A method for dividing traffic in an Internet marketing system

Abstract

The present invention discloses a method for dividing traffic of an Internet marketing system, and relates to the technical field of Internet marketing, including integrating multimodal information and environmental data, laying a foundation for accurate analysis through data cleaning and conversion, using sentiment analysis to build a dynamic user sentiment model, predicting sentiment trends in real time, enhancing the emotional resonance of recommended content, mapping environmental data to scene categories, combining machine learning to achieve intelligent push according to the scene where the user is located, a multimodal neural network model performs cross-modal matching according to user emotions and content preferences, improving recommendation accuracy, establishing a multi-objective optimization model, comprehensively considering user status and system resources, dynamically adjusting traffic distribution, and achieving efficient promotion of personalized content.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet marketing, and in particular to a method for dividing traffic of an Internet marketing system. Background Art

[0002] In recent years, the rise of big data and artificial intelligence technologies has brought unprecedented opportunities to Internet marketing systems, but challenges still exist in the process of achieving precision marketing. The integration and processing of multimodal information is still imperfect. Although data sources are rich and diverse, due to different formats and uneven quality, how to effectively clean and convert these data to make them effective inputs for model analysis is a major technical problem. The accuracy of user sentiment analysis and sentiment modeling needs to be improved. User emotional fluctuations have a significant impact on consumer decisions, but existing sentiment analysis technologies often rely on limited sentiment lexicons, making it difficult to capture subtle emotional changes in complex situations, resulting in the constructed user sentiment model not being dynamic and accurate enough. Summary of the invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides a traffic segmentation method for an Internet marketing system to solve the problems of effective integration of multimodal information and environmental data in Internet marketing, insufficient dynamic perception of user emotions, and lack of personalized accuracy in traffic allocation strategies.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a method for segmenting traffic in an Internet marketing system, which includes integrating multimodal information and environmental data, performing data cleaning and converting data types;

[0007] Perform sentiment analysis based on the converted multimodal information and build a dynamic user sentiment model;

[0008] Extract environmental data and apply machine learning algorithms to map them to different scene categories;

[0009] Build a multimodal neural network model to accurately match user preferences for different types of content;

[0010] Establish a multi-objective optimization model and dynamically adjust the traffic distribution strategy.

[0011] As a preferred solution of the method for dividing traffic of the Internet marketing system of the present invention, wherein: the multimodal information includes text, image and video;

[0012] Environmental data includes geographic location, weather conditions, time of day, and holidays.

[0013] As a preferred solution of the Internet marketing system traffic segmentation method of the present invention, the converted multimodal information refers to the modal features extracted from the diversified information of text, image and video, and the expression is:

[0014] ;

[0015] in, Indicates The overall feature vector of samples after comprehensive processing, i represents the sample index, M represents the number of feature types, Indicates The weight of the class feature, Indicates The sample in Eigenvectors on class features.

[0016] It should be noted that by integrating diversified information and performing refined feature extraction, it not only promotes the possibility of personalized processing and accurate recommendations, but also improves the pertinence and accuracy of feature extraction, so that subsequent processing can focus more on the core part of the information, thereby improving processing efficiency and effectiveness.

[0017] As a preferred solution of the method for segmenting the traffic of the Internet marketing system of the present invention, the sentiment analysis based on the converted multimodal information refers to performing sentiment scoring on each sample based on the extracted modal features, and the expression is:

[0018] ;

[0019] in, Indicates The sentiment score of samples, represents the sentiment score variable, represents the minimum value of the sentiment score, represents the maximum value of the sentiment score, represents the prior probability distribution, represents the variance of the normal distribution, Represents the comprehensive feature vector The converted value of sentiment intensity extracted from Represents the normal distribution function.

[0020] It should be noted that the use of the normal distribution function combined with the prior probability distribution to calculate the sentiment score improves the robustness and adaptability of the score, and can better handle the diversity of individual emotional expressions. By setting the minimum and maximum values ​​of the sentiment score, the sentiment analysis results are consistent and interpretable in different samples and scenarios, which facilitates sentiment analysis and comparison across scenarios.

[0021] As a preferred solution of the method for dividing the traffic of the Internet marketing system described in the present invention, the construction of a dynamic user emotion model refers to taking the emotion score as the instant emotion and combining the emotion state at the previous moment to predict the emotion state at the current moment, and the expression is:

[0022] ;

[0023] in, represents the user's emotional state at time t, Indicates the weight of the impact of historical emotional state on current emotion, represents the emotional state at the previous time point t-1, β represents the influence factor of environmental factors on emotions, represents the environmental impact value, Indicates the specific location at the current moment. Indicates weather conditions. Indicates time information;

[0024] Will The value range of is mapped to [−1,1] to compare the emotional states at different time points. Positive values ​​represent positive emotional trends, negative values ​​represent negative emotional trends, and values ​​near zero indicate relatively stable emotions.

[0025] It should be noted that by using the emotional score as the real-time emotion, the user's current emotional state can be captured instantly. By combining the emotional state of the previous moment and giving it an influence weight, the emotional consistency and changing trend can be taken into consideration, and the emotional state can be predicted more accurately. Through environmental factors and their corresponding environmental impact values, a comprehensive assessment of the user's environment can be achieved, and the emotional state is mapped to the [-1,1] interval, which enables the quantification and intuitive comparison of the emotional state, making it easier for users to understand their own emotional change trends. At the same time, it also provides intuitive data support for emotional management and personalized services.

[0026] As a preferred solution of the method for dividing the traffic of the Internet marketing system of the present invention, the following specific steps are taken to extract environmental data and map it to different scene categories using a machine learning algorithm:

[0027] Obtain the GPS coordinates of the user's device in real time, real-time weather data for the user's location, record the exact time of data collection, and query the public holiday database based on the current date to determine whether it is a holiday;

[0028] Integrate the acquired environmental data to form a comprehensive environmental feature vector;

[0029] Based on the comprehensive environmental feature vector, a scene recognition model is constructed and the output probability distribution is expressed as:

[0030] ;

[0031] in, Indicates at a point in time User in scene The probability of represents the comprehensive environment feature vector, Indicates a specific scene. Indicates a specific scene The variance of Indicates a specific scene The mean of represents the scene category label, Representation scene The mean of the eigenvectors of Representation scene The variance of the eigenvector of

[0032] According to the probability distribution of scene categories output by the model, a threshold Tc is set for each scene;

[0033] Design a multi-label assignment algorithm, the expression is:

[0034] ;

[0035] in, Indicates that in a given environment Under the condition of , the probability that the user is in scene c is, represents the set of scene labels assigned to the user at time t, and Tc represents the assignment threshold of scene c;

[0036] Traverse all scene categories and calculate the probability of each scene. If the probability exceeds the set threshold, add the scene to the user's scene label list.

[0037] It should be noted that by acquiring the GPS coordinates of the user's device, real-time weather data, data collection time, and querying the holiday database in real time, the accuracy and real-time performance of scene recognition are improved, providing a solid data foundation for subsequent personalized services, safety warnings, activity recommendations and other applications. The machine learning algorithm is used to build a scene recognition model and output the probability distribution, so that the system can flexibly adapt to various complex environmental conditions and predict the user's scene category and its probability. By setting the threshold Tc of the scene category probability distribution and designing a multi-label assignment algorithm, flexible and accurate assignment of user scene labels is achieved.

[0038] As a preferred solution of the method for dividing the traffic of the Internet marketing system described in the present invention, the multimodal neural network model is constructed to achieve accurate matching of user preferences for different types of content, which means forming multimodal features based on text, picture and information features, combining the user's emotional state and the fused multimodal features, performing cross-modal modeling, and generating a preference prediction model, which is expressed as:

[0039] ;

[0040] in, Indicates the user at a point in time and multimodal features Under the condition of degree of preference, is the multimodal fusion feature vector, represents the model parameter vector associated with scene category c, Indicates any specific category The latent feature vector of preferences, represents the user’s emotional state vector at time point t, An exponential function representing the sum of all scenarios of a specific category;

[0041] The value range of the preference probability distribution is [0,1], which indicates the user's preference for content category c. The larger the value, the stronger the preference.

[0042] It should be noted that by integrating text, image and information features to form a multimodal feature vector, a multi-angle understanding of user content consumption habits is achieved. Combining the user's emotional state with multimodal features, content preference matching that is sensitive to the user's psychological state is achieved. Through cross-modal modeling technology, a preference prediction model is generated, which realizes a direct mapping from multi-source heterogeneous data to user preference prediction. The preference prediction model outputs the preference probability distribution of content category c, which realizes a quantitative evaluation of the user's preference intensity.

[0043] As a preferred solution of the method for dividing the traffic of the Internet marketing system described in the present invention, a multi-objective optimization model is established to dynamically adjust the traffic allocation strategy. The specific steps are as follows:

[0044] Assign weights to the user emotion model, scene label importance, and content preference features, define the weighted fusion function, build a multi-objective optimization model, and obtain the user's comprehensive score, which is expressed as:

[0045] ;

[0046] in, represents the comprehensive score of user u, The weight factor representing the emotional state, represents the scene relevance weight factor, represents the content preference matching weight factor, The gain coefficient representing the importance of scene labels to recommended content, represents the weight of the content preference feature vector in preference matching, Indicates the user at a point in time Any scene in the set of all scene tags assigned , Indicates the weight of the impact of historical emotional state on current emotion;

[0047] Prioritize content based on the user's comprehensive score, and display the content with the highest score first;

[0048] Based on the user's current status and system resources, traffic allocation decisions are made to recommend personalized content to users through the most appropriate channels.

[0049] It should be noted that by assigning weights to user emotion models, scene label importance, and content preference characteristics, we have achieved a detailed consideration of the multi-dimensional user personalized needs, built a multi-objective optimization model, and used a weighted fusion function to obtain the user's comprehensive score, which simplified the decision-making process. The content recommendation system can efficiently and intuitively identify which content is most attractive to specific users, and then make fast and accurate recommendation decisions. It prioritizes content according to the user's comprehensive score, improves user click-through rate and satisfaction, reduces the time cost of users looking for content of interest, and enhances the smoothness and pleasure of the user experience. It makes traffic allocation decisions based on the user's current status and system resource conditions, and recommends personalized content through the most appropriate channels. While ensuring the quality of recommendations, it optimizes server load and reduces operating costs.

[0050] In a second aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the Internet marketing system traffic segmentation method described in the first aspect of the present invention is implemented.

[0051] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the Internet marketing system traffic segmentation method as described in the first aspect of the present invention is implemented.

[0052] The beneficial effects of the present invention are: integrating multimodal information and environmental data, laying the foundation for accurate analysis through data cleaning and conversion, using sentiment analysis to build a dynamic user emotion model, predicting emotional trends in real time, enhancing the emotional resonance of recommended content, mapping environmental data to scene categories, combining machine learning to achieve intelligent push based on the user's scene, multimodal neural network models perform cross-modal matching based on user emotions and content preferences, improving recommendation accuracy, establishing a multi-objective optimization model, comprehensively considering user status and system resources, dynamically adjusting traffic distribution, and achieving efficient promotion of personalized content. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0054] Figure 1 This is a flow chart of the Internet marketing system traffic segmentation method in Example 1.

[0055] Figure 2 This is a flow chart of the preference probability distribution in Example 1. DETAILED DESCRIPTION

[0056] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0057] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0058] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0059] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a method for dividing traffic of an Internet marketing system, comprising the following steps:

[0060] S1, integrate multimodal information and environmental data, perform data cleaning and data type conversion;

[0061] Perform sentiment analysis based on the converted multimodal information and build a dynamic user sentiment model;

[0062] Extract environmental data and apply machine learning algorithms to map them to different scene categories;

[0063] Build a multimodal neural network model to accurately match user preferences for different types of content;

[0064] Establish a multi-objective optimization model and dynamically adjust the traffic distribution strategy.

[0065] S2, multimodal information includes text, images, and videos;

[0066] Environmental data includes geographic location, weather conditions, time of day, and holidays.

[0067] S3. The converted multimodal information refers to the modal features extracted from the diversified information of text, image, and video, and the expression is:

[0068] ;

[0069] in, Indicates The overall feature vector of samples after comprehensive processing, i represents the sample index, M represents the number of feature types, Indicates The weight of the class feature, Indicates The sample in Eigenvectors on class features.

[0070] Furthermore, the specific steps for extracting features from diversified information such as text, images, and videos are as follows:

[0071] Use the BERT model to encode the processed text and obtain the embedded representation of the text;

[0072] ResNet is used to adjust the image size, extract the high-level feature map of the image, and convert the feature map into a fixed-length feature vector that represents the overall characteristics of the image;

[0073] The key frame features are extracted using 3D and combined with the audio features, which are then fused through a time series model to obtain the overall feature vector of the video.

[0074] S4. Sentiment analysis based on the converted multimodal information refers to scoring the sentiment of each sample based on the extracted modal features. The expression is:

[0075] ;

[0076] in, Indicates The sentiment score of samples, represents the sentiment score variable, represents the minimum value of the sentiment score, represents the maximum value of the sentiment score, represents the prior probability distribution, represents the variance of the normal distribution, Represents the comprehensive feature vector The converted value of sentiment intensity extracted from Represents the normal distribution function.

[0077] Furthermore, each sample refers to all modal information of each user's behavior.

[0078] Furthermore, based on the sentiment score, in the sentiment analysis stage, the value range of the sentiment score is usually [−∞,+∞], where positive numbers represent positive emotions, negative numbers represent negative emotions, and the absolute value reflects the intensity of the emotion.

[0079] S5. Building a dynamic user emotion model means taking the emotion score as the immediate emotion and combining it with the emotion state at the previous moment to predict the emotion state at the current moment. The expression is:

[0080] ;

[0081] in, represents the user's emotional state at time t, Indicates the weight of the impact of historical emotional state on current emotion, represents the emotional state at the previous time point t-1, β represents the influence factor of environmental factors on emotions, represents the environmental impact value, Indicates the specific location at the current moment. Indicates weather conditions. Indicates time information;

[0082] Will The value range of is mapped to [−1,1] to compare the emotional states at different time points. Positive values ​​represent positive emotional trends, negative values ​​represent negative emotional trends, and values ​​near zero indicate relatively stable emotions.

[0083] S6. Extract environmental data and apply machine learning algorithms to map them to different scene categories. The specific steps are as follows:

[0084] Obtain the GPS coordinates of the user's device in real time, real-time weather data for the user's location, record the exact time of data collection, and query the public holiday database based on the current date to determine whether it is a holiday;

[0085] Integrate the acquired environmental data to form a comprehensive environmental feature vector;

[0086] Based on the comprehensive environmental feature vector, a scene recognition model is constructed and the output probability distribution is expressed as:

[0087] ;

[0088] in, Indicates at a point in time User in scene The probability of represents the comprehensive environment feature vector, Indicates a specific scene. Indicates a specific scene The variance of Indicates a specific scene The mean of represents the scene category label, Representation scene The mean of the eigenvectors of Representation scene The variance of the eigenvector of

[0089] According to the probability distribution of scene categories output by the model, a threshold Tc is set for each scene;

[0090] Design a multi-label assignment algorithm, the expression is:

[0091] ;

[0092] in, Indicates that in a given environment Under the condition of , the probability that the user is in scene c is, represents the set of scene labels assigned to the user at time t, and Tc represents the assignment threshold of scene c;

[0093] Traverse all scene categories and calculate the probability of each scene. If the probability exceeds the set threshold, add the scene to the user's scene label list.

[0094] Furthermore, we set the environmental feature vector of user U1 at a certain time point t , the scene category probability distribution of the model output is as follows:

[0095] Weekday commute (Scenario A): =0.75;

[0096] Weekend leisure (Scenario B): =0.45;

[0097] Severe weather at home (Scenario C): =0.6.

[0098] According to the set threshold, label allocation is performed:

[0099] For scenario A, since 0.75>0.7, the label “weekday commuting” is assigned;

[0100] For scenario B, since 0.45 < 0.50, the “weekend leisure” label is not assigned;

[0101] For scenario C, since 0.6 < 0.8, the “bad weather at home” label is not assigned;

[0102] Therefore, the scene label set assigned to user U1 at time point t is: weekday commuting.

[0103] S7. Constructing a multimodal neural network model to achieve accurate matching of user preferences for different types of content means forming multimodal features based on text, image and information features, combining the user's emotional state and the fused multimodal features, performing cross-modal modeling, and generating a preference prediction model, which is expressed as:

[0104] ;

[0105] in, Indicates the user at a point in time and multimodal features Under the condition of degree of preference, is the multimodal fusion feature vector, represents the model parameter vector associated with scene category c, Indicates any specific category The latent feature vector of preferences, represents the user’s emotional state vector at time point t, An exponential function representing the sum of all scenarios of a specific category;

[0106] The value range of the preference probability distribution is [0,1], which indicates the user's preference for content category c. The larger the value, the stronger the preference.

[0107] S8. Establish a multi-objective optimization model and dynamically adjust the traffic allocation strategy. The specific steps are as follows:

[0108] Assign weights to the user emotion model, scene label importance, and content preference features, define the weighted fusion function, build a multi-objective optimization model, and obtain the user's comprehensive score, which is expressed as:

[0109] ;

[0110] in, represents the comprehensive score of user u, The weight factor representing the emotional state, represents the scene relevance weight factor, represents the content preference matching weight factor, The gain coefficient representing the importance of scene labels to recommended content, represents the weight of the content preference feature vector in preference matching, Indicates the user at a point in time Any scene in the set of all scene tags assigned , Indicates the weight of the impact of historical emotional state on current emotion;

[0111] Prioritize content based on the user's comprehensive score, and display the content with the highest score first;

[0112] Based on the user's current status and system resources, traffic allocation decisions are made to recommend personalized content to users through the most appropriate channels.

[0113] This embodiment also provides a computer device, which is suitable for the case of the Internet marketing system traffic segmentation method, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the Internet marketing system traffic segmentation method proposed in the above embodiment.

[0114] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0115] The present embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for implementing traffic segmentation of an Internet marketing system as proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, disk or optical disk.

[0116] In summary, the present invention lays the foundation for accurate analysis by integrating multimodal information and environmental data, data cleaning and conversion, uses sentiment analysis to build a dynamic user emotion model, predicts emotional trends in real time, and enhances the emotional resonance of recommended content. Environmental data is mapped to scene categories, combined with machine learning to achieve intelligent push based on the user's scene. The multimodal neural network model performs cross-modal matching based on user emotions and content preferences to improve recommendation accuracy. A multi-objective optimization model is established, which comprehensively considers user status and system resources, dynamically adjusts traffic distribution, and achieves efficient promotion of personalized content.

[0117] Example 2

[0118] Referring to Table 1, which is the second embodiment of the present invention, experimental simulation data of the traffic segmentation method of the Internet marketing system is provided to further verify the present invention.

[0119] By integrating multiple information of text, images, and videos, combined with environmental factors such as geographic location, weather, time, and holidays, and taking the dynamic user emotion model as the core, accurate content matching and traffic optimization allocation are achieved. Six virtual users (U1 to U6) are selected as test subjects. Each user has their own unique information flow sample set, covering interaction data in different situations in daily life, as shown in the following table:

[0120] Table 1 Experimental record table

[0121]

[0122] By analyzing the table data, users U1 and U3 have the highest comprehensive scores, thanks to their high sentiment scores and content preference matching in travel and outdoor activities scenarios, respectively. U1's strong preference for travel content is consistent with the "travel" scenario label identified by the system, reflecting the accuracy of the model in sentiment understanding and scenario matching. In contrast, although U2 and U5 had low emotional states in specific situations (such as rainy weather and home environment), the system successfully recommended content that matched their preferences at the time, such as reading and music, which proved the model's adaptability in different emotional states and scenarios. By comparing the cases of U4 and U6, it can be seen that the system can also effectively distinguish the differences in user preferences in family and nightlife scenarios. U4 prefers movie content, while U6 prefers entertainment content, which further proves the innovation and advantages of the fusion of multimodal information and environmental data in improving the personalization and accuracy of recommendation systems.

[0123] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for dividing traffic in an Internet marketing system, characterized by: include, Integrate multimodal information and environmental data, perform data cleaning and convert data types; Perform sentiment analysis based on the converted multimodal information and build a dynamic user sentiment model; Extract environmental data and apply machine learning algorithms to map them to different scene categories; Build a multimodal neural network model to accurately match user preferences for different types of content; Establish a multi-objective optimization model to dynamically adjust the traffic allocation strategy; The construction of a dynamic user emotion model refers to taking the emotion score as the instant emotion and combining it with the emotion state at the previous moment to predict the emotion state at the current moment. The expression is: ； in, represents the user's emotional state at time t, Indicates the weight of the impact of historical emotional state on current emotion, represents the emotional state at the previous time point t-1, β represents the influence factor of environmental factors on emotions, represents the environmental impact value, Indicates the specific location at the current moment. Indicates weather conditions. Indicates time information; Will The value range of is mapped to [−1,1], and the emotional states at different time points are compared. Positive values ​​represent positive emotional trends, negative values ​​represent negative emotional trends, and values ​​near zero indicate relatively stable emotions. The construction of a multimodal neural network model to achieve accurate matching of user preferences for different types of content refers to forming multimodal features based on text, image and information features, combining the user's emotional state and the fused multimodal features, performing cross-modal modeling, and generating a preference prediction model, which is expressed as: ； in, Indicates the user at a point in time and multimodal features Under the condition of , the preference degree for scene category c is, is the multimodal fusion feature vector, represents the model parameter vector associated with scene category c, Indicates any specific category The latent feature vector of preferences, represents the user’s emotional state vector at time point t, An exponential function representing the sum of all scenarios of a specific category; The value range of the preference probability distribution is [0,1], which indicates the user's preference for content category c. The larger the value, the stronger the preference. Establish a multi-objective optimization model and dynamically adjust the traffic allocation strategy. The specific steps are as follows: Assign weights to the user emotion model, scene label importance, and content preference features, define the weighted fusion function, build a multi-objective optimization model, and obtain the user's comprehensive score, which is expressed as: ； in, represents the comprehensive score of user u, The weight factor representing the emotional state, represents the scene relevance weight factor, represents the content preference matching weight factor, The gain coefficient representing the importance of scene labels to recommended content, represents the weight of the content preference feature vector in preference matching, Indicates the user at a point in time Any scene c in the set of all assigned scene labels, Indicates the weight of the impact of historical emotional state on current emotion; Prioritize content based on the user's comprehensive score, and display the content with the highest score first; Based on the user's current status and system resources, traffic allocation decisions are made to recommend personalized content to users through the most appropriate channels.

2. The method for dividing traffic of an Internet marketing system according to claim 1, characterized in that: The specific steps of extracting environmental data and applying machine learning algorithms to map them to different scene categories are as follows: Obtain the GPS coordinates of the user's device in real time, real-time weather data for the user's location, record the exact time of data collection, and query the public holiday database based on the current date to determine whether it is a holiday; Integrate the acquired environmental data to form a comprehensive environmental feature vector; Based on the comprehensive environmental feature vector, a scene recognition model is constructed and the output probability distribution is expressed as: ； in, Indicates at a point in time The probability that the user is in scene c, represents the comprehensive environment feature vector, Indicates a specific scene. Indicates a specific scene The variance of Indicates a specific scene The mean of represents the scene category label, Representation scene The mean of the eigenvectors of Representation scene The variance of the eigenvector of According to the probability distribution of scene categories output by the model, a threshold Tc is set for each scene; Design a multi-label assignment algorithm, the expression is: ； in, Indicates that in a given environment Under the condition of , the probability that the user is in scene c is, represents the set of scene labels assigned to the user at time t, and Tc represents the assignment threshold of scene c; Traverse all scene categories and calculate the probability of each scene. If the probability exceeds the set threshold, add the scene to the user's scene label list.

3. The method for dividing traffic of an Internet marketing system according to claim 2, characterized in that: The multimodal information includes text, images and videos; The environmental data includes geographical location, weather conditions, time and holidays.

4. The method for dividing traffic of an Internet marketing system according to claim 3, characterized in that: The converted multimodal information refers to the modal features extracted from the diversified information of text, image, and video, and the expression is: ； in, Indicates The overall feature vector of samples after comprehensive processing, i represents the sample index, M represents the number of feature types, Indicates The weight of the class feature, Indicates The sample in Eigenvectors on class features.

5. The method for dividing traffic of an Internet marketing system according to claim 4, characterized in that: The sentiment analysis based on the converted multimodal information refers to performing sentiment scoring on each sample based on the extracted modal features, and the expression is: ； in, Indicates The sentiment score of samples, represents the sentiment score variable, represents the minimum value of the sentiment score, represents the maximum value of the sentiment score, represents the prior probability distribution, represents the variance of the normal distribution, Represents the comprehensive feature vector The converted value of sentiment intensity extracted from Represents the normal distribution function.

Images