Gpt-combined commodity marketing ai dialogue method and electronic equipment thereof
By combining GPT model and machine learning algorithms, analyzing user behavior and generating personalized marketing conversations, the problem of inability to accurately match user needs in traditional marketing is solved, and more efficient marketing results are achieved.
Patent Information
- Application Number
- CN202410294249.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-14
- Publication Date
- 2024-12-20
AI Technical Summary
Traditional intelligent marketing technology uses a fixed knowledge base, and users need to enter keywords in advance, which limits the nature of precise marketing and cannot effectively make personalized recommendations based on users' unique needs and preferences.
Combining the GPT model, by analyzing users' historical purchase records, browsing behaviors and social media interactions, building user portraits, using machine learning algorithms to predict user needs and preferences, and generating mixed recommendations based on needs and preferences, using the GPT model to generate personalized marketing Conversation content.
It realizes intelligent marketing based on users' unique needs and preferences, improves the pertinence of marketing conversations and user satisfaction, overcomes the limitations of the system's passive feedback keywords, and achieves more accurate marketing effects.
Smart Images
Figure CN119168726A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of AI applications, and in particular to a commodity marketing AI dialogue method combined with GPT and its electronic device. Background Art
[0002] Intelligent marketing is a product of technological progress and a bridge for communication between enterprises and users. It integrates large-scale knowledge processing, natural language understanding, knowledge management, automatic question answering and reasoning technology, enabling enterprises to conduct fine-grained knowledge management more effectively.
[0003] Traditional technology uses a fixed knowledge base, which requires users to enter in advance and the system passively feedbacks keywords, which limits the effectiveness of precision marketing. Summary of the invention
[0004] The purpose of this application is to overcome the above-mentioned prior art and provide a commodity marketing AI dialogue method combined with GPT and its electronic device.
[0005] This application provides a commodity marketing AI dialogue method combined with GPT, including:
[0006] Build user profiles based on their historical purchase records, browsing behavior, and social media interactions;
[0007] Based on the user profile, use machine learning algorithms to predict user needs and preferences;
[0008] Filtering a first product or service according to the demand, filtering a second product or service according to the preference, and generating a mixed recommendation based on the first product or service and the second product or service;
[0009] Based on the hybrid recommendation, the GPT model is used to generate conversation content for marketing conversation.
[0010] Optionally, build a user profile based on the user's historical purchase history, browsing behavior, and social media interactions, including:
[0011] Collect data from users’ purchase history, browsing behavior, and social media interactions;
[0012] Preprocess the collected raw data, including cleaning, normalization, and filling or deleting missing data;
[0013] Extract features related to user profiles from preprocessed data;
[0014] A user profile is constructed using a classifier or a clustering algorithm based on the preprocessed data.
[0015] Optionally, based on the user portrait, a machine learning algorithm is used to predict user needs and preferences, including:
[0016] Extracting features related to user needs and preferences from the user portrait data;
[0017] Select logistic regression, support vector machine, decision tree, random forest or neural network algorithm for prediction, including: training the selected machine learning algorithm to build a prediction model; using an independent data set to evaluate the trained model; tuning the model based on the evaluation results; deploying the trained model to the production environment to achieve real-time prediction of user needs and preferences.
[0018] Optionally, generating a mixed recommendation based on the first product or service and the second product or service includes:
[0019] Based on user needs, collaborative filtering algorithms are used to filter the first product or service;
[0020] Based on user preferences, use text analysis technology to filter the second product or service;
[0021] According to the matching degree between the first commodity or service and the second commodity or service, a personalized mixed recommendation is generated, and the expression is as follows:
[0022] H q =λ1u i +λ2p j ,u i ∈U,p j ∈P
[0023] Among them, the u i is the matching degree between the first product or service and the demand, the p j is the matching degree between the second commodity and service and the preference, λ1 and λ2 are the weights of the first commodity and service and the second commodity and service, which are determined by the entropy weight method, and U and P are the first commodity and service set and the second commodity and service set.
[0024] Optionally, based on the hybrid recommendation, a GPT model is used to generate dialogue content for a marketing dialogue, including:
[0025] Based on the features, using the GPT model to generate sales pitch statements or questions to guide users to learn or purchase; and
[0026] The GPT model is updated and optimized based on the user's response rate and conversation duration to improve the pertinence of conversation content and user satisfaction.
[0027] The present application also provides an electronic device for commodity marketing AI dialogue combined with GPT, including:
[0028] A memory for storing a computer executable program of the commodity marketing AI dialogue method combined with GPT as described in any one of claims 1 to 5;
[0029] A processor is used to call the computer executable program and execute: building a user portrait based on the user's historical purchase records, browsing behaviors, and social media interactions; predicting user needs and preferences based on the user portrait using a machine learning algorithm; screening a first product or service based on the needs, and screening a second product or service based on the preferences, and generating a mixed recommendation based on the first product or service and the second product or service; and generating dialogue content for marketing dialogue using a GPT model based on the mixed recommendation.
[0030] Optionally, the processor constructs a user profile based on the user's historical purchase records, browsing behaviors, and social media interactions, including:
[0031] Collect data from users’ purchase history, browsing behavior, and social media interactions;
[0032] Preprocess the collected raw data, including cleaning, normalization, and filling or deleting missing data;
[0033] Extract features related to user profiles from preprocessed data;
[0034] A user profile is constructed using a classifier or a clustering algorithm based on the preprocessed data.
[0035] Optionally, the processor predicts user needs and preferences using a machine learning algorithm based on the user portrait, including:
[0036] Extracting features related to user needs and preferences from the user portrait data;
[0037] Select logistic regression, support vector machine, decision tree, random forest or neural network algorithm for prediction, including: training the selected machine learning algorithm to build a prediction model; using an independent data set to evaluate the trained model; tuning the model based on the evaluation results; deploying the trained model to the production environment to achieve real-time prediction of user needs and preferences.
[0038] Optionally, the processor generates a mixed recommendation according to the first product or service and the second product or service, including:
[0039] Based on user needs, collaborative filtering algorithms are used to filter the first product or service;
[0040] Based on user preferences, use text analysis technology to filter the second product or service;
[0041] According to the matching degree between the first commodity or service and the second commodity or service, a personalized mixed recommendation is generated, and the expression is as follows:
[0042] H q =λ1u i +γ2p j ,u i ∈U,p j ∈P
[0043] Among them, the u i is the matching degree between the first product or service and the demand, the p j is the matching degree between the second commodity and service and the preference, λ1 and λ2 are the weights of the first commodity and service and the second commodity and service, which are determined by the entropy weight method, and U and P are the first commodity and service set and the second commodity and service set.
[0044] Optionally, the processor generates dialogue content using a GPT model according to the hybrid recommendation to conduct a marketing dialogue, including:
[0045] Based on the features, using the GPT model to generate sales pitch statements or questions to guide users to learn or purchase; and
[0046] The GPT model is updated and optimized based on the user's response rate and conversation duration to improve the pertinence of conversation content and user satisfaction.
[0047] The beneficial effects of this application are:
[0048] Invention point 1: Hybrid recommendation based on needs and preferences
[0049] Invention point 2: Filtering the first product or service and the second product or service based on demand and preference respectively Invention point 3: Product recommendation by integrating hybrid recommendation and GPT
[0050] The present application provides a product marketing AI dialogue method combined with GPT, including: constructing a user portrait based on the user's historical purchase records, browsing behavior, and social media interactions; predicting user needs and preferences based on the user portrait using a machine learning algorithm; screening a first product or service based on the needs, and screening a second product or service based on the preferences, and generating a mixed recommendation based on the first product or service and the second product or service; and using a GPT model to generate dialogue content for marketing dialogue based on the mixed recommendation. The present application analyzes and conducts intelligent marketing based on the user's unique needs and preferences, solving the problem of the system passively feeding back keywords and limiting precision marketing. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] Figure 1 It is a flowchart of the product marketing AI dialogue method combined with GPT in this application;
[0052] Figure 2 This is a schematic diagram of the process of building a user portrait in this application;
[0053] Figure 3 It is a schematic diagram of an electronic device for the product marketing ai dialogue combined with gpt in this application; DETAILED DESCRIPTION
[0054] The present application is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present application and implement it.
[0055] The following contents are all examples of specific implementation processes provided for describing in detail the technical solutions to be protected by this application. However, this application is also implemented in other ways different from the descriptions herein. Under the guidance of the concept of this application, those skilled in the art may implement this application using different technical means. Therefore, this application is not limited to the specific embodiments below.
[0056] Please refer to Figure 1 As shown, the present application provides a commodity marketing AI dialogue method combined with GPT, the steps of which include:
[0057] S101 builds user portraits based on their historical purchase records, browsing behaviors, and social media interactions.
[0058] This application builds an accurate user profile based on the user's historical purchase records, browsing behavior and social media interactions. This profile helps to better understand the user's needs and preferences, so as to provide more personalized services and products.
[0059] like Figure 2 As shown in the figure, the specific steps of building a user portrait include:
[0060] S201 collects data from users’ purchase history, browsing behavior, and social media interactions.
[0061] Get raw data from various sources. This data may come from your company’s database, user logs from your website or app, social media platforms, etc. The collected data should include the following:
[0062] Purchase records: This includes the name, price, quantity, purchase date, etc. of the goods or services purchased by the user.
[0063] Browsing behavior: This includes the pages users visit, dwell time, browsing paths, click behavior, etc.
[0064] Social media engagement: This includes user interactions with a brand or product on social media, such as posts, comments, likes, etc.
[0065] S202 pre-processes the collected raw data.
[0066] Data cleaning: Eliminate errors and outliers in the raw data. For example, if the value of a field does not fit the expected range or contains illegal characters, cleaning is required.
[0067] Data normalization: Since the dimensions and magnitudes of features may be different, using the original data directly will cause some algorithms to fail to learn correctly. Therefore, the data needs to be converted to the same dimension to make them comparable. Common normalization methods include minimum-maximum normalization and Z-score normalization.
[0068] Handling missing values: Common methods include filling missing values (for example, using the mean or median) or deleting rows or columns with missing values.
[0069] S203 extracts features related to the user portrait from the preprocessed data.
[0070] Feature extraction is the process of converting raw data into a format that can be understood by machine learning algorithms. These features should be able to describe user preferences, behaviors, and attributes. For example, extract the type, frequency, and total amount of purchased items from purchase records; extract the type of browsed items, number of pages browsed, and dwell time from browsing behavior; extract positive or negative comments on brands or products from social media interactions, etc.
[0071] S204 constructs a user portrait using a classifier or clustering algorithm based on the preprocessed data.
[0072] This step involves using classifiers or clustering algorithms to analyze the preprocessed data and build user profiles. Classifiers predict a user's likely preferences or behaviors based on their behavior and attributes (e.g., using logistic regression or support vector machines for classification). Clustering algorithms group users with similar behaviors together (e.g., using K-means clustering to divide users into different groups).
[0073] Based on the results of the classifier or clustering algorithm, the characteristics of each group are further analyzed to build a more accurate user portrait. For example, based on the user's purchase history and browsing behavior, their preference for different types of goods and their purchasing habits are analyzed to recommend more suitable goods or services for them.
[0074] Furthermore, we will optimize and improve the user portrait from the following aspects:
[0075] In addition to purchase history, browsing behavior, and social media interactions, consider adding other data sources such as the user’s geographic location, occupation, age, etc.
[0076] For existing data, we conduct in-depth mining to discover users’ potential needs and preferences. For example, we analyze users’ browsing behavior to find out the types of products, brands, and price ranges they are interested in.
[0077] Users’ interests and needs are constantly changing, so the user profile should be updated regularly to keep it accurate. This can be done by regularly collecting new data about the user or updating the profile based on the user’s latest behavior.
[0078] In this application, machine learning technology is used to automatically learn and identify user interests and needs, so as to build user portraits more accurately. For example, clustering algorithms are used to divide users into different groups, or association rule mining algorithms are used to discover the associations between users' purchases of goods.
[0079] S102 predicts user needs and preferences using a machine learning algorithm based on the user portrait.
[0080] Specifically, features related to user needs and preferences are extracted from the user portrait data, including:
[0081] User's basic information: age, gender, geographic location, etc.
[0082] User purchase history: type of goods purchased, frequency, price, etc.
[0083] User browsing behavior: type of products browsed, dwell time, browsing frequency, etc.
[0084] User social media interactions: product-related posts, likes, comments, etc.
[0085] Furthermore, standardization or normalization is performed on numerical features, such as age, purchase frequency, etc., and labeling is performed on categorical features, such as gender, product type, etc.
[0086] In this application, not all features are helpful for predicting user needs and preferences. Therefore, feature selection is performed to select the features that are most relevant to the prediction target. Feature selection algorithms are used, such as model-based filtering, rule-based filtering, etc.
[0087] Suppose you want to predict whether a user will buy a certain brand of clothing, and extract the following features from the user portrait:
[0088] Age: Users of different age groups have different preferences and demands for clothing.
[0089] Gender: Users of different genders also have different preferences and demands for clothing.
[0090] Purchase history: the brand, type, price, etc. of clothing that the user has purchased in the past, reflecting his / her purchasing habits and preferences.
[0091] Browsing behavior: the clothing brands, types, prices, browsing frequency, etc. that users have browsed, reflecting their potential needs and interests.
[0092] Social media interaction: Users’ interactions with clothing on social media, such as likes and comments, reflect their attitudes and preferences towards clothing.
[0093] In the feature selection stage, a model-based filtering method is used. For example, a logistic regression model is used to predict whether a user will buy a certain brand of clothing. The expression is as follows:
[0094] The formula for the logistic regression model is:
[0095]
[0096] Where Z = β0 + Σ(β i *X i )+Σ(β j *X j )+Σ(β k *X k ); f(Z)=α1*log(Z)+α2*Z 2 +α3*tanh(Z); Y is the target variable of purchase behavior (0 means no purchase, 1 means purchase); X1, X2, ..., X n are characteristic variables; β0, β1, β2, ..., β n are the coefficients of the model, and α1, α2, and α3 are additional parameters introduced to enhance the complexity of the model and improve the prediction performance.
[0097] Among them, the f(Z) function is a composite function, including logarithmic function, quadratic function and hyperbolic tangent function. Such a design can capture the nonlinear relationship between features and targets, thereby better fitting the data.
[0098] α parameters: α1, α2, and α3 are parameters that adjust the weights of the various functions in f(Z). These parameters are optimized through methods such as cross-validation or Bayesian optimization to find the best model configuration. The introduction of these parameters makes the model more flexible and can better adapt to different data distributions and business scenarios.
[0099] By training the model, we can get the coefficient of each feature, and then determine the relevance of the feature to the prediction target. In this example, we found that the coefficients of "purchase history" and "browsing behavior" are large, indicating that these two features are highly correlated with whether the user purchases a certain brand of clothing. Therefore, we select "purchase history" and "browsing behavior" as the most relevant features to build the prediction model.
[0100] At the same time, rule-based filtering is also used, such as judging whether a feature is relevant to the prediction target based on business rules or experience. For example, "age" and "gender" are considered to be features related to clothing purchase preferences, so these features are selected as relevant features.
[0101] Choose an appropriate machine learning model based on the nature of the features and target variables. For example, if the goal is to predict whether a user will buy a product, use a logistic regression or decision tree classifier; if the goal is to predict the user's preference for a product, use a support vector machine or random forest regressor; if the goal is to discover the user's purchase pattern, use a clustering algorithm such as K-means clustering.
[0102] The model was trained using historical data and the performance of the model was evaluated using cross-validation.
[0103] According to the performance and evaluation results of the model, the model is optimized and adjusted, for example, adjusting the parameters of the model, selecting different kernel functions, etc.
[0104] Use the trained model to predict new user data. Evaluate the performance of the model by comparing the predicted results with the actual results. Common evaluation indicators include accuracy, recall, F1 score, etc.
[0105] Deploy the trained model to the production environment, monitor the performance of the model in real time, and update the model regularly to adapt to data changes.
[0106] Finally, based on the trained model, predict user needs and preferences.
[0107] S103: filtering the first product or service according to the demand, filtering the second product or service according to the preference, and generating a mixed recommendation based on the first product or service and the second product or service.
[0108] According to the needs and preferences of the user, the first product or service and the second product or service are screened out. In order to generate mixed recommendations, these products or services need to be further processed and integrated to provide more personalized recommendations that meet the needs of the user.
[0109] Specifically, the first product or service is combined with the second product or service to generate a mixed recommendation. For example, if a user has a purchase demand for a certain brand of clothing and also shows a preference for a certain brand of shoes, these two categories of products are combined together to generate a mixed recommendation, recommending the user to purchase both categories of products at the same time.
[0110] According to the needs and preferences of the user, the first product or service and the second product or service are screened out, and according to their matching degree with the needs and matching degree with the preferences, a personalized mixed recommendation is generated. Specifically, the following expression is used to describe:
[0111] H q =λ1u i +λ2p j ,u i ∈U,p j ∈P
[0112] in:
[0113] H q Indicates the matching degree of mixed recommendations.
[0114] λ1 and λ2 are weights used to balance the matching degree u between the first product or service and the demand i The matching degree p between the second product or service and the preference j .
[0115] u i It is the matching degree between the first product or service and the demand, indicating the degree to which the product or service meets the user's needs.
[0116] p j It is the matching degree between the second product or service and the preference, indicating the degree to which the product or service meets the user's preference.
[0117] u i ∈U indicates that the first commodity or service belongs to the first commodity and service set U.
[0118] p j ∈P indicates that the second product or service belongs to the second product and service set P.
[0119] The determination of weights λ1 and λ2 is critical because they determine the weight of the matching degree between the first product or service and the demand and the matching degree between the second product or service and the preference in the mixed recommendation. In order to determine these two weights, the entropy weight method is used.
[0120] Specifically, the entropy weight method is a weight determination method based on information theory, which objectively determines the weight of each indicator according to the characteristics of the data itself. For the matching degree between the first commodity or service and the demand and the matching degree between the second commodity or service and the preference, their entropy values are calculated respectively, and then the values of λ1 and λ2 are determined according to the size of the entropy values.
[0121] S104 generates dialogue content for marketing dialogue using the GPT model based on the hybrid recommendation.
[0122] In this application, the marketing dialogue based on hybrid recommendation is a personalized and intelligent marketing method that provides more accurate and appropriate recommendations and services based on the interests and needs of users.
[0123] Use the GPT model to generate conversation content for marketing conversations, further improving marketing effectiveness and user experience. The GPT model is a natural language processing model based on deep learning that generates natural and fluent language, making marketing conversations more humane and friendly.
[0124] Specifically, based on the user's preference for the first product or service and the second product or service, some questions about product features, prices, and usage methods are generated to guide the user to further understand and purchase. At the same time, the GPT model is used to generate sales statements, such as emphasizing the advantages of the product and providing limited-time discounts, to stimulate the user's willingness to buy.
[0125] In addition, in order to improve the pertinence of conversation content and user satisfaction, the GPT model is updated and optimized based on the user's response rate and conversation duration. Specifically, user feedback and behavior data, such as response rate and conversation duration, are collected to evaluate the effectiveness of conversation content. If certain sales pitch statements or questions are not effective, they are adjusted or replaced to improve pertinence and satisfaction.
[0126] The process of updating and optimizing the GPT model can be achieved through machine learning technology. For example, through supervised learning or reinforcement learning methods, the model can be trained and optimized using user feedback and behavior data. Through continuous iteration and optimization, the performance of the model can be improved, making marketing conversations more accurate and effective.
[0127] In short, using the GPT model to generate sales statements or questions and updating and optimizing them based on user feedback and behavioral data can improve the targeting of marketing conversations and user satisfaction, thereby increasing the company's sales and user loyalty.
[0128] Specifically, based on the user's mixed recommendations, the GPT model is used to generate personalized marketing dialogue content. For example, based on the user's needs and preferences, different dialogue contents such as recommendations, greetings, product introductions, and discount information are generated. These contents are customized according to the user's interests and needs to meet the needs and expectations of different users.
[0129] By using the GPT model to generate conversation content for marketing conversations, we can further improve marketing effectiveness and user experience, and provide users with more accurate and appropriate recommendations and services. At the same time, it will also help companies expand their market share and increase sales.
[0130] The present application also provides an electronic device for commodity marketing AI dialogue combined with GPT, which electronic device includes at least a memory and a processor.
[0131] like Figure 3 As shown, specifically, the memory 301 is used to store the computer executable program of the commodity marketing AI dialogue method combined with GPT;
[0132] The processor 302 is used to call the computer executable program and execute: building a user portrait based on the user's historical purchase records, browsing behaviors, and social media interactions; predicting user needs and preferences based on the user portrait using a machine learning algorithm; screening a first product or service based on the needs, and screening a second product or service based on the preferences, and generating a mixed recommendation based on the first product or service and the second product or service; and generating dialogue content for a marketing dialogue using a GPT model based on the mixed recommendation.
[0133] Furthermore, the processor 301 constructs a user profile based on the user's historical purchase records, browsing behaviors, and social media interactions, including:
[0134] Collect data from users’ purchase history, browsing behavior, and social media interactions;
[0135] Preprocess the collected raw data, including cleaning, normalization, and filling or deleting missing data;
[0136] Extract features related to user profiles from preprocessed data;
[0137] A user profile is constructed using a classifier or a clustering algorithm based on the preprocessed data.
[0138] Furthermore, the processor 301 uses a machine learning algorithm to predict user needs and preferences based on the user portrait, including:
[0139] Extracting features related to user needs and preferences from the user portrait data;
[0140] Select logistic regression, support vector machine, decision tree, random forest or neural network algorithm for prediction, including: training the selected machine learning algorithm to build a prediction model; using an independent data set to evaluate the trained model; tuning the model based on the evaluation results; deploying the trained model to the production environment to achieve real-time prediction of user needs and preferences.
[0141] Furthermore, the processor 301 generates a mixed recommendation according to the first product or service and the second product or service, including:
[0142] Based on user needs, collaborative filtering algorithms are used to filter the first product or service;
[0143] Based on user preferences, use text analysis technology to filter the second product or service;
[0144] According to the matching degree between the first commodity or service and the second commodity or service, a personalized mixed recommendation is generated, and the expression is as follows:
[0145] H q =λ1u i +λ2p j ,u i ∈U,p j ∈P
[0146] Among them, the u i is the matching degree between the first product or service and the demand, the p j is the matching degree between the second commodity and service and the preference, λ1 and λ2 are the weights of the first commodity and service and the second commodity and service, which are determined by the entropy weight method, and U and P are the first commodity and service set and the second commodity and service set.
[0147] Furthermore, the processor 301 generates dialogue content using the GPT model according to the hybrid recommendation to conduct a marketing dialogue, including:
[0148] Based on the features, using the GPT model to generate sales pitch statements or questions to guide users to learn or purchase; and
[0149] The GPT model is updated and optimized based on the user's response rate and conversation duration to improve the pertinence of conversation content and user satisfaction.
Claims
1. A commodity marketing AI dialogue method combined with GPT, characterized in that: include: Build user profiles based on their historical purchase records, browsing behavior, and social media interactions; Based on the user profile, use machine learning algorithms to predict user needs and preferences; Filtering a first product or service according to the demand, filtering a second product or service according to the preference, and generating a mixed recommendation based on the first product or service and the second product or service; Based on the hybrid recommendation, the GPT model is used to generate conversation content for marketing conversation.
2. According to claim 1, the commodity marketing AI dialogue method combined with GPT is characterized in that: Build user profiles based on their purchase history, browsing behavior, and social media interactions, including: Collect data from users’ purchase history, browsing behavior, and social media interactions; Preprocess the collected raw data, including cleaning, normalization, and filling or deleting missing data; Extract features related to user profiles from preprocessed data; A user profile is constructed using a classifier or a clustering algorithm based on the preprocessed data.
3. According to claim 1, the commodity marketing AI dialogue method combined with GPT is characterized in that: Based on the user profile, use machine learning algorithms to predict user needs and preferences, including: Extracting features related to user needs and preferences from the user portrait data; Select logistic regression, support vector machine, decision tree, random forest or neural network algorithm for prediction, including: training the selected machine learning algorithm to build a prediction model; using an independent data set to evaluate the trained model; tuning the model based on the evaluation results; deploying the trained model to the production environment to achieve real-time prediction of user needs and preferences.
4. According to claim 1, the commodity marketing AI dialogue method combined with GPT is characterized in that: Generating a mixed recommendation based on the first product or service and the second product or service includes: Based on user needs, use collaborative filtering algorithms to filter the first product or service; Based on user preferences, use text analysis technology to select the second product or service; According to the matching degree between the first commodity or service and the second commodity or service, a personalized mixed recommendation is generated, and the expression is as follows: H q =λ1u i +λ2p j ,u i ∈U,p j ∈P Among them, the u i is the matching degree between the first product or service and the demand, the p j is the matching degree between the second commodity and service and the preference, λ1 and λ2 are the weights of the first commodity and service and the second commodity and service, which are determined by the entropy weight method, and U and P are the first commodity and service set and the second commodity and service set.
5. According to claim 1, the commodity marketing AI dialogue method combined with GPT is characterized in that: Based on the hybrid recommendation, the GPT model is used to generate conversation content for marketing conversation, including: Based on the features, using the GPT model to generate sales pitch statements or questions to guide users to learn or purchase; and The GPT model is updated and optimized based on the user's response rate and conversation duration to improve the pertinence of conversation content and user satisfaction.
6. An electronic device for commodity marketing AI dialogue combined with GPT, characterized in that: include: A memory for storing a computer executable program of the commodity marketing AI dialogue method combined with GPT as described in any one of claims 1 to 5; A processor is used to call the computer executable program and execute: building a user portrait based on the user's historical purchase records, browsing behaviors, and social media interactions; predicting user needs and preferences based on the user portrait using a machine learning algorithm; screening a first product or service based on the needs, and screening a second product or service based on the preferences, and generating a mixed recommendation based on the first product or service and the second product or service; and generating dialogue content for marketing dialogue using a GPT model based on the mixed recommendation.
7. The electronic device for commodity marketing AI dialogue combined with GPT according to claim 6 is characterized in that: The processor builds a user profile based on the user's historical purchase records, browsing behavior, and social media interactions, including: Collect data from users’ purchase history, browsing behavior, and social media interactions; Preprocess the collected raw data, including cleaning, normalization, and filling or deleting missing data; Extract features related to user profiles from preprocessed data; A user profile is constructed using a classifier or a clustering algorithm based on the preprocessed data.
8. The electronic device for commodity marketing AI dialogue combined with GPT according to claim 6 is characterized in that: The processor predicts user needs and preferences using a machine learning algorithm based on the user profile, including: Extracting features related to user needs and preferences from the user portrait data; Select logistic regression, support vector machine, decision tree, random forest or neural network algorithm for prediction, including: training the selected machine learning algorithm to build a prediction model; using an independent data set to evaluate the trained model; tuning the model based on the evaluation results; deploying the trained model to the production environment to achieve real-time prediction of user needs and preferences.
9. The electronic device for commodity marketing AI dialogue combined with GPT according to claim 6 is characterized in that: The processor generates a mixed recommendation based on the first product or service and the second product or service, including: Based on user needs, use collaborative filtering algorithms to filter the first product or service; Based on user preferences, use text analysis technology to select the second product or service; According to the matching degree between the first commodity or service and the second commodity or service, a personalized mixed recommendation is generated, and the expression is as follows: H q =λ1u i +λ2p j ,u i ∈U,p j ∈P Among them, the u i is the matching degree between the first product or service and the demand, the p j is the matching degree between the second commodity and service and the preference, λ1 and λ2 are the weights of the first commodity and service and the second commodity and service, which are determined by the entropy weight method, and U and P are the first commodity and service set and the second commodity and service set.
10. The electronic device for commodity marketing AI dialogue combined with GPT according to claim 6, characterized in that: The processor generates dialogue content using the GPT model according to the hybrid recommendation to conduct a marketing dialogue, including: Based on the features, using the GPT model to generate sales pitch statements or questions to guide users to learn or purchase; and The GPT model is updated and optimized based on the user's response rate and conversation duration to improve the pertinence of conversation content and user satisfaction.