A live e-commerce intelligent guidance system and promotion method

Abstract

The present invention relates to a live broadcast e-commerce intelligent guidance system and method, including an anchor form processing module and a purchase desire cooling module. The anchor form processing module provides an anchor viewing mode and a full-item viewing mode, and a real anchor introduces the product or views the product display separately. The purchase desire cooling module is implemented through multiple sub-modules, including: a behavior prediction sub-module, which uses LSTM and behavioral economics models to predict user purchase behavior; an emotion-decision linkage sub-module, which adjusts user emotions and shopping decisions through emotion recognition and cognitive neuroscience models; a game and collaborative filtering sub-module, which combines the herd effect and game theory to optimize the recommendation strategy; an instant feedback and reinforcement learning sub-module, which adjusts shopping prompts and cooling period suggestions in real time. The system can dynamically regulate the recommended content according to the user's emotions and behaviors, provide cooling period prompts, guide users to make more rational shopping decisions, avoid impulsive consumption, improve user experience and optimize inventory management.

Description

Technical Field

[0001] The present invention relates to the technical field of live broadcast e-commerce, and in particular to a live broadcast e-commerce intelligent guidance system and promotion method. Background Art

[0002] With the rapid development of e-commerce, live streaming has become one of the mainstream ways to attract users and promote products. Traditional e-commerce live streaming models mostly rely on real anchors to introduce products through language and video. However, with the increase in the variety of products, it is difficult for a single anchor to fully cover the display and explanation of all products within a limited time, which may cause users to miss the products that suit them. In addition, when watching live streaming, users often have impulsive consumption behaviors in a high emotional state, and lack reasonable purchasing decisions, which in turn leads to regret after shopping and decreased satisfaction.

[0003] Existing technologies have made some progress in improving user experience and optimizing product display, often using collaborative filtering algorithms, recommendation systems and other means to improve the accuracy of recommendations, but they are still insufficient in dealing with user impulse purchases and providing personalized consumption suggestions. Therefore, there is an urgent need for a more intelligent live e-commerce system that comprehensively considers user behavior, emotional state and market dynamics to help users make more rational purchase decisions and improve their consumption experience. Summary of the invention

[0004] In response to the problems in the related technology, the present invention proposes a live e-commerce intelligent guidance system and promotion method to overcome the above-mentioned technical problems existing in the existing related technology.

[0005] To this end, the specific technical solution adopted by the present invention is as follows:

[0006] A live e-commerce intelligent guidance method comprises the following steps:

[0007] Step S1: Anchor form processing and mode selection

[0008] When a user enters the live e-commerce platform, the system provides two modes to choose from through the anchor processing module:

[0009] Anchor viewing mode: users choose a real person or digital human anchor to introduce the product, and convey product information through visual, voice and other forms;

[0010] All-item viewing mode: Display products independently without involving anchors, and users can view product appearance and parameters from multiple angles;

[0011] Based on the user's historical preferences, the system uses the LSTM behavior prediction algorithm to analyze the user's viewing pattern tendencies and provide personalized recommendations;

[0012] Step S2: Behavior prediction analysis

[0013] The system starts the behavior prediction submodule, which uses the LSTM model to predict the user's potential shopping desire by analyzing the user's historical browsing history, click data, and shopping cart addition behavior;

[0014] When the user's behavior data shows that they have a strong desire to shop, the system combines the behavioral economics model to correct the user's impulsive consumption tendency through the behavioral selection bias model and determines whether it is necessary to trigger the purchase desire cooling mechanism;

[0015] Step S3: Emotion recognition and decision-making linkage

[0016] The system monitors the user's emotional state in real time through the emotion-decision linkage submodule:

[0017] The emotion recognition submodule uses CNN and RNN to analyze the user's voice intonation, facial expressions and barrage content to identify whether the user is in a period of emotional fluctuation;

[0018] The cognitive neuroscience submodule combines emotion recognition results to simulate users’ cognitive biases when they are emotionally charged, and adjusts recommendation strategies, such as delaying reminders or providing cooling-off period suggestions, to help users avoid impulsive purchases.

[0019] Step S4: Group behavior and game analysis

[0020] The system dynamically adjusts product recommendations through game theory and collaborative filtering submodules:

[0021] The multi-agent game model analyzes the interaction between users and other live broadcast viewers, and adjusts the priority of product recommendations through the herd effect in the user group;

[0022] The collaborative filtering submodule calculates the recommendation weight based on the similarity of the user's historical behavior and dynamically adjusts the recommended products according to the game results;

[0023] Step S5: Immediate feedback and reinforcement learning optimization

[0024] The system enables instant feedback and reinforcement learning submodules to adjust the prompt mechanism in real time based on user behavior feedback:

[0025] The instant feedback mechanism adjusts the frequency and intensity of cooling reminders according to the user's purchase desire;

[0026] The reinforcement learning model continuously learns from user feedback, optimizes consumption prompt strategies, and gradually enhances the system's adaptability to changes in user behavior;

[0027] Step S6: Time series prediction and personalized regulation

[0028] The system uses a time series prediction algorithm to analyze users’ historical consumption patterns and seasonal behaviors, and dynamically adjusts the triggering conditions of the purchase desire cooling mechanism;

[0029] Based on the prediction results, the system will strengthen the cooling-off period reminder at specific time points or seasonal peaks in advance to prevent users from making impulse purchases when their shopping desire is high;

[0030] Step S7: Assisting the digital human to interact with the user

[0031] The auxiliary digital human module uses the integrated large language model LLM and natural language processing NLP technology to interact with users in real time through voice and text;

[0032] When the system detects that the user's emotions are fluctuating greatly, the digital human will proactively provide cooling-off period suggestions or rational consumption tips based on the user's emotional state to guide the user to make more rational consumption decisions;

[0033] Step S8: System learning and continuous optimization

[0034] During operation, the system continuously records user behavioral feedback, emotional changes, consumption decisions and other information, combines reinforcement learning and behavioral analysis models, and continuously optimizes the triggering conditions and cooling prompt strategies of the purchase desire cooling mechanism to ensure that the system can adapt to the behavioral patterns and emotional changes of different users.

[0035] A live e-commerce intelligent guidance system, including an anchor format processing module and a purchase desire cooling module;

[0036] The anchor format processing module includes anchor viewing mode and all-item viewing mode. The anchor viewing mode is used for customers to select anchors to introduce products, and the all-item viewing mode is used to display products independently.

[0037] The purchase desire cooling module is implemented through multiple sub-modules, including:

[0038] The behavior prediction submodule integrates behavioral economics and LSTM behavior prediction algorithm;

[0039] The emotion-decision linkage submodule, through emotion recognition and cognitive neuroscience models;

[0040] Game and collaborative filtering submodule, through the herd effect and multi-agent game theory;

[0041] The instant feedback and reinforcement learning submodule combines the instant feedback mechanism with the reinforcement learning model to achieve dynamic regulation.

[0042] Furthermore, the behavior prediction submodule includes:

[0043] LSTM behavior prediction model is used to capture the time series characteristics of user behavior and predict the user's potential shopping desire by analyzing the user's historical browsing records, click data, and shopping cart addition behavior;

[0044] The behavioral economics sub-module combines rational choice theory and expected utility theory to correct users' impulsive consumption tendencies through a behavioral selection bias model.

[0045] Furthermore, the emotion-decision linkage submodule operates in the following specific ways:

[0046] The emotion recognition submodule uses the convolutional neural network (CNN) and RNN models to analyze the user's voice intonation, facial expressions, and comment content to identify the user's emotional fluctuations and emotional state.

[0047] The cognitive neuroscience sub-module simulates the cognitive behavioral biases of users when they are emotionally high or anxious through emotion regulation and decision-making theory.

[0048] Furthermore, the game and collaborative filtering submodules include:

[0049] Multi-agent game model, used to simulate the interaction of purchase strategies among user groups, and adjust the recommended content through the strategy of users' shopping behaviors and interactions with others;

[0050] The collaborative filtering submodule calculates the recommendation weight of products based on the similarity analysis of user behaviors, such as the similarity of purchasing and browsing behaviors.

[0051] Furthermore, the instant feedback and reinforcement learning submodules include:

[0052] Instant feedback mechanism, through the analysis of user behavior and psychological state, real-time adjustment of cooling reminder frequency and intensity;

[0053] The reinforcement learning model uses user feedback to continuously optimize the consumption reminder mechanism and enhance the system's adaptability.

[0054] Furthermore, the behavior prediction submodule combines the time series prediction algorithm to analyze the user's historical consumption patterns and seasonal behaviors, and dynamically adjusts the triggering conditions of the purchase desire cooling mechanism.

[0055] Furthermore, the auxiliary digital human module integrates the large language model LLM and natural language processing NLP technology to interact with users in real time through voice and text, and provides cooling-off period suggestions when users are emotionally aroused, helping users make rational decisions.

[0056] The present invention proposes a live e-commerce intelligent guidance system that integrates multiple algorithm models. By combining advanced technologies such as behavioral economics, deep learning, emotion recognition, and multi-agent game theory, it can not only effectively capture users' purchasing behavior and emotional state, but also actively provide cooling-off period suggestions through the system's purchase desire cooling module when users have large emotional fluctuations or impulsive consumption, helping users make more rational shopping decisions. The specific beneficial effects are as follows:

[0057] Personalized consumption prompts and cooling-off mechanism: Through the LSTM behavior prediction model and behavioral economics model, the system can predict the user's purchasing desire in real time and dynamically adjust the cooling-off period prompts. When it detects that the user has irrational consumption tendencies, it actively provides delayed purchase suggestions or product comparison analysis to avoid impulse purchases.

[0058] Emotion-decision linkage: By introducing emotion recognition algorithms and cognitive neuroscience models, the system can analyze the emotional fluctuations of users' voices, expressions, and comments during live broadcasts, identify the users' emotional states in a timely manner, and adjust the intensity of product recommendations or prompt a cooling-off period based on emotional fluctuations, helping users make decisions more calmly.

[0059] Bandwagon effect and game analysis: The system combines a multi-agent game model with a collaborative filtering algorithm. It can not only automatically adjust the weight of product recommendations when users interact with groups, but also adjust the recommended content based on the behavioral interaction strategy between users and other users, thereby increasing the accuracy of recommended products.

[0060] Instant feedback and adaptive reinforcement learning: Through the instant feedback mechanism, the system can adjust the intensity of consumption prompts in real time according to the user's instant clicking behavior and psychological state during the shopping process, and use the reinforcement learning model to continuously optimize the consumption prompt mechanism based on user feedback, further enhancing the system's adaptability.

[0061] Inventory management and market dynamics regulation: The system combines big data analysis and trend prediction algorithms to optimize recommendation strategies based on the market trends and inventory dynamics of commodities, thus avoiding unnecessary consumption burdens caused by impulse purchases when inventory is tight.

[0062] Budget management and rational consumption guidance: The system analyzes the user's historical consumption data and combines it with a linear regression algorithm to predict future consumption behavior. When the user's shopping behavior exceeds the set budget, the system will proactively provide rational consumption suggestions to prevent the user from making irrational consumption when the budget is overspent. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0064] Figure 1 It is a promotion method flow chart of a live broadcast e-commerce intelligent guidance system and promotion method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] According to an embodiment of the present invention, a live e-commerce intelligent guidance system is provided.

[0067] It includes anchor format processing module and purchase desire cooling module;

[0068] The anchor format processing module includes anchor viewing mode and all-item viewing mode. The anchor viewing mode is used for customers to select anchors to introduce products, and the all-item viewing mode is used to display products independently.

[0069] The purchase desire cooling module is implemented through multiple sub-modules, including:

[0070] The behavior prediction submodule integrates behavioral economics and LSTM behavior prediction algorithm;

[0071] The emotion-decision linkage submodule, through emotion recognition and cognitive neuroscience models;

[0072] Game and collaborative filtering submodule, through the herd effect and multi-agent game theory;

[0073] The instant feedback and reinforcement learning submodule combines the instant feedback mechanism with the reinforcement learning model to achieve dynamic regulation.

[0074] Embodiment 1:

[0075] The behavior prediction submodule includes:

[0076] LSTM behavior prediction model is used to capture the time series characteristics of user behavior and predict the user's potential shopping desire by analyzing the user's historical browsing records, click data, and shopping cart addition behavior;

[0077] The behavioral economics sub-module combines rational choice theory and expected utility theory to correct users' impulsive consumption tendencies through a behavioral selection bias model.

[0078] Traditional behavior prediction mostly uses long short-term memory networks (LSTM) to predict future behaviors based on users' historical behavior data such as browsing, clicking, and purchasing. LSTM can capture the time series characteristics of behavior, but it lacks consideration of user emotions, cognitive states, etc.

[0079] By introducing the "rational choice theory" and "expected utility theory" in behavioral economics, the LSTM model is expanded so that it can not only make predictions based on behavioral data, but also combine users' cognitive biases, risk preferences, and long-term and short-term utility trade-offs to improve the accuracy of predictions of purchasing decisions.

[0080] Behavioral selection bias model: By combining the user's purchase behavior history, a bias parameter such as "time discount" or "loss aversion coefficient" is designed to reflect the user's cognitive tendency when facing different products. For example, if the user shows a strong impulse to buy higher-priced products, the system can add a delayed decision or a cooling-off period prompt based on historical bias.

[0081] Fusion mechanism: The LSTM model combines the predicted purchase behavior output with the results of the behavioral economics model. When it detects that a user may make an irrational purchase decision, it adjusts the prediction result through further probabilistic reasoning and generates timely feedback to cool down the purchase desire.

[0082] The innovation of this algorithm lies in introducing user behavioral biases such as impulsive consumption tendencies into the deep learning model, thereby achieving more intelligent personalized consumption intervention. It not only takes into account the user's historical data, but also makes dynamic adjustments to the user's risk preferences and emotional state to adapt to consumption behaviors in different purchasing scenarios.

[0083] The behavior prediction submodule

[0084] Core algorithm: Fusion model of LSTM and behavioral economics

[0085] 1. Input data:

[0086] User historical behavior data: such as browsing history, number of times a product is clicked, shopping cart addition behavior, etc.

[0087] Time interval (Δt): The time interval between browsing, clicking, adding to shopping cart, etc., measured in seconds.

[0088] Purchase history: The price and type of items that the user has previously purchased.

[0089] 2. Behavioral selection bias parameters:

[0090] Time discount factor ;

[0091] Indicates the importance users place on future purchases. A low value indicates that users tend to consume immediately, while a high value indicates that users are willing to wait.

[0092] Loss Aversion Coefficient ;

[0093] Indicates the user's sensitivity to price changes, with high values ​​indicating that users are more sensitive to price increases.

[0094] Behavior threshold (θb): A numerical indicator based on the desire to purchase, with the trigger point threshold set at 0.75.

[0095] 3. Calculation method:

[0096] LSTM captures time series behavior data and outputs purchase intention value (P).

[0097] ,

[0098] Among them, Wt represents the weight matrix of the time series, Ht is the user behavior historical data matrix, and ε is the noise term.

[0099] Behavioral bias correction: Introduce the time discount coefficient and loss aversion coefficient to correct the purchase intention value output by LSTM.

[0100] ,

[0101] The revised purchase intention value is compared with the behavior threshold (θb). If , then the purchase desire cooling mechanism is triggered and a cooling-off period suggestion is generated.

[0102] Embodiment 2:

[0103] The emotion-decision linkage submodule operates in the following specific ways:

[0104] The emotion recognition submodule uses the convolutional neural network (CNN) and RNN models to analyze the user's voice intonation, facial expressions, and comment content to identify the user's emotional fluctuations and emotional state.

[0105] The cognitive neuroscience sub-module simulates the cognitive behavioral biases of users when they are emotionally high or anxious through emotion regulation and decision-making theory.

[0106] Current emotion recognition algorithms mostly rely on convolutional neural networks (CNN) and RNN, which analyze users' voice intonation, facial expressions and other features to identify users' emotional states in real time. Although they can capture emotional fluctuations, they pay little attention to the relationship between emotions and behaviors.

[0107] By introducing the "emotion and decision-making" theory in cognitive neuroscience, combining the results of emotion recognition with the user's cognitive load, emotion regulation ability, etc., a multi-level emotion-decision model is designed to regulate the user's shopping behavior in real time.

[0108] Emotion-cognition linkage model: The emotion recognition results are input into a cognitive control layer, which simulates the user's cognitive biases under high-intensity emotions, such as over-optimism or impulsive behavior when nervous, and adjusts the recommendation strategy based on the real-time detected emotion data. For example, when the user is in a high mood, he may not be sensitive to the price of the product, and the system can provide delayed reminders.

[0109] Adjustment based on cognitive load: By real-time monitoring the degree of information processing load of users during the shopping process, the frequency and intensity of emotional feedback can be dynamically adjusted. For example, when the system detects that the user's emotions tend to be unstable after browsing large-value items multiple times, the density of product recommendations can be reduced, and a cooling-off period or rational suggestions can be added.

[0110] This algorithm improves the adaptability and effectiveness of emotion recognition algorithms by combining emotional regulation in cognitive neuroscience with shopping decisions. It can identify the user's emotional state in specific situations and provide personalized shopping guidance through cognitive-level control, making the system more intelligent and humane.

[0111] The emotion-decision linkage submodule

[0112] Core algorithm: Linkage model of CNN and cognitive neuroscience

[0113] Input data:

[0114] Speech Intonation Parameter (V): The pitch frequency change based on speech recognition, in Hertz (Hz), ranging from 150Hz to 250Hz.

[0115] Facial expression feature vector (F): 6 emotional states (happy, angry, anxious, surprised, sad, calm) based on facial expression recognition.

[0116] Bullet comment content keywords (K): emotional keywords extracted through NLP and assigned emotional weights.

[0117] Parametric Model:

[0118] Emotional fluctuation coefficient (αe): CNN and RNN are used to identify the user’s emotional state through voice intonation, facial expressions, and comments, and calculate the fluctuation amplitude.

[0119] ,

[0120] Among them, wi is the weight of each expression state, Fi is the facial expression feature, wv is the weight of voice intonation, and g(V) is the voice fluctuation feature function.

[0121] Emotion threshold (θe): The threshold of emotional excitement is defined as 0.8. , triggers a cooldown period suggestion.

[0122] Cognitive bias correction coefficient (βc): simulates the cognitive behavioral bias of users when their emotions are high. The constant is set to 1.2, which means that the higher the emotion, the greater the user's cognitive bias.

[0123] ,

[0124] Specific operations:

[0125] The amplitude of emotional fluctuation is output by joint analysis of voice and expression αe. , the system further adjusts the purchase intention based on the cognitive bias correction coefficient and gives rational prompts.

[0126] Embodiment three:

[0127] The game and collaborative filtering submodules include:

[0128] Multi-agent game model, used to simulate the interaction of purchase strategies among user groups, and adjust the recommended content through the strategy of users' shopping behaviors and interactions with others;

[0129] The collaborative filtering submodule calculates the recommendation weight of products based on the similarity analysis of user behaviors, such as the similarity of purchasing and browsing behaviors.

[0130] Game and collaborative filtering submodules

[0131] Core algorithm: Multi-agent game and collaborative filtering combined model

[0132] Input data:

[0133] User similarity matrix (S): Based on the similarity of users' browsing, clicking, and purchasing behaviors, the value range is between 0 and 1.

[0134] Conformity coefficient (κc): indicates the intensity of the influence of group behavior on users, ranging from [0.2, 0.8].

[0135] Product click-through rate (CR): The unit is clicks / minute.

[0136] Collaborative filtering model:

[0137] Collaborative filtering calculates the recommendation weights of similar user groups:

[0138] ,

[0139] Among them, Sij is the similarity between user i and user j, and Rj is the rating of the product by user j.

[0140] Herd coefficient adjustment:

[0141] ,

[0142] when , it means that the user is under a strong herd influence and the system triggers a cooling-off period prompt.

[0143] Gaming strategy adjustment:

[0144] Multi-agent game formula: Introduce the cost function and benefit function of game strategy selection.

[0145] ,

[0146] Among them, σ is the strategy selection, U is the utility function of user i, and Ci is the cost of choosing the strategy. Users will adjust their own strategies according to the strategies of other users. If the utility of the selected strategy exceeds a certain critical point, such as increasing the recommendation weight by 30%, then alternative product recommendations will be provided.

[0147] Embodiment 4:

[0148] The instant feedback and reinforcement learning submodules include:

[0149] Instant feedback mechanism, through the analysis of user behavior and psychological state, real-time adjustment of cooling reminder frequency and intensity;

[0150] The reinforcement learning model uses user feedback to continuously optimize the consumption reminder mechanism and enhance the system's adaptability.

[0151] Current reinforcement learning algorithms are mostly used to dynamically adjust recommendation systems and continuously optimize based on user clicks, purchases, and other behaviors. However, existing reinforcement learning models mainly focus on user behavioral feedback and lack consideration of immediate psychological states and cognitive biases.

[0152] By combining the immediate feedback mechanism and self-control theory in psychology, a reinforcement learning model based on psychological feedback is designed, which can adjust the frequency and intensity of shopping prompts and cooling suggestions in real time, thereby helping users make more rational purchasing decisions.

[0153] Instant psychological feedback mechanism: By introducing an instant feedback model and combining it with the reward value in the reinforcement learning algorithm, the system dynamically adjusts the feedback mechanism based on the user's instant psychological state, such as anxiety, excitement, etc. For example, when the user shows a high desire to buy, the system reduces the frequency of instant feedback to reduce shopping prompts and avoid stimulating further impulsive behavior.

[0154] Multi-step decision feedback based on self-control: Using the theory of self-control in psychology, the system simulates the user's decision-making process through reinforcement learning and optimizes the recommendation strategy based on multi-step feedback. For example, when the user repeatedly ignores the delayed purchase suggestion, the system can gradually weaken the intensity of the cooling-off period prompt and turn to providing other shopping suggestions such as alternative products.

[0155] The algorithm enhances the adaptability of the reinforcement learning algorithm by introducing the psychological instant feedback and self-control model. The system can not only adjust the recommendation strategy based on the user's behavioral feedback, but also respond to changes in the user's psychological state in real time and provide more flexible shopping suggestions.

[0156] The instant feedback and reinforcement learning submodules

[0157] Core algorithm: instant feedback and reinforcement learning adaptive model

[0158] Input data:

[0159] Click frequency (CF): The number of times a user clicks on a product per minute, measured in times / minute.

[0160] Browsing Time (BT): The time a user stays on a specific product page, measured in seconds.

[0161] Feedback signal (F): The user's response to the cooling-off period suggestion, which is a binary value (0 means ignore, 1 means accept).

[0162] Instant feedback mechanism:

[0163] Determine the intensity of the user's desire to buy based on the user's click frequency and browsing time:

[0164] ,

[0165] like , indicating that the user has a strong purchase intention, the system gradually reduces the frequency of cooling prompts.

[0166] Reinforcement Learning Model:

[0167] Update the policy via Q-learning:

[0168] ,

[0169] Among them, α is the learning rate, γ is the discount factor, and r is the user's feedback signal (0 or 1) for the cooling period prompt.

[0170] Behavior adjustment: Based on user feedback signals, the system updates the cooling-off period strategy. If the user ignores the prompt multiple times, the system reduces the prompt intensity and provides alternative suggestions.

[0171] The behavior prediction submodule combines the time series prediction algorithm to analyze the user's historical consumption patterns and seasonal behaviors, and dynamically adjusts the triggering conditions of the purchase desire cooling mechanism.

[0172] The auxiliary digital human module integrates the large language model (LLM) and natural language processing (NLP) technology to interact with users in real time through voice and text, and provides cooling-off period suggestions when users are emotionally aroused, helping users make rational decisions.

[0173] A live e-commerce intelligent guidance method comprises the following steps:

[0174] S1: Anchor format processing and mode selection

[0175] When a user enters the live e-commerce platform, the system provides two modes to choose from through the anchor processing module:

[0176] Anchor viewing mode: users choose a real person or digital human anchor to introduce the product, and convey product information through visual, voice and other forms;

[0177] All-item viewing mode: Display products independently without involving anchors, and users can view product appearance and parameters from multiple angles;

[0178] Based on the user's historical preferences, the system uses the LSTM behavior prediction algorithm to analyze the user's viewing pattern tendencies and provide personalized recommendations;

[0179] S2: Behavior Prediction Analysis

[0180] The system starts the behavior prediction submodule, which uses the LSTM model to predict the user's potential shopping desire by analyzing the user's historical browsing history, click data, and shopping cart addition behavior;

[0181] When the user's behavior data shows that they have a strong desire to shop, the system combines the behavioral economics model to correct the user's impulsive consumption tendency through the behavioral selection bias model and determines whether it is necessary to trigger the purchase desire cooling mechanism;

[0182] S3: Emotion recognition and decision-making linkage

[0183] The system monitors the user's emotional state in real time through the emotion-decision linkage submodule:

[0184] The emotion recognition submodule uses CNN and RNN to analyze the user's voice intonation, facial expressions and barrage content to identify whether the user is in a period of emotional fluctuation;

[0185] The cognitive neuroscience submodule combines emotion recognition results to simulate users’ cognitive biases when they are emotionally charged, and adjusts recommendation strategies, such as delaying reminders or providing cooling-off period suggestions, to help users avoid impulsive purchases.

[0186] S4: Group Behavior and Game Analysis

[0187] The system dynamically adjusts product recommendations through game theory and collaborative filtering submodules:

[0188] The multi-agent game model analyzes the interaction between users and other live broadcast viewers, and adjusts the priority of product recommendations through the herd effect in the user group;

[0189] The collaborative filtering submodule calculates the recommendation weight based on the similarity of the user's historical behavior and dynamically adjusts the recommended products according to the game results;

[0190] S5: Immediate feedback and reinforcement learning optimization

[0191] The system enables instant feedback and reinforcement learning submodules to adjust the prompt mechanism in real time based on user behavior feedback:

[0192] The instant feedback mechanism adjusts the frequency and intensity of cooling reminders according to the user's purchase desire;

[0193] The reinforcement learning model continuously learns from user feedback, optimizes consumption prompt strategies, and gradually enhances the system's adaptability to changes in user behavior;

[0194] S6: Time series prediction and personalized regulation

[0195] The system uses a time series prediction algorithm to analyze users’ historical consumption patterns and seasonal behaviors, and dynamically adjusts the triggering conditions of the purchase desire cooling mechanism;

[0196] Based on the prediction results, the system will strengthen the cooling-off period reminder at specific time points or seasonal peaks in advance to prevent users from making impulse purchases when their shopping desire is high;

[0197] S7: Assist digital humans to interact with users

[0198] The auxiliary digital human module uses the integrated large language model LLM and natural language processing NLP technology to interact with users in real time through voice and text;

[0199] When the system detects that the user's emotions are fluctuating greatly, the digital human will proactively provide cooling-off period suggestions or rational consumption tips based on the user's emotional state to guide the user to make more rational consumption decisions;

[0200] S8: System learning and continuous optimization

[0201] During operation, the system continuously records user behavioral feedback, emotional changes, consumption decisions and other information, combines reinforcement learning and behavioral analysis models, and continuously optimizes the triggering conditions and cooling prompt strategies of the purchase desire cooling mechanism to ensure that the system can adapt to the behavioral patterns and emotional changes of different users.

[0202] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for intelligently guiding live e-commerce, characterized in that: The following steps are involved: Step S1: Anchor form processing and mode selection When a user enters the live e-commerce platform, the system provides two modes to choose from through the anchor processing module: Anchor viewing mode: users choose a real person or digital anchor to introduce the product, and convey product information through visual and voice forms; All-item viewing mode: Display products independently without involving anchors, and users can view product appearance and parameters from multiple angles; Based on the user's historical preferences, the system uses the LSTM behavior prediction algorithm to analyze the user's viewing pattern tendencies and provide personalized recommendations; Step S2: Behavior prediction analysis The system starts the behavior prediction submodule, which uses the LSTM model to predict the user's potential shopping desire by analyzing the user's historical browsing records, click data, and shopping cart behavior; When the user's behavior data shows that they have a strong desire to shop, the system combines the behavioral economics model to correct the user's impulsive consumption tendency through the behavioral selection bias model and determines whether it is necessary to trigger the purchase desire cooling mechanism; Step S3: Emotion recognition and decision-making linkage The system monitors the user's emotional state in real time through the emotion-decision linkage submodule: The emotion recognition submodule uses convolutional neural networks and RNNs to analyze the user's voice intonation, facial expressions, and barrage content to identify whether the user is in a period of emotional fluctuation; The cognitive neuroscience submodule combines emotion recognition results to simulate users’ cognitive biases when they are emotionally charged, and adjusts recommendation strategies, such as delaying reminders or providing cooling-off period suggestions, to help users avoid impulsive purchases. Step S4: Group behavior and game analysis The system dynamically adjusts product recommendations through game theory and collaborative filtering submodules: The multi-agent game model analyzes the interaction between users and other live broadcast viewers, and adjusts the priority of product recommendations through the herd effect in the user group; The collaborative filtering submodule calculates the recommendation weight based on the similarity of the user's historical behavior and dynamically adjusts the recommended products according to the game results; Step S5: Immediate feedback and reinforcement learning optimization The system enables instant feedback and reinforcement learning submodules to adjust the prompt mechanism in real time based on user behavior feedback: The instant feedback mechanism adjusts the frequency and intensity of cooling reminders according to the user's purchase desire; The reinforcement learning model continuously learns from user feedback, optimizes consumption prompt strategies, and gradually enhances the system's adaptability to changes in user behavior; Step S6: Time series prediction and personalized regulation The system uses a time series prediction algorithm to analyze users’ historical consumption patterns and seasonal behaviors, and dynamically adjusts the triggering conditions of the purchase desire cooling mechanism; Based on the prediction results, the system will strengthen the cooling-off period reminder at specific time points or seasonal peaks in advance to prevent users from making impulse purchases when their shopping desire is high; Step S7: Assisting the digital human to interact with the user The auxiliary digital human module uses the integrated large language model LLM and natural language processing NLP technology to interact with users in real time through voice and text; When the system detects that the user's emotions fluctuate greatly, the digital human will proactively provide cooling-off period suggestions or rational consumption tips based on the user's emotional state to guide the user to make more rational consumption decisions; Step S8: System learning and continuous optimization During operation, the system continuously records users' behavioral feedback, emotional changes, and consumption decision information. By combining reinforcement learning and behavioral analysis models, it continuously optimizes the triggering conditions and cooling prompt strategies of the purchase desire cooling mechanism, ensuring that the system can adapt to the behavioral patterns and emotional changes of different users.

2. A live e-commerce intelligent guidance system, used for the live e-commerce intelligent guidance method according to claim 1, characterized in that: It includes anchor format processing module and purchase desire cooling module; The anchor format processing module includes an anchor viewing mode and a full item viewing mode. The anchor viewing mode is used for customers to select an anchor to introduce products, and the full item viewing mode is used to independently display products. The purchase desire cooling module is implemented through multiple sub-modules, specifically including: The behavior prediction submodule integrates behavioral economics and LSTM behavior prediction algorithm; The emotion-decision linkage submodule, through emotion recognition and cognitive neuroscience models; Game and collaborative filtering submodule, through the herd effect and multi-agent game theory; The instant feedback and reinforcement learning submodule combines the instant feedback mechanism with the reinforcement learning model to achieve dynamic regulation.

3. According to claim 2, a live broadcast e-commerce intelligent guidance system is characterized in that: The behavior prediction submodule includes: LSTM behavior prediction model is used to capture the time series characteristics of user behavior and predict the user's potential shopping desire by analyzing the user's historical browsing records, click data, and shopping cart addition behavior; The behavioral economics sub-module combines rational choice theory and expected utility theory to correct users' impulsive consumption tendencies through a behavioral selection bias model.

4. According to claim 3, a live broadcast e-commerce intelligent guidance system is characterized in that: The emotion-decision linkage submodule includes: The emotion recognition submodule uses convolutional neural networks and RNN models to analyze the user's voice intonation, facial expressions, and bullet comments to identify the user's emotional fluctuations and emotional state; The cognitive neuroscience sub-module simulates the cognitive behavioral biases of users when they are emotionally high or anxious through emotion regulation and decision-making theory.

5. According to claim 4, a live broadcast e-commerce intelligent guidance system is characterized in that: The game and collaborative filtering submodule includes: Multi-agent game model, used to simulate the interaction of purchase strategies among user groups, and adjust the recommended content through the strategy of users' shopping behaviors and interactions with others; The collaborative filtering submodule calculates the recommendation weight of products based on the similarity analysis of user behaviors, such as the similarity of purchasing and browsing behaviors.

6. A live broadcast e-commerce intelligent guidance system according to claim 5, characterized in that: The instant feedback and reinforcement learning submodule includes: Instant feedback mechanism, through the analysis of user behavior and psychological state, real-time adjustment of cooling reminder frequency and intensity; The reinforcement learning model uses user feedback to continuously optimize the consumption reminder mechanism and enhance the system's adaptability.

7. A live broadcast e-commerce intelligent guidance system according to claim 6, characterized in that: The behavior prediction submodule combines the time series prediction algorithm to analyze the user's historical consumption patterns and seasonal behaviors, and dynamically adjusts the triggering conditions of the purchase desire cooling mechanism.

8. A live broadcast e-commerce intelligent guidance system according to claim 7, characterized in that: The auxiliary digital human module integrates the large language model LLM and natural language processing NLP technology, interacts with users in real time through voice and text, and provides cooling-off period suggestions when users are emotionally aroused, helping users make rational decisions.

Images