in the embodiments of the present application can also be executed by an electronic device, which can be a server, a server cluster, or a cloud server. The electronic device can also be a terminal device such as a mobile phone, a computer, a tablet computer, a wearable device, or a dedicated device (such as a dedicated terminal device with a large model recommendation system based on personal user-side behavior). These electronic devices can also be equipped with the chips introduced in the above embodiments. Alternatively, these electronic devices can also be installed with a service program for executing a large model recommendation scheme based on personal user-side behavior.

[0037] Figure 1 A schematic diagram of a large model recommendation method based on individual user terminal behavior provided in an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0038] 101. Obtain user behavior data of target users;

[0039] 102. Input the user behavior data into an active analysis model to perform user profiling, and obtain the behavior preference information of the target user;

[0040] 103. Based on the behavior preference information and the current geographical location of the target user, predict the behavior trajectory information of the target user within a preset time period in the future through a behavior construction model;

[0041] 104. Recommending a comprehensive service unit matching the current geographical location to the target user based on the behavior trajectory information.

[0042] In an embodiment of the present application, the user behavior data includes at least: online behavior data and offline behavior data. Online behavior data includes the user's operations on various network platforms, such as web browsing records, which can reflect their interest tendencies; search history, which can provide insights into their needs and concerns. Online transaction information presents consumer preferences and capabilities, etc. Offline behavior data involves the user's activities in real-life scenarios, such as physical store consumption records, which show actual consumption choices. Participation in offline activities reflects social, hobbies and other aspects. The integration of these two types of data can comprehensively outline the overall picture of user behavior, provide a rich and multi-dimensional information foundation for subsequent analysis, portrait construction and precise recommendations based on user behavior, and help to gain a deeper understanding of user needs and behavior patterns.

[0043] In 101, obtaining the user behavior data of target users is the basis for accurate analysis and service to users. This data can be collected through various channels, covering multiple areas online and offline.

[0044] Online, various Internet platforms are important sources of data. For example, e-commerce platforms can record users' browsing history, which can provide insights into which categories of goods users are interested in, as well as how long they stay when browsing goods, reflecting their level of attention. Search records clearly show users' immediate needs and exploration directions. Transaction records present users' consumption preferences in detail, including the types of goods purchased, brands, price ranges, purchase frequencies, etc. These data combined outline the outline of users' online consumption behavior. Social platforms can also provide rich information. Users' interactions with friends, posting content, participating in topic discussions, and other behaviors reflect their social circles, interests, hobbies, values, etc.

[0045] Offline, through the consumption records of physical stores, we can understand the consumption choices of users in real scenarios. For example, the shopping list in the supermarket reflects daily consumption needs, and the purchase records in the specialty store show the preference for specific brands or categories. In addition, offline activity participation is also important data. Users' participation in various exhibitions, lectures, sports events and other activities can reflect their interests and lifestyles. Through smart devices such as mobile payment terminals and mobile phones with positioning functions, users' geographic location information can also be obtained, combined with time data, to analyze the user's travel patterns and frequented places.

[0046] In this way, the user behavior data collected through multiple online and offline channels can comprehensively and three-dimensionally present the overall behavior of the target users, provide a solid data foundation for the subsequent construction of user portraits, analysis of behavioral preferences, and precise recommendation services, and help enterprises and service providers better meet user needs and improve service quality and user experience.

[0047] In an embodiment of the present application, the active analysis model is used to mine and analyze the behavioral preference characteristics in the user behavior data from multiple dimensions.

[0048] For example, the principle of the active analysis model can be based on deep learning and data mining technology, using neural networks, decision trees and other algorithms to build a complex model structure. The model first pre-processes the acquired user behavior data, unifies the format and cleans up the noise data, then extracts features from the data and converts the original data into a vector form that can be understood by the model. Then, the model's multi-layer structure is used to automatically learn hidden patterns and relationships in the data, such as capturing local features in the data through the convolution layer, processing sequence data through the loop layer to analyze the time series characteristics of the behavior, and mining behavioral preference characteristics from multiple dimensions such as consumption, social interaction, and interests.

[0049] Therefore, the active analysis model can provide deep insights into the real preferences behind user behavior, accurately extract users' interests and behavioral tendencies in different fields, and provide accurate basis for user portraits. The recommendation system can then provide users with highly personalized and demand-oriented recommendations based on these accurate preference characteristics, effectively improve users' acceptance and satisfaction with recommended content, and enhance the interactive stickiness between users and the system.

[0050] In the embodiment of the present application, the behavior trajectory information includes the behavior nodes of the target user whose predicted execution probability reaches the preset threshold within the future preset time period. Specifically, in the embodiment of the present application, the behavior trajectory information is crucial to accurately grasp the future behavior of the target user. The behavior nodes included therein are derived based on the prediction of the target user's behavior. These predictions are not random guesses, but are scientifically calculated through the behavior construction model in combination with multiple factors such as user behavior preference information and current geographical location. The predicted execution probability reflects the possibility of the actual occurrence of the behavior node within the future preset time period. The preset threshold is an artificially set standard value. Only the behavior nodes whose predicted execution probability reaches this threshold will be included in the behavior trajectory information. This setting effectively filters out behavior predictions with low probability, ensures that the behavior trajectory information focuses on behaviors with high probability of occurrence, and makes the subsequent comprehensive service units recommended to users based on this information more targeted and practical, which can better meet the real needs of users and improve the accuracy and effectiveness of recommendations.

[0051] In an embodiment of the present application, the comprehensive service unit includes a service unit that matches the behavior node in the behavior trajectory information, and the service unit is at least one of an offline product, an offline store, an offline service business, an online service business, and an online product.

[0052] For example, if the target user's behavior trajectory information predicts that he has a high probability of purchasing sports equipment in a nearby shopping mall in the next hour (behavior node), the comprehensive service unit will match this behavior. This may include offline products, such as sports shoes and sportswear of a certain brand in a shopping mall; offline stores, such as specialty stores of well-known sports brands; offline service businesses, such as on-site try-on and customization services for sports equipment. If the user is in an online scene, the service unit may also have online service businesses, such as online push of sports and fitness courses; online products, such as the same online sports equipment recommended by sports apps.

[0053] Obviously, this matching method can greatly improve the accuracy of recommendations and accurately hit the potential needs of users. For users, they can quickly obtain services and products that they are interested in and that are in line with the current situation, improve user experience, and increase user attention and participation in recommended content. For merchants, accurate recommendations help improve the conversion rate of goods and services, optimize resource allocation, reduce ineffective promotion costs, improve operational efficiency and commercial benefits, and promote the coordinated development of online and offline businesses.

[0054] As an optional embodiment, the active analysis model includes at least the following structures: an extraction layer, a construction layer, an analysis layer, a prediction layer, and an output layer.

[0055] Based on the above model structure, in 102, the user behavior data is input into the active analysis model to perform user profiling, and the behavior preference information of the target user is obtained, including:

[0056] Extracting multi-dimensional user behavior features from the user behavior data through an extraction layer;

[0057] Through the construction layer, the multidimensional user behavior features are projected into the multidimensional portrait space according to the projection strategies corresponding to their respective dimensions to obtain the corresponding multidimensional user portrait; through the analysis layer, the multidimensional user portrait is adaptively segmented using pre-configured slice fitting parameters to obtain multiple behavior slices corresponding to the multidimensional user portrait; through the prediction layer, the matching probability between multiple behavior slices and preset behavior types is predicted to select candidate behavior types for the multidimensional user portrait; through the output layer, the behavior preference information of the target user is generated according to the candidate behavior types.

[0058] In the embodiment of the present application, each portrait coordinate point in the multi-dimensional user portrait corresponds to at least one behavior feature point in the multi-dimensional user behavior feature.

[0059] In the process of constructing a multidimensional user portrait, the multidimensional user portrait is like a complex coordinate space, and each of its portrait coordinate points has special significance. These coordinate points do not exist in isolation, but are closely connected with the multidimensional user behavior characteristics. Each portrait coordinate point corresponds to at least one behavioral feature point in the multidimensional user behavior characteristics. For example, in a multidimensional user portrait space constructed with user consumption frequency, consumption amount, and consumption category as dimensions, a specific portrait coordinate point may correspond to both the user's high-frequency consumption behavior feature point in the electronic product category and the user's high-consumption amount behavior feature point in this category. This correspondence enables the multidimensional user portrait to accurately map the user's complex behavioral characteristics. By analyzing these coordinate points, we can deeply understand the user's behavior patterns in different dimensions, providing a rich and accurate information basis for further user behavior analysis, personalized recommendations and other applications.

[0060] In an embodiment of the present application, multiple behavior slices are used to represent the behavior change characteristics of the multi-dimensional user portrait under different analysis angles; the slice fitting parameters are used to represent the slicing method of the multi-dimensional user portrait and the method of selecting the slice cutting position.

[0061] It is understandable that multiple behavior slices are used to represent the behavior change characteristics of the multi-dimensional user portrait from different analysis angles. These behavior slices divide the user portrait into multiple parts, which helps to analyze the user's behavior from multiple dimensions and perspectives. For example, in a three-dimensional user portrait, one behavior slice may represent a specific behavior pattern of the user in a certain time period and scenario, while another behavior slice may represent another behavior pattern of the user under different conditions.

[0062] By analyzing different behavior slices, we can understand the user's behavior preferences in more detail and provide richer information for the subsequent prediction layer. Different behavior slices can reflect the user's behavioral tendencies in different aspects. For example, in a shopping scenario, one slice may reflect the user's behavioral characteristics when purchasing high-value goods, and another slice may reflect the user's behavioral characteristics when purchasing daily necessities.

[0063] Specifically, in 102, first, the extraction layer acts as a data filter to select representative multi-dimensional user behavior features from complex user behavior data, such as extracting features such as consumption amount and interaction frequency from shopping, social and other multi-faceted behavior data. Then, the construction layer projects these features into the multi-dimensional portrait space according to the projection strategy corresponding to each dimension, just like placing different information points in a multi-dimensional coordinate system to form a multi-dimensional user portrait and comprehensively outline the user behavior profile. Then, the analysis layer uses the pre-configured slice fitting parameters to adaptively segment the multi-dimensional user portrait according to the data characteristics and analysis requirements, and obtains multiple behavior slices, which reflect the changes in user behavior from different angles. On this basis, the prediction layer evaluates the matching probability of these behavior slices with the preset behavior types, such as judging the degree of fit between the behavior pattern represented by a certain behavior slice and the preset types such as "sports enthusiasts" and "fashion seekers", and selects the behavior type with a high matching probability as the candidate behavior type. Finally, the output layer generates the behavior preference information of the target user based on the candidate behavior type, completing the key step of user portrait.

[0064] Exemplarily, assume that user behavior data covers information such as online shopping and social platform interaction. The extraction layer extracts multidimensional features such as the monthly online clothing purchase amount of users and the frequency of speaking in fitness social groups. The construction layer projects these features into a multidimensional portrait space composed of consumption dimensions, social dimensions, etc. to form a user portrait. The analysis layer slices according to preset parameters, such as the consumption amount range and social activity range, to obtain multiple behavior slices. The prediction layer determines the matching probability of these slices with preset behavior types such as "fashion consumer group" and "fitness enthusiast". If a slice has a high matching probability with "fitness enthusiast", it will be used as a candidate type. Based on this, the output layer generates behavioral preference information of users' preference for fitness-related activities and products.

[0065] Therefore, this method can deeply mine user behavior data, accurately build user portraits, and comprehensively and meticulously grasp user behavior preferences. Based on the generated behavior preference information, it provides strong support for applications such as personalized recommendations and precision marketing, improves the accuracy of recommendations and the pertinence of recommended information, and enhances user experience.

[0066] Further optionally, in the above steps, the matching probability between the plurality of behavior slices and the preset behavior types is predicted through the prediction layer to select the candidate behavior types for obtaining the multi-dimensional user portrait, including:

[0067] For the j-th preset behavior type Tj among multiple preset behavior types, calculate the matching degree P(Tj) between multiple behavior slices and the preset behavior type Tj; summarize the matching degrees between the multiple preset behavior types and the preset behavior type Tj to obtain the matching probability P(Tj) between the multiple preset behavior types and the preset behavior type Tj; select the candidate behavior type whose matching probability meets the preset screening condition from the multiple preset behavior types.

[0068] Among them, P(Tj)i between the i-th behavior slice Si and the preset behavior type Tj is expressed as the following formula: cosine(Si,Tj) is the cosine similarity between the behavior slice Si and the preset behavior type Tj, N(Si,Tj) is the matching probability vector obtained by processing the behavior slice Si and the preset behavior type Tj through the preset neural network, n is the number of behavior slices, and wi is the weight coefficient corresponding to the behavior slice Si.

[0069] In the prediction process, this method comprehensively considers multiple ways of measuring matching, fully utilizes the intuitiveness of cosine similarity and the ability of neural networks to mine potential relationships, and flexibly adjusts the contribution of different behavior slices to matching through weight coefficients, making the matching probability calculation more comprehensive and accurate. This helps to accurately identify candidate behavior types that are highly correlated with user behavior characteristics, and thus lays the foundation for generating accurate user behavior preference information, improves the effectiveness and accuracy of applications such as recommendation systems and marketing activities based on user portraits, better meets user personalized needs, and improves user experience and business efficiency.

[0070] Further optionally, in the above steps, the matching probability P(Tj) between the plurality of preset behavior types and the preset behavior type Tj is expressed as the following formula:

[0071]

[0072] Among them, μi and μj are the means of the behavior slice Si and another behavior slice Sj, respectively. is an exponential term based on the covariance matrix, which is used to represent the correlation between a behavior slice Si and another behavior slice Sj.

[0073] This formula further considers the correlation between behavior slices when calculating the matching probability between multiple preset behavior types and preset behavior types. First, the cosine similarity between the behavior slice and the preset behavior type and the matching probability vector obtained by the preset neural network processing are combined with the weight coefficient corresponding to the behavior slice to obtain the preliminary matching degree between the behavior slice and the preset behavior type. On this basis, the exponential term based on the covariance matrix is ​​introduced. The covariance matrix can reflect the relationship between the behavior slices. and are the means of the behavior slice and another behavior slice respectively. By calculating the exponential term, the degree of correlation between different behavior slices is measured. Behavior slices with high correlation have a more significant impact on the matching probability. The exponential term incorporates this correlation into the calculation of the matching probability, so that the final matching probability not only considers the matching of a single behavior slice with the preset behavior type, but also takes into account the internal connection between the behavior slices, thereby more comprehensively and accurately reflecting the degree of fit between the user's behavior characteristics and the preset behavior type.

[0074] This method of calculating matching probability can more finely characterize user behavior patterns. Considering the correlation between behavior slices avoids viewing each behavior slice in isolation, digs out the potential logical relationship between user behaviors, and improves the accuracy of matching probability. Selecting candidate behavior types based on more accurate matching probabilities helps to build more accurate user portraits, so that subsequent recommendations, analysis, and other applications based on user portraits can better meet users' actual needs, provide users with more personalized, high-quality services, and enhance the system's competitiveness in the field of user behavior analysis.

[0075] Further optionally, in the above steps, the multi-dimensional user portrait is adaptively segmented by using the pre-configured slice fitting parameters through the analysis layer to obtain a plurality of behavior slices corresponding to the multi-dimensional user portrait, including:

[0076] Based on the selection method of the slice interception position, the starting position and the ending position of each behavior slice are set; the computational complexity corresponding to each behavior slice under different analysis angle combinations is obtained, and the target analysis angle combination matching each behavior slice is selected based on the obtained computational complexity; based on the starting position and the ending position of each behavior slice, starting from the slice perspective corresponding to the target analysis angle combination, the multi-dimensional user portrait is adaptively segmented to obtain multiple behavior slices; wherein the adaptive segmentation is at least one of an equidistant division method, a peak division based on data distribution, and a cluster division.

[0077] Specifically, first, according to the pre-set slice interception position selection method, the starting and ending positions of each behavior slice in the multi-dimensional user portrait space are clarified. This setting is the basis for segmentation and determines the specific range of the behavior slice in the space. Next, the computational complexity of each behavior slice under different analysis angle combinations is calculated. Different analysis angle combinations have different requirements for data processing and analysis, and the computational complexity will also vary. By evaluating the computational complexity, we can understand the difficulty and resource consumption of different analysis angle combinations when processing specific behavior slices. Then, based on the computational complexity, the most matching target analysis angle combination is selected for each behavior slice to ensure that in subsequent analysis, the characteristics of the behavior slice can be effectively mined and a balance can be achieved between computing resources and efficiency. Finally, starting from the slice perspective corresponding to the selected target analysis angle combination, adaptive segmentation methods such as equidistant segmentation, peak segmentation based on data distribution, or clustering segmentation are used to segment the multi-dimensional user portrait to obtain multiple behavior slices. Equidistant partitioning is based on fixed intervals and is suitable for situations where data distribution is relatively uniform. Peak partitioning based on data distribution divides data at peaks and divides data at valleys according to the density of data distribution, which can better reflect the actual distribution of data. Cluster partitioning groups data points with similar behavioral characteristics into one category to form slices, highlighting differences in behavioral patterns.

[0078] It can be understood that the following is a specific introduction to the equal distance division method, the peak division based on data distribution, and the cluster division:

[0079] Equidistant division method: Each dimension of the multidimensional user portrait space is divided according to a fixed length or interval. In a two-dimensional space, for example, a fixed division length is set on the x-axis and y-axis respectively, and the entire space is cut into small areas of equal size. Each small area is a behavior slice. This division method does not consider the specific distribution of the data, and operates entirely based on the equidistant rule, so that each slice has the same length or range in each dimension.

[0080] This method is simple and intuitive, easy to understand and implement. The partitioning process does not require complex statistical analysis or calculation of the data. It can quickly partition the multidimensional space to obtain a certain number of uniformly sized behavioral slices, which facilitates the subsequent unified processing and analysis of each slice. It is suitable for scenarios where the data is relatively evenly distributed in each dimension, or there is no prior knowledge of the data distribution, and it is hoped that the space can be partitioned in a simple and unified way.

[0081] Peak division based on data distribution: First, perform statistical analysis on the user portrait data in each dimension, such as using methods such as kernel density estimation to obtain the probability density function on that dimension. By analyzing the probability density function, find the peak and valley of the data distribution. At the peak, the data distribution is denser, indicating that the user's behavior in these areas is more concentrated, so these areas are divided into finer slices to capture the user's behavior characteristics more carefully; at the valley, the data distribution is relatively sparse, and the user behavior is relatively small, so it is divided into coarser slices. In this way, according to the actual situation of data distribution, different division granularities are used at different locations, so that the slices can more accurately reflect the concentration of user behavior in different areas.

[0082] In this way, it is possible to flexibly divide the data according to its actual distribution, provide finer slices in data-dense areas, and more comprehensively capture user behavior characteristics. In data-sparse areas, the division granularity is appropriately relaxed to avoid too many meaningless divisions in areas with small amounts of data, thereby improving the efficiency and accuracy of the division and making the resulting behavior slices better reflect the true distribution of user behavior.

[0083] Therefore, this method is suitable for scenarios where data distribution is uneven and it is hoped that a more detailed description of the concentration of user behavior in different areas can be achieved, such as analyzing the distribution of user behavior in different time periods, different consumption amount ranges, and other dimensions.

[0084] Clustering: Use clustering algorithms, such as K-Means clustering algorithm, to process multi-dimensional user portrait data. Divide data points into different clusters based on metrics such as similarity or distance between them. Each cluster can be regarded as a behavior slice, and its boundary is determined by the range of data points in the cluster. The clustering algorithm automatically classifies data points with similar behavior characteristics into the same category, so that data points between different clusters have obvious behavioral differences, thus forming different behavior slices, which can well reflect the different behavior patterns of users.

[0085] Therefore, it is possible to automatically discover the inherent structure and regularity in the data, cluster data points with similar behavioral characteristics together, and form behavioral slices that can accurately reflect the different behavioral patterns of users, which helps to deeply understand the diversity and differences of user behaviors. There is no need to determine the position and shape of the slices in advance, and the division is completely based on the characteristics of the data itself, which has strong adaptability.

[0086] Therefore, it is suitable for scenarios where it is necessary to mine different user behavior patterns and the data volume is large and the data structure is relatively complex. For example, the comprehensive behavioral characteristics of users can be analyzed and users can be divided into different behavioral groups for targeted marketing or services.

[0087] Obviously, this adaptive segmentation method is highly flexible and targeted. By setting the slice position reasonably, it is ensured that each behavior slice can effectively cover valuable user behavior characteristics. The target analysis angle combination is selected based on the computational complexity to avoid resource waste or poor analysis results caused by improper analysis angles, and improve analysis efficiency and accuracy. By adopting a variety of adaptive segmentation methods, the most appropriate division method can be selected according to the characteristics of multi-dimensional user portrait data, so that the obtained behavior slices can more accurately reflect the characteristics of user behavior in different dimensions and scenarios. Whether it is evenly distributed data, data with obvious peaks, or data based on behavioral pattern clustering, it can be effectively processed. These behavioral slices provide a high-quality data foundation for the subsequent in-depth understanding of user behavior, accurate prediction, and personalized recommendations, and enhance the comprehensiveness and effectiveness of the entire system's user behavior analysis.

[0088] As an optional embodiment, in 103, based on the behavior preference information and the current geographical location of the target user, the behavior trajectory information of the target user within a preset time period in the future is predicted by building a behavior model, including:

[0089] Get the target user's current geographic location;

[0090] Based on the current geographical location and the mapping relationship between the geographical location type and the duration configuration strategy, set the future preset duration corresponding to the target user;

[0091] Obtain corresponding surrounding environment data based on the current geographical location;

[0092] Based on the configured future preset duration, the behavior preference information, the surrounding environment data and the current geographical location are input into the behavior construction model to predict the behavior trajectory to obtain the behavior trajectory information.

[0093] Specifically, we first need to obtain the current geographic location of the target user, which is the basis for subsequent analysis. After that, we set the future preset duration that fits the target user based on the mapping relationship between the geographic location type and the duration configuration strategy. For example, if the user is in a shopping mall, a shorter preset duration may be configured considering the general duration characteristics of activities in the mall; if the user is in an airport, a longer preset duration may be configured due to activities such as waiting for the flight. Doing so can more reasonably plan the prediction time range based on the common behavior duration patterns in different scenarios.

[0094] Next, obtain the surrounding environment data corresponding to the current location of the target user, such as the type of surrounding venues, traffic flow, etc. These environmental data have an important impact on user behavior. For example, if there is a movie theater nearby, the user may have a tendency to watch a movie.

[0095] Finally, the configured future preset duration is input into the behavior construction model together with the behavior preference information, surrounding environment data, and current geographical location. The model integrates this information and uses its algorithms and mechanisms to predict the behavior trajectory of the target user within the future preset duration, thereby obtaining the behavior trajectory information. In this way, factors such as the user's location, environment, and own preferences are fully considered, making the predicted behavior trajectory information more in line with the actual situation and providing an accurate basis for subsequent service recommendations.

[0096] As an optional embodiment, in 103, it is assumed that the behavior construction model includes at least the following structures: extraction layer, fusion layer, prediction layer, and screening layer. Based on this, in 104, based on the future preset time obtained by configuration, the behavior preference information, the surrounding environment data, and the current geographical location are input into the behavior construction model to predict the behavior trajectory, and the behavior trajectory information is obtained, including:

[0097] Through the extraction layer, the behavior preference features are extracted from the behavior preference information, and the surrounding environment features are extracted from the surrounding environment data; through the fusion layer, the behavior preference features, the surrounding environment features and the current geographical location are associated and fused to obtain the fused behavior features of the target user; the fused behavior features include each behavior preference feature associated with multiple combinations of different surrounding environment features and different geographical locations; through the prediction layer, a multi-branch decision layer is used to predict the changes of the fused behavior features in the future preset time length, and the predicted behavior nodes of the target user in the future preset time length and the corresponding predicted execution probability are obtained; through the screening layer, the predicted behavior nodes with the preset threshold of the predicted execution probability are constructed as the behavior trajectory information.

[0098] Extraction layer: This layer is mainly responsible for feature extraction. It accurately extracts behavior preference features from behavior preference information, which cover various preferences of users, such as shopping preferences, entertainment preferences, social preferences, etc. At the same time, it also extracts surrounding environment features from surrounding environment data, including surrounding facility information, human traffic, environment type, etc. The purpose of this layer is to convert raw data into more representative and analyzable feature data, providing a basis for subsequent processing.

[0099] Fusion layer: This layer associates and fuses the extracted behavior preference features, surrounding environment features, and current geographical location. Through a certain fusion strategy, different types of information are organically combined to form a fusion behavior feature of the target user. For example, the user's shopping preferences are combined with surrounding shopping mall information and current location information into a fusion behavior feature, which contains multiple different combinations, so that each behavior preference feature can be associated with different surrounding environment features and different geographical locations to reflect the comprehensive behavior characteristics that users may exhibit in different environments and locations.

[0100] Prediction layer: A multi-branch decision layer is used here to predict the changes of the fused behavior features in the future preset time. The multi-branch decision layer can be understood as having multiple decision branches, each branch predicts different situations, and can consider multiple possibilities and conditions. By analyzing the fused behavior features, various predicted behavior nodes that may appear in the target user in the future preset time are predicted, as well as the predicted execution probability corresponding to each behavior node. This layer uses its powerful decision-making ability to evaluate the user's behavior trends under the influence of different environments and preferences, and derive the probability of various behaviors occurring.

[0101] Screening layer: This layer is the last checkpoint to screen the prediction results. The prediction behavior nodes whose prediction execution probability does not reach the preset threshold are excluded, and only those prediction behavior nodes that reach or exceed the preset threshold are retained and constructed as behavior trajectory information. This ensures the reliability and effectiveness of the behavior trajectory information and avoids interference from too many low-probability behaviors.

[0102] Based on the above structure, first, in the extraction layer, data processing and feature extraction techniques are used, and methods such as data cleaning, feature encoding, and feature selection may be used to convert the original behavior preference information and surrounding environment data into more valuable behavior preference features and surrounding environment features. For example, the user's shopping records are converted into the user's preference features for different categories, and the store type, distance, and other information in the surrounding environment information are extracted as environmental features. Then, the fusion layer fuses the above features, and can use feature splicing, weighted fusion, and other methods to deeply fuse the behavior preference features, surrounding environment features, and current geographic location information. For example, the user's preferred sports category is combined with the sports venue information around the current location to form a fused behavior feature, showing the potential possibility of the user to perform related sports near the location. Then, the prediction layer uses the complex structure of the multi-branch decision layer and adopts machine learning algorithms or deep learning algorithms to predict the user's behavior within a preset time period in the future based on historical data and existing behavior patterns. For the fused behavior features, the multi-branch decision layer can perform different prediction paths based on different feature combinations and conditions, and calculate the predicted execution probability of each possible behavior node. For example, based on the user's shopping preferences, surrounding shopping mall information and location, the probability of entering a specific store within a preset time in the future is predicted. Finally, at the screening layer, based on the preset threshold, the most likely behavior nodes are screened out through simple probability comparison. For example, if the preset threshold is set to 60%, the behavior nodes with a predicted execution probability greater than or equal to 60% will be retained to form the final behavior trajectory information for subsequent recommendation or service decision-making.

[0103] Through this model structure and implementation method, we can analyze user behavior information from multiple perspectives, comprehensively consider user preferences, environment and location factors, and more accurately predict user behavior trajectories in the future, providing users with more targeted and practical services or recommendations.

[0104] Further optionally, the multi-branch decision layer predicts the fusion behavior characteristics, which is a relatively complex but orderly process. The following is its working principle and method:

[0105] First, the multi-branch decision layer will take the fused behavioral features as input. These fused behavioral features contain information such as the target user's behavioral preference characteristics, surrounding environment characteristics, and current geographical location. They are a comprehensive and rich feature set.

[0106] Then, based on these input features, the multi-branch decision layer uses its internal pre-trained multiple branch structures to perform analysis and processing in different dimensions or directions. Each branch may be responsible for processing a specific type of information or making predictions for a specific behavior pattern. For example, one branch may focus on time-related features to analyze the possible changing trends of behavior over time within a preset period of time in the future; another branch may focus on geographic location-related information to predict possible behavior nodes in different geographic locations; and another branch may focus on the interaction between behavioral preferences and the surrounding environment to determine the more likely behavior based on user preferences in a specific environment.

[0107] Each branch usually contains multiple neurons or processing units, which perform weighted summation, nonlinear transformation and other operations on the input fusion behavior features to extract higher-level feature representations. These operations are performed based on pre-trained weight parameters, which are continuously adjusted and learned through a large amount of training data and optimization algorithms, so that each branch can accurately capture the feature patterns related to its own field of responsibility.

[0108] Then, each branch will generate preliminary prediction results based on its own processing results. These results may include predicted behavior nodes and corresponding probability distributions under different assumptions or conditions. For example, one branch may predict that the user has a certain probability of appearing at a certain location and performing a certain behavior in a certain period of time in the future, while another branch may predict that the user has a different probability of performing another behavior.

[0109] Finally, the multi-branch decision layer will integrate and comprehensively evaluate the prediction results of each branch. This may involve weighted fusion and probability combination of prediction results of different branches to obtain a final prediction result that comprehensively considers multiple factors, that is, the predicted behavior node of the target user at a preset time in the future and the corresponding predicted execution probability. This comprehensive result is obtained after in-depth analysis and prediction of the fused behavior features from different angles by multiple branches, and can more comprehensively and accurately reflect the possible behavior trajectory and trend of the target user in the future.

[0110] In 104, based on the behavior trajectory information, a comprehensive service unit matching the current geographical location is recommended to the target user.

[0111] In an optional embodiment of 104, the operation is first carried out according to the previously predicted behavior trajectory information. The behavior trajectory information covers the behavior nodes of the target user whose predicted execution probability reaches the preset threshold within the preset time in the future, and these nodes reflect the possible behavior direction of the user. At the same time, combined with the current geographical location of the target user, the system aims to provide users with accurate and practical service recommendations.

[0112] The system will combine the behavior nodes in the behavior trajectory information with the actual situation of the current geographic location. For example, if the behavior trajectory information shows that the user may have a behavior node of buying sports equipment in the future, and there is a shopping mall near the current geographic location, the system will search for service units matching the behavior node of buying sports equipment within the scope of the shopping mall.

[0113] The comprehensive service units here include various types, such as offline products, that is, specific sports brand products in shopping malls; offline stores, such as professional sporting goods stores; offline service businesses, such as trying on and customizing sports equipment; if the user is in an online scene, there will also be online service businesses, such as online push of sports and fitness courses; and online products, such as the same sports equipment recommended online by sports apps.

[0114] In this way, the system recommends comprehensive service units to target users based on the matching of behavioral trajectory information with current geographic location, so that the recommended services are consistent with the user's behavioral tendencies and the actual location at the moment, greatly improving the accuracy and practicality of the recommendations, meeting the user's potential needs in actual scenarios, and improving the user experience. At the same time, it also helps to improve the resource utilization efficiency and commercial benefits of service providers.

[0115] The embodiment of the present application can obtain comprehensive and detailed insights into the user's real behavior patterns and preferences by acquiring user behavior data covering both online and offline, laying a solid foundation for subsequent precise analysis. The active analysis model deeply mines and analyzes behavioral preference characteristics from multiple dimensions, can accurately construct user portraits, outline user behavior preference information, and allow the system to have a more thorough understanding of the user. Based on the behavioral preference information and the user's current geographic location, the behavioral trajectory information within a preset time period in the future is predicted with the help of the behavior construction model, which can proactively predict the user's possible behavior, and filter out high-probability behavior nodes by setting a preset threshold for the prediction execution probability, making the prediction more reliable and practical. Finally, based on the behavioral trajectory information, a comprehensive service unit matching the current geographic location is recommended, which includes various service units matching the behavior nodes, achieving a high degree of fit between the service recommendation and the user's immediate needs and the scene in which they are located, greatly improving the accuracy and relevance of the recommendation, providing users with services that are more in line with actual needs, enhancing the user experience, and also helping to improve the operating efficiency and resource utilization of service providers.

[0116] Figure 2 A structural diagram of a large model recommendation system based on individual user terminal behavior provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the system comprises the following steps:

[0117] An acquisition unit, configured to acquire user behavior data of a target user; the user behavior data at least includes: online behavior data and offline behavior data;

[0118] An analysis unit is used to input the user behavior data into an active analysis model to perform user profiling and obtain the behavior preference information of the target user; the active analysis model is used to mine and analyze the behavior preference features in the user behavior data from multiple dimensions;

[0119] A prediction unit, configured to predict the behavior trajectory information of the target user within a preset time period in the future through a behavior construction model based on the behavior preference information and the current geographical location of the target user; the behavior trajectory information includes the behavior nodes of which the predicted execution probability of the target user within the preset time period in the future reaches a preset threshold;

[0120] A recommendation unit is used to recommend a comprehensive service unit that matches the current geographical location to the target user based on the behavior trajectory information; the comprehensive service unit includes a service unit that matches the behavior node in the behavior trajectory information, and the service unit is at least one of an offline product, an offline store, an offline service business, an online service business, and an online product.

[0121] Further optionally, the active analysis model at least includes the following structures: an extraction layer, a construction layer, an analysis layer, a prediction layer, and an output layer; an analysis unit, which inputs the user behavior data into the active analysis model to perform user profiling and obtain the behavior preference information of the target user, specifically used for:

[0122] Extracting multi-dimensional user behavior features from the user behavior data through an extraction layer;

[0123] By constructing the layers, according to the projection strategies corresponding to the respective dimensions, the multidimensional user behavior features are projected into the multidimensional portrait space to obtain the corresponding multidimensional user portrait; wherein each portrait coordinate point in the multidimensional user portrait corresponds to at least one behavior feature point in the multidimensional user behavior features;

[0124] The multi-dimensional user portrait is adaptively segmented by using the pre-configured slice fitting parameters through the analysis layer to obtain a plurality of behavior slices corresponding to the multi-dimensional user portrait; wherein the plurality of behavior slices are respectively used to represent the behavior change characteristics of the multi-dimensional user portrait under different analysis angles; the slice fitting parameters are used to represent the slice interception method of the multi-dimensional user portrait and the selection method of the slice interception position;

[0125] Through the prediction layer, the matching probability between multiple behavior slices and preset behavior types is predicted to select candidate behavior types for obtaining the multi-dimensional user portrait;

[0126] Through the output layer, the behavior preference information of the target user is generated according to the candidate behavior type.

[0127] Further optionally, the analysis unit predicts the matching probability between the plurality of behavior slices and the preset behavior type through the prediction layer to select the candidate behavior type of the multi-dimensional user portrait, specifically for:

[0128] For the j-th preset behavior type Tj among the multiple preset behavior types, calculating the matching degree P(Tj) between the multiple behavior slices and the preset behavior type Tj;

[0129] Among them, P(Tj)i between the i-th behavior slice Si and the preset behavior type Tj is expressed as the following formula: cosine(Si,Tj) is the cosine similarity between the behavior slice Si and the preset behavior type Tj, N(Si,Tj) is the matching probability vector obtained by processing the behavior slice Si and the preset behavior type Tj through the preset neural network, n is the number of behavior slices, wi is the weight coefficient corresponding to the behavior slice Si;

[0130] Summarize the matching degrees between the multiple preset behavior types and the preset behavior type Tj to obtain the matching probability P(Tj) between the multiple preset behavior types and the preset behavior type Tj;

[0131] A candidate behavior type whose matching probability reaches a preset screening condition is selected from multiple preset behavior types.

[0132] Further optionally, the matching probability P(Tj) between the plurality of preset behavior types and the preset behavior type Tj is expressed as the following formula:

[0133]

[0134] Among them, μi and μj are the means of the behavior slice Si and another behavior slice Sj, respectively. is an exponential term based on the covariance matrix, which is used to represent the correlation between a behavior slice Si and another behavior slice Sj.

[0135] Further optionally, the analysis unit, through the analysis layer, uses pre-configured slice fitting parameters to adaptively segment the multi-dimensional user portrait to obtain a plurality of behavior slices corresponding to the multi-dimensional user portrait, specifically for:

[0136] Based on the selection method of the slice interception position, set the starting position and the ending position of each behavior slice;

[0137] Obtain the computational complexity corresponding to each behavior slice under different analysis angle combinations, and select a target analysis angle combination matching each behavior slice based on the obtained computational complexity;

[0138] Based on the starting position and the ending position of each behavior slice, the corresponding slice perspectives are combined from the perspective of target analysis to adaptively segment the multi-dimensional user portrait to obtain multiple behavior slices; wherein the adaptive segmentation is at least one of an equidistant division method, a peak division based on data distribution, and a cluster division.

[0139] Further optionally, the prediction unit, based on the behavior preference information and the current geographical location of the target user, predicts the behavior trajectory information of the target user within a preset time period in the future through a behavior construction model, specifically for:

[0140] Get the target user's current geographic location;

[0141] Based on the current geographical location and the mapping relationship between the geographical location type and the duration configuration strategy, set the future preset duration corresponding to the target user;

[0142] Obtain corresponding surrounding environment data based on the current geographical location;

[0143] Based on the configured future preset duration, the behavior preference information, the surrounding environment data and the current geographical location are input into the behavior construction model to predict the behavior trajectory to obtain the behavior trajectory information.

[0144] Further optionally, the behavior construction model includes at least the following structures: an extraction layer, a fusion layer, a prediction layer, and a screening layer; a prediction unit, based on the configured future preset time, inputs the behavior preference information, the surrounding environment data, and the current geographical location into the behavior construction model to predict the behavior trajectory, and obtains the behavior trajectory information, which is specifically used for:

[0145] extracting behavior preference features from the behavior preference information and extracting surrounding environment features from the surrounding environment data through an extraction layer;

[0146] The behavior preference feature, the surrounding environment feature and the current geographical location are associated and fused through the fusion layer to obtain the fused behavior feature of the target user; the fused behavior feature includes each behavior preference feature associated with multiple combinations of different surrounding environment features and different geographical locations;

[0147] Through the prediction layer, a multi-branch decision layer is used to predict the change of the fusion behavior feature in the future preset time length, and obtain the predicted behavior node of the target user in the future preset time length and the corresponding predicted execution probability;

[0148] Through the screening layer, the predicted behavior nodes with the preset threshold of the predicted execution probability are constructed as the behavior trajectory information.

[0149] See also Figure 3 , Figure 3 This is a schematic diagram of an electronic device provided in an embodiment of the present application. Figure 3 As shown, an embodiment of the present application provides an electronic device 500, including a memory 510, a processor 520, and a computer program 511 stored in the memory 510 and executable on the processor 520. When the processor 520 executes the computer program 511, the aforementioned embodiment is implemented.

[0150] See also Figure 4 , Figure 4 A schematic diagram of an embodiment of a computer-readable storage medium provided in an embodiment of the present application. Figure 4 As shown, this embodiment provides a computer-readable storage medium 600 on which a computer program 611 is stored. When the computer program 611 is executed by a processor, the aforementioned embodiment is implemented.

[0151] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0152] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0153] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0154] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0155] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0156] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0157] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A large model recommendation method based on individual user behavior, characterized in that: The method at least comprises: Obtaining user behavior data of the target user; the user behavior data at least includes: online behavior data and offline behavior data; The user behavior data is input into an active analysis model to perform user profiling, thereby obtaining the behavior preference information of the target user; the active analysis model is used to mine and analyze the behavior preference features in the user behavior data from multiple dimensions; Based on the behavior preference information and the current geographic location of the target user, predict the behavior trajectory information of the target user within a preset time period in the future through a behavior construction model; the behavior trajectory information includes the behavior nodes of the target user whose predicted execution probability reaches a preset threshold within the preset time period in the future; Based on the behavior trajectory information, a comprehensive service unit matching the current geographical location is recommended to the target user; the comprehensive service unit includes a service unit matching the behavior node in the behavior trajectory information, and the service unit is at least one of an offline product, an offline store, an offline service business, an online service business, and an online product.

2. The large model recommendation method based on individual user terminal behavior according to claim 1 is characterized in that: The active analysis model at least includes the following structures: extraction layer, construction layer, analysis layer, prediction layer, and output layer; The inputting the user behavior data into the active analysis model to perform user profiling and obtain the behavior preference information of the target user includes: Extracting multi-dimensional user behavior features from the user behavior data through an extraction layer; By constructing the layers, according to the projection strategies corresponding to the respective dimensions, the multidimensional user behavior features are projected into the multidimensional portrait space to obtain the corresponding multidimensional user portrait; wherein each portrait coordinate point in the multidimensional user portrait corresponds to at least one behavior feature point in the multidimensional user behavior features; The multi-dimensional user portrait is adaptively segmented by using the pre-configured slice fitting parameters through the analysis layer to obtain a plurality of behavior slices corresponding to the multi-dimensional user portrait; wherein the plurality of behavior slices are respectively used to represent the behavior change characteristics of the multi-dimensional user portrait under different analysis angles; the slice fitting parameters are used to represent the slice interception method of the multi-dimensional user portrait and the selection method of the slice interception position; Through the prediction layer, the matching probability between multiple behavior slices and preset behavior types is predicted to select candidate behavior types for obtaining the multi-dimensional user portrait; Through the output layer, the behavior preference information of the target user is generated according to the candidate behavior type.

3. The large model recommendation method based on individual user terminal behavior according to claim 2 is characterized in that: The prediction layer predicts the matching probability between the plurality of behavior slices and the preset behavior types to select candidate behavior types for the multi-dimensional user portrait, including: For the j-th preset behavior type Tj among the multiple preset behavior types, calculating the matching degree P(Tj) between the multiple behavior slices and the preset behavior type Tj; Among them, P(Tj)i between the i-th behavior slice Si and the preset behavior type Tj is expressed as the following formula: cosine(Si,Tj) is the cosine similarity between the behavior slice Si and the preset behavior type Tj, N(Si,Tj) is the matching probability vector obtained by processing the behavior slice Si and the preset behavior type Tj through the preset neural network, n is the number of behavior slices, wi is the weight coefficient corresponding to the behavior slice Si; Summarize the matching degrees between the multiple preset behavior types and the preset behavior type Tj to obtain the matching probability P(Tj) between the multiple preset behavior types and the preset behavior type Tj; A candidate behavior type whose matching probability reaches a preset screening condition is selected from multiple preset behavior types.

4. The large model recommendation method based on individual user terminal behavior according to claim 3 is characterized in that: The matching probability P(Tj) between multiple preset behavior types and the preset behavior type Tj is expressed as the following formula: Among them, μi and μj are the means of the behavior slice Si and another behavior slice Sj, respectively. is an exponential term based on the covariance matrix, which is used to represent the correlation between a behavior slice Si and another behavior slice Sj.

5. According to claim 2, the large model recommendation method based on individual user terminal behavior is characterized in that: The analysis layer uses pre-configured slice fitting parameters to adaptively segment the multi-dimensional user portrait to obtain multiple behavior slices corresponding to the multi-dimensional user portrait, including: Based on the selection method of the slice interception position, set the starting position and the ending position of each behavior slice; Obtain the computational complexity corresponding to each behavior slice under different analysis angle combinations, and select a target analysis angle combination matching each behavior slice based on the obtained computational complexity; Based on the starting position and the ending position of each behavior slice, the corresponding slice perspectives are combined from the perspective of target analysis to adaptively segment the multi-dimensional user portrait to obtain multiple behavior slices; wherein the adaptive segmentation is at least one of an equidistant division method, a peak division based on data distribution, and a cluster division.

6. The large model recommendation method based on individual user terminal behavior according to claim 1 is characterized in that: The predicting of the target user's behavior trajectory information within a preset time period in the future by building a behavior model based on the behavior preference information and the target user's current geographical location includes: Get the target user's current geographic location; Based on the current geographical location and the mapping relationship between the geographical location type and the duration configuration strategy, set the future preset duration corresponding to the target user; Obtain corresponding surrounding environment data based on the current geographical location; Based on the configured future preset duration, the behavior preference information, the surrounding environment data and the current geographical location are input into the behavior construction model to predict the behavior trajectory to obtain the behavior trajectory information.

7. The large model recommendation method based on individual user terminal behavior according to claim 1 is characterized in that: The behavior construction model at least includes the following structures: extraction layer, fusion layer, prediction layer, and screening layer; The future preset time obtained based on the configuration, the behavior preference information, the surrounding environment data and the current geographical location are input into the behavior construction model to predict the behavior trajectory, and the behavior trajectory information is obtained, including: extracting behavior preference features from the behavior preference information and extracting surrounding environment features from the surrounding environment data through an extraction layer; The behavior preference feature, the surrounding environment feature and the current geographical location are associated and fused through the fusion layer to obtain the fused behavior feature of the target user; the fused behavior feature includes each behavior preference feature associated with multiple combinations of different surrounding environment features and different geographical locations; Through the prediction layer, a multi-branch decision layer is used to predict the change of the fusion behavior feature in the future preset time, and obtain the predicted behavior node of the target user in the future preset time and the corresponding predicted execution probability; Through the screening layer, the predicted behavior nodes with the preset threshold of the predicted execution probability are constructed as the behavior trajectory information.

8. A large model recommendation system based on individual user behavior, characterized in that: The system comprises at least the following units: An acquisition unit, used to acquire user behavior data of a target user; The user behavior data at least includes: online behavior data and offline behavior data; An analysis unit is used to input the user behavior data into an active analysis model to perform user profiling and obtain the behavior preference information of the target user; the active analysis model is used to mine and analyze the behavior preference features in the user behavior data from multiple dimensions; A prediction unit, configured to predict the behavior trajectory information of the target user within a preset time period in the future through a behavior construction model based on the behavior preference information and the current geographical location of the target user; the behavior trajectory information includes the behavior nodes of which the predicted execution probability of the target user within the preset time period in the future reaches a preset threshold; A recommendation unit is used to recommend a comprehensive service unit that matches the current geographical location to the target user based on the behavior trajectory information; the comprehensive service unit includes a service unit that matches the behavior node in the behavior trajectory information, and the service unit is at least one of an offline product, an offline store, an offline service business, an online service business, and an online product.

9. An electronic device, characterized in that: including a memory for storing a computer software program; A processor is used to read and execute the computer software program, thereby realizing the functions of each component part in the large model recommendation system based on personal user terminal behavior as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium, characterized in that: The storage medium stores a computer software program, which, when executed by a processor, implements the functions of each component part of the large model recommendation system based on individual user-side behavior as described in any one of claims 1 to 7.

Images