Virtual human marketing strategy method, system and equipment based on AI personality analysis

Abstract

The invention discloses a virtual human marketing method, system and device based on AI personality analysis, and the method comprises the following steps: collecting interactive dialogue data of a user and a virtual human in real time through a natural language processing technology, and extracting a multi-dimensional personality feature vector of the user through a deep learning model; according to the user personality feature vector, virtual human interaction parameters most matched with user personality communication are generated through a dynamic parameter adjustment model, and an emotional value is provided for the user; analyzing a short-term emotion demand vector of the user based on the real-time dialogue data; the system background constructs an emotion demand value vector matrix of the commodity library. Based on personality features and emotional features of the user, the virtual human can be close to the emotional demand of the user, and the use stickiness of the user to the system is improved; on the basis, the win-win situation of commercial value is achieved, virtual human parameter dynamic adjustment is added to enable the average dialogue duration of the user, more information of the user can be obtained, and the user portrait is more accurate.

Description

Technical Field

[0001] The present invention belongs to the cross - field of artificial intelligence and digital marketing, and particularly relates to a virtual human marketing method, system and device based on AI personality analysis. Background Art

[0002] Currently, artificial intelligence is widely applied in the field of digital marketing, but the existing technologies mainly have the following three technical bottlenecks:

[0003] 1. Timeliness and dimensional defects in user portrait construction:

[0004] Traditional user portrait technologies mainly rely on static data such as historical browsing and purchase records, and use clustering algorithms to generate user classification labels. There are two major limitations in such methods: 1. Data lag: It is impossible to capture the user's immediate psychological state through real - time conversations. For example, a user may show a consumption tendency in the current conversation that is completely different from historical behavior; 2. Dimensional singularity: Existing technologies mostly focus on demographic characteristics (such as age, gender) and behavior preferences (such as price sensitivity), while psychological research has proven that personality traits have a huge impact on product selection during online consumption.

[0005] 2. Matching gap between recommendation systems and psychological needs:

[0006] Current mainstream recommendation algorithms mainly use collaborative filtering (recommending based on the similarity of user group behavior) and content filtering (relying on keyword matching of product attributes), ignoring the emotional value dimension.

[0007] 3. Insufficient adaptability of virtual human interaction systems:

[0008] Existing virtual human systems generally adopt relatively fixed conversation scripts and content, and the style cannot be adjusted according to user characteristics, resulting in a decline in the efficiency of virtual human communication and recommendation. Summary of the Invention

[0009] This application provides an optimization design method for the impeller of a transcritical carbon dioxide centrifugal pump to at least solve the problem that the blade design optimization method provided in the related art has limitations and cannot be applied to the impeller of a transcritical carbon dioxide centrifugal pump.

[0010] A virtual human marketing method based on AI personality analysis, where each user has a virtual human service that fits the user's personality characteristics, including the following steps:

[0011] Step S1: Real - time collect the interaction conversation data between the user and the virtual human through natural language processing technology, and extract the multi - dimensional personality feature vectors of the user through a deep learning model;

[0012] Step S2: Based on the user personality feature vector, generate the virtual human interaction parameters that best match the user personality communication through a dynamic parameter adjustment model, and provide emotional value to the user;

[0013] Step S3: Analyze the short-term emotional demand vector of the user based on real-time conversation data;

[0014] Step S4: The system background constructs an emotional demand value vector matrix of the product library, and the matrix includes multi-dimensional feature vectors of the functional attributes, usage scenarios, and emotional values of the products;

[0015] Step S5: Calculate the matching degree between the user demand vector and the product feature vector through a vector space similarity algorithm, generate a personalized product recommendation sequence, and output it to the user terminal.

[0016] Further, the specific steps of step S1 include:

[0017] Use a pre-trained language model to extract semantic features from the user conversation text. The pre-trained language model includes the BERT or GPT architecture. Build a personality analysis MBTI model based on psychological personality theory, and map the user conversation features to the four-dimensional personality vector space of extraversion vs introversion, sensing vs intuition, logic vs emotion, and planning vs spontaneity. The construction of the four-dimensional personality vector space further includes: setting the weight coefficients of each dimension based on the Myers-Briggs Type Indicator theory (MBTI).

[0018] Further, the dynamic parameter adjustment model in step S2 is implemented using a reinforcement learning framework, specifically including:

[0019] Construct an adversarial generation network (GAN) for virtual human behavior parameters. The parameters include digital representations of language style, response speed, and expression actions. Design a three-dimensional virtual human parameter space, and the dimensions cover affinity, professionalism, and initiative indicators. Use the user interaction duration, number of dialogue turns, and satisfaction feedback as the reward function, and optimize the virtual human parameter generation strategy through the Q-learning algorithm.

[0020] Further, the specific steps of step S3 include:

[0021] Capture the emotional fluctuations during the conversation through a time series analysis module, generate dynamic emotional parameter sub-vectors, and use a pre-trained language model to extract semantic features from the user conversation text to analyze the user's short-term demand parameter sub-vectors.

[0022] The construction of the user's short-term emotional sub-vector further includes:

[0023] Through the correlation analysis of user historical behavior data and personality vectors, the dimension weights are dynamically calibrated. Through the correlation analysis of user historical interaction language data and user historical order placement behavior data, the dimension weights are dynamically calibrated.

[0024] Furthermore, the construction method of the commodity emotional demand value vector matrix in the step S4 includes:

[0025] Based on the commodity description text and user review data, the Doc2Vec algorithm is used to generate commodity semantic feature vectors. Through the knowledge graph technology, the functional attributes, usage scenarios and emotional labels of the commodity are associated to form multi-dimensional feature vectors. A dynamic update mechanism for commodity vectors is established, and the weights of commodity feature vectors are corrected in real time according to user purchase feedback data.

[0026] Furthermore, the vector space similarity algorithm in the step S5 adopts an improved cosine similarity calculation, specifically including:

[0027] Normalize the user demand vector and the commodity feature vector;

[0028] Calculate the comprehensive matching degree through the fusion algorithm of weighted Euclidean distance and cosine similarity, set a threshold to filter out low-matching commodities, and generate a recommendation sequence in descending order of matching degree.

[0029] Furthermore, a virtual human marketing system based on AI personality analysis includes:

[0030] Data acquisition module: used for dialogue text input and audio input;

[0031] Analysis engine module: used to collect the interactive dialogue data between the user and the virtual human in real time, and then extract the multi-dimensional personality feature vectors of the user through a deep learning model;

[0032] Virtual adaptation module: used to generate the virtual human interaction parameters that best match the user's personality and provide emotional value to the user;

[0033] Commodity eigenvalue generation module: used to construct the emotional demand value vector matrix of the commodity library, and the matrix contains multi-dimensional feature vectors of the functional attributes, usage scenarios and emotional values of the commodity;

[0034] Commodity matching engine: used to calculate the matching degree between the user demand vector and the commodity feature vector, generate a personalized commodity recommendation sequence and output it to the user terminal.

[0035] Even further, an electronic device includes:

[0036] Memory,

[0037] Processor, and

[0038] A computer program stored on the memory and executable on the processor, which, when executed by the processor, implements the virtual human marketing method based on AI personality analysis provided above.

[0039] Compared with the related art, a virtual human marketing method, system and device based on AI personality analysis provided by the present application have the following beneficial effects compared with the prior art:

[0040] 1) Long-term user operation: The core of this system is to provide emotional value to users. Based on the user's personality characteristics and emotional characteristics, the virtual human can be close to the user's emotional needs and increase the user's stickiness to the system; on this basis, a win-win situation in realizing commercial value can be achieved.

[0041] 2) Interaction experience optimization: Increasing the dynamic adjustment of virtual human parameters can extend the average conversation duration of users, enabling the acquisition of more information about users and making the user portrait more accurate;

[0042] 3) Improvement of marketing conversion rate: Considering the user's long-term personality characteristics and short-term emotions comprehensively, the accuracy of product recommendation is improved. Description of the Drawings

[0043] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0044] Figure 1 is a flowchart of a virtual human marketing method based on AI personality analysis shown according to an exemplary embodiment;

[0045] Figure 2 is a block diagram of a virtual human marketing system based on AI personality analysis shown according to an exemplary embodiment. Detailed Embodiments

[0046] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be described and explained below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. Based on the embodiments provided by the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0047] Obviously, the accompanying drawings in the following description are merely some examples or embodiments of the present application. For those of ordinary skill in the art, without creative efforts, the present application can also be applied to other similar scenarios based on these drawings. In addition, it can also be understood that although the efforts made in such a development process may be complex and lengthy, for those of ordinary skill in the art related to the content disclosed in the present application, some design, manufacturing, or production changes based on the technical content disclosed in the present application are only conventional technical means and should not be understood as insufficient disclosure of the content of the present application.

[0048] References to "embodiments" in this application mean that the specific features, structures, or characteristics described in connection with the embodiments can be included in at least one embodiment of the application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those of ordinary skill in the art will explicitly and implicitly understand that the embodiments described in this application can be combined with other embodiments without conflict.

[0049] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the ordinary meaning understood by those of ordinary skill in the technical field to which this application belongs. The words such as "a", "an", "one", "the" and the like involved in this application do not indicate a quantity limitation and can represent a singular or plural number. The terms "including", "comprising", "having" and any variations thereof involved in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product or device including a series of steps or modules (units) is not limited to the listed steps or units, but may further include unlisted steps or units, or may further include other steps or units inherent to these processes, methods, products or devices. The terms "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The term "plurality" involved in this application means two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships can exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0050] Embodiment 1

[0051] An embodiment of the present invention provides a virtual human marketing method based on AI personality analysis. Each user has an exclusive virtual human service that matches the user's personality characteristics, including the following steps:

[0052] Step S1: Real-time collect the interactive dialogue data between the user and the virtual human through natural language processing technology, and extract the multi-dimensional personality feature vectors of the user through a deep learning model. Specifically, the user interacts with the virtual human through the user terminal, and the terminal transmits the user's text or audio input to the server; the terminal transmits scene information such as geographical location obtained through longitude and latitude to the server;

[0053] The specific operation of step S1 includes: using a pre-trained language model to extract semantic features from the user dialogue text, and the pre-trained language model includes a BERT or GPT architecture;

[0054] Construct a personality analysis MBTI model based on the psychological personality theory, and map the user dialogue features to the four-dimensional personality vector space of extraversion vs introversion, sensing vs intuition, logic vs emotion, and planning vs spontaneity;

[0055] The construction of the four-dimensional personality vector space further includes: setting the weight coefficients of each dimension based on the Myers-Briggs Type Indicator theory (MBTI)

[0056] In step S1, the user terminal is generally a mobile terminal used by the user, such as a mobile phone and a tablet computer, or it can also be other desktop terminals, such as a computer, etc., which will not be elaborated here. The user terminal needs to have a stable network connection to the server. In the user terminal, the virtual human image rendering and voice / text interaction functions are implemented through an APP or a Web. The virtual human display technology in the front end directly adopts a mature virtual human display solution in the industry, which is not the key point of this solution and will not be described in detail. The front-end software layer communicates with the back-end server in real time through the WebSocket protocol, supports multi-modal input (text, voice, emoji), and there are the following types of input sources in the front end: First, text dialogue, the real-time chat record between the user and the virtual human, which is stored using UTF-8 encoding and supports mixed Chinese and English input; Second, voice data, which is converted into text through automatic speech recognition (ASR) technology, with a sampling rate of 16 kHz and an identification accuracy rate ≥ 95%; Third, behavior logs, which record buried point data such as the user's click frequency, page stay duration, and jump path, and are used to assist in personality analysis.

[0057] Step S2: According to the user personality feature vectors, generate the virtual human interaction parameters that best match the user personality communication through a dynamic parameter adjustment model, and provide emotional value to the user. Specifically, the server extracts the long-term multi-dimensional personality feature vectors and short-term emotion and demand vectors of the user through the historical and current interaction information between the user and the virtual human through a deep learning model;

[0058] The dynamic parameter adjustment model in S2 is implemented using a reinforcement learning framework. The specific operations include:

[0059] Construct an adversarial generative network (GAN) for the behavior parameters of the virtual human. The parameters include the digital representation of language style, response speed, and expression actions.

[0060] Design a three-dimensional virtual human parameter space, with dimensions covering affinity, professionalism, and initiative indicators.

[0061] Use the user interaction duration, number of dialogue turns, and satisfaction feedback as the reward function, and optimize the virtual human parameter generation strategy through the Q-learning algorithm.

[0062] In step S2 of this embodiment, the backend needs to implement a common Web server to complete communication with the mobile terminal, and the information collected from the terminal system. Since the development and deployment of such Web servers are widely used by various Internet service providers and are not the focus of this solution, no more details will be described. The backend needs to implement a user personality analysis engine. A natural language processing (NLP) node cluster orchestrated by Kubernetes is used. Each node deploys a fine-tuned BERT model to process high-concurrency requests in parallel. The engine analyzes the user's historical interaction data, page click data, order placement data, review data, etc. in the form of background tasks, updates the user's long-term personality characteristics, and can be scheduled to execute during the low-traffic period at 0:00 every day. Considering the long-term stability of the user's long-term personality characteristics, the calculation strategy can be optimized to reduce the calculation amount to save resources. The engine analyzes the user's short-term dialogue data in real-time to extract the short-term emotional demand vector of the personality. Based on the psychological personality theory, a personality analysis MBTI model is constructed. All the user's historical conversations are input into the natural language processing (NLP) node cluster and mapped to the four-dimensional personality vector space V of social orientation S: extraversion vs introversion, perception mode P: sensing vs intuition, decision-making logic D: logic vs emotion, and behavior tendency B: planning vs casualness user =[S, P, D, B]; The user's emotions are designed as 3 dimensions [pleasure degree, anxiety degree, social tendency]. The information of the user's current communication is input into the natural language processing (NLP) node cluster and mapped to the above 3 dimensions to obtain the emotion vector V emotion ; Through the attention mechanism, extract the demand keywords in the conversation (such as "gift giving", "promotion", "urgent", "Women's Day", etc.) and map them to the demand vector V demand =[sociality, timeliness, budget sensitivity, usage scenario].

[0063] Step S3: Analyze the user's short-term emotional demand vector based on real-time conversation data. Specifically, the server corrects the personality feature vector of the virtual human to the personality feature most suitable for the customer according to the user's personality vector feature through the MBTI personality analysis theory, and provides the best emotional value to the user;

[0064] The specific operations of Step S3 include:

[0065] Capture the emotional fluctuations during the conversation through the time series analysis module to generate a dynamic emotional parameter sub-vector;

[0066] Use a pre-trained language model to extract semantic features from the user's conversation text and analyze the user's short-term demand parameter sub-vector.

[0067] The construction of the user's short-term emotional sub-vector further includes:

[0068] Dynamically calibrate the dimension weights through the correlation analysis of the user's historical behavior data and personality vector;

[0069] Dynamically calibrate the dimension weights through the correlation analysis of the user's historical interaction language data and the user's historical order placement behavior data.

[0070] In Step S3, a virtual personality adapter needs to be implemented. In this solution, the core is that the virtual human can obtain useful information that may be converted into benefits through natural communication with the user. Therefore, it is necessary for the virtual human to be able to fit the user's personality and emotional needs as much as possible. Change the personality characteristics of the virtual human to achieve changes in topics, words, and emotions during communication with the user, so that the user can feel pleasure during communication. According to the MBTI personality theory in psychology, based on the principles of similarity and complementarity, the most suitable partner personality can be found for each personality characteristic, such as ISTJ and ISFJ, ENFP and INFJ, etc. Based on this, according to the dynamic feedback in the user's communication, adjust the personality parameters of the virtual human to complete dynamic adaptation.

[0071] Step S4: The system background constructs an emotional demand value vector matrix of the product library. The matrix contains multi-dimensional feature vectors of the product's functional attributes, usage scenarios, and emotional values. Specifically, the server background module calculates the feature vectors of each product based on the product description text, user comment data, and the historical click and transaction data information of all users. Establish a dynamic update mechanism for product vectors to improve the accuracy of product vectors;

[0072] The construction method of the product emotional demand value vector matrix in Step S4 includes:

[0073] Based on the product description text and user comment data, use the Doc2Vec algorithm to generate product semantic feature vectors;

[0074] Associate the functional attributes, usage scenarios, and sentiment tags of products through knowledge graph technology to form multi-dimensional feature vectors;

[0075] Establish a dynamic update mechanism for product vectors, and correct the weights of product feature vectors in real time according to user purchase feedback data.

[0076] In step S4, it is necessary to implement a product feature vector generation module. Based on the distributed index library built by Elasticsearch, store the multi-dimensional feature vectors (functional attributes, sentiment values, usage scenarios) of products, and support real-time query and incremental update. Build a knowledge graph, complete the product-scenario-sentiment triples, and store them using the graph database Neo4j. This module analyzes and calculates product feature vectors in the form of background tasks, which can be executed by scheduling during the low-traffic period at 0:00 every day. The main analysis data of this module are: product price, category, sales volume and other attributes (extracted directly from the database); product description text, encoded into multi-dimensional semantic vectors through the BERT model; user comments, calculate the sentiment polarity (-1 to +1) and intensity value (0-1) using sentiment analysis tools (such as SnowNLP). Fuse the semantic vector, sentiment vector and scenario vector according to the weights of 5:3:2 to generate the final product feature vector.

[0077] In step S4 of this embodiment, it is necessary to implement a product recommendation and display module. Through the foregoing steps, it is possible that the system has not yet collected enough information about the user, or when the system determines that there are no suitable recommended products according to the user's current emotional scenario, then our virtual human does not actively make product recommendations, mainly providing communication and emotional value for the user. This module adopts an improved hybrid similarity algorithm: comprehensively considering the cosine similarity (measuring the direction consistency) and the standardized Euclidean distance (measuring the absolute distance), with weights of 0.6 and 0.4 respectively; the system retains products with a similarity ≥ 0.75, and only selects the top 1 or 2 products for recommendation. When the system obtains the recommended products, it completes the recommendation of the products in the natural conversation with the user in the text and tone that conform to the current virtual human personality characteristics.

[0078] Step S5: Calculate the matching degree between the user demand vector and the product feature vector through the vector space similarity algorithm, generate a personalized product recommendation sequence and output it to the user terminal. Specifically, the server calculates the recommended products through the cosine similarity algorithm based on the user's long-term personality feature vector, short-term emotional demand vector, and product feature vector, and generates a sequence in descending order of the matching degree and returns it to the client. The virtual human completes the recommendation through natural interaction with the user.

[0079] The vector space similarity algorithm in step S5 adopts an improved cosine similarity calculation, specifically including:

[0080] Perform standardization processing on the user demand vector and the product feature vector;

[0081] Calculate the comprehensive matching degree through the fusion algorithm of weighted Euclidean distance and cosine similarity;

[0082] Set a threshold to filter out products with low matching degrees, and generate a recommendation sequence in descending order of matching degrees.

[0083] In summary, it can be seen that the difference between the present invention and the prior art lies in that through continuous communication with users by virtual personalities, the system background can specifically analyze the personality characteristics and emotional characteristics of users, and dynamically adjust the topics, tones, etc. of the virtual human communication in real time, so that the virtual human can provide more emotional value to users, and thereby strengthen the trust of users in virtual characters. On this basis, the system can specifically and in a way that users do not dislike complete recommendation marketing and improve the efficiency of marketing

[0084] Embodiment 2

[0085] The second embodiment of this application provides an analysis engine module: used to collect the interactive dialogue data between users and virtual humans in real time, and then extract the multi-dimensional personality feature vectors of users through a deep learning model;

[0086] Data collection module: used for dialogue text input and audio input;

[0087] The analysis engine module includes:

[0088] Long-term personality vector analysis unit: construct a dialogue analysis model using the Transformer architecture, map the user dialogue text to the MBTI four-dimensional personality vector space (social orientation S, perception mode P, decision-making logic D, behavior tendency B), and the output format is V user =[S, P, D, B];

[0089] Short-term emotion and demand analysis unit:

[0090] Use the Bi-LSTM model to process the dialogue flow in real time, and generate a dynamic emotion vector V every 30 seconds emotion =[pleasure degree, anxiety degree, social tendency] (value range 0-1);

[0091] Extract the demand keywords in the dialogue (such as "gift giving", "promotion", "urgent") through the attention mechanism, and map them to the demand vector V demand =[sociality, timeliness, budget sensitivity];

[0092] Finally, fuse the long-term personality and short-term state to generate a composite user feature vector V final :

[0093]

[0094] where γ is the attenuation coefficient (default 0.7), Indicates a vector concatenation operation.

[0095] Virtual adaptation module: Used to generate virtual human interaction parameters that best match the user's personality communication, provide emotional value to the user. Specifically, when operating, it designs a parameter-adjustable model that supports 16 MBTI personality types, and dynamically optimizes the personality parameters of the virtual human through reinforcement learning;

[0096] Commodity eigenvalue generation module: Used to construct an emotional demand value vector matrix for the commodity library, and the matrix contains multi-dimensional feature vectors of the functional attributes, usage scenarios, and emotional values of the commodity;

[0097] The commodity eigenvalue generation module includes:

[0098] Multi-source data collection:

[0099] Structured data: Commodity SKU attributes (price, category, sales volume);

[0100] Unstructured data: Commodity description text, user reviews, social media related content;

[0101] Feature vector generation process:

[0102] Semantic feature extraction: Use the BERT model to encode the commodity description text to generate a multi-dimensional semantic vector V semantic ;

[0103] Emotional value quantification: Analyze user reviews based on an emotional dictionary (such as BosonNLP), calculate the emotional polarity value P (-1 to +1) and the emotional intensity value I (0-1);

[0104] Knowledge graph association: Construct a commodity-scenario-emotion triple, and extract the scenario-based feature vector V sence = [weekday, holiday, social occasion, personal use];

[0105] Commodity feature vector synthesis: Weight and fuse the above features into a commodity feature vector V product :

[0106] V product = W1·V semantic + W1·[P, I] + W3·V scene

[0107] Among them, W1, W2, and W3 are the weights of each factor in calculating the summary vector, and their sum is 1. Here, W1 = 0.5, W2 = 0.3, W3 = 0.2

[0108] Commodity matching engine: Used to calculate the matching degree between the user demand vector and the commodity feature vector, generate a personalized commodity recommendation sequence and output it to the user terminal:

[0109] Specifically, an improved cosine similarity algorithm is adopted to calculate the user composite vector V final and the matching degree S with the commodity vector V product is as follows:

[0110]

[0111] where α and β are the weights of direction similarity and distance similarity, and their sum is 1. Here, α = 0.6, β = 0.4, and D max is the maximum distance in the vector space.

[0112] Embodiment 3

[0113] Embodiment 3 of the present application provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the virtual human marketing method based on AI personality analysis provided above is implemented.

[0114] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0115] The above embodiments only represent several implementation manners of the present application. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several deformations and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A virtual human marketing method based on AI personality analysis, characterized in that Each user has a virtual human service tailored to their personality traits, including the following steps: Step S1: Real-time collect the interaction dialogue data between the user and the virtual human through natural language processing technology, and extract the multi-dimensional personality feature vectors of the user through a deep learning model; Step S2: According to the user personality feature vectors, generate the virtual human interaction parameters that best match the user's personality communication through a dynamic parameter adjustment model, and provide emotional value to the user; Step S3: Analyze the user's short-term emotional demand vectors based on real-time dialogue data; Step S4: The system background constructs an emotional demand value vector matrix for the product library, and the matrix contains multi-dimensional feature vectors of the product's functional attributes, usage scenarios, and emotional values; Step S5: Calculate the matching degree between the user demand vector and the product feature vector through a vector space similarity algorithm, generate a personalized product recommendation sequence, and output it to the user terminal.

2. The method according to claim 1, characterized in that The specific content of step S1 includes: Use a pre-trained language model to extract semantic features from the user's dialogue text; Build a personality analysis MBTI model based on psychological personality theory, and map the user's dialogue features to a four-dimensional personality vector space of extraversion vs introversion, sensing vs intuition, logic vs emotion, and planning vs spontaneity.

3. The method according to claim 1, wherein In step S2, the dynamic parameter adjustment model is implemented using a reinforcement learning framework, specifically including: Construct an adversarial generation network (GAN) for the virtual human behavior parameters, and the parameters include digital representations of language style, response speed, and expression actions; Design a three-dimensional virtual human parameter space, and the dimensions cover affinity, professionalism, and initiative indicators; Use the user interaction duration, number of dialogue turns, and satisfaction feedback as the reward function, and optimize the virtual human parameter generation strategy through the Q-learning algorithm.

4. The method according to claim 1, characterized in that, The specific content of step S3 includes: Capture the emotional fluctuations during the dialogue through a time series analysis module, and generate dynamic emotion parameter sub-vectors; Use a pre-trained language model to extract semantic features from the user's dialogue text, and analyze the user's short-term demand parameter sub-vectors.

5. The method according to claim 1, wherein The construction method of the product emotional demand value vector matrix in step S4 includes: Based on the product description text and user review data, use the Doc2Vec algorithm to generate product semantic feature vectors; Associate the functional attributes, usage scenarios, and emotional labels of the product through knowledge graph technology to form multi-dimensional feature vectors; Establish a dynamic update mechanism for product vectors, and correct the weights of product feature vectors in real time according to user purchase feedback data.

6. The method according to claim 1, characterized in that, The vector space similarity algorithm in step S5 uses an improved cosine similarity calculation, specifically including: Normalize the user demand vector and the product feature vector; Calculate the comprehensive matching degree through a fusion algorithm of weighted Euclidean distance and cosine similarity; Set a threshold to filter out products with low matching degrees, and generate a recommendation sequence in descending order of matching degree.

7. The method according to claim 2, characterized in that The construction of the four-dimensional personality vector space further includes: Set the weight coefficients for each dimension based on the Myers-Briggs Type Indicator theory (MBTI).

8. The method according to claim 4, wherein The construction of the user's short-term demand parameter sub-vectors further includes: Dynamically calibrate the dimension weights through the correlation analysis of the user's historical behavior data and personality vectors; Dynamically calibrate the dimension weights through the correlation analysis of the user's historical interaction language data and the user's historical order placement behavior data.

9. A virtual human marketing system based on AI personality analysis, which is used for the method described in any one of claims 1-8, and is characterized in that, Including: Data collection module: used for dialogue text input and audio input; Analysis engine module: used to collect the interaction dialogue data between the user and the virtual human in real time, and then extract the multi-dimensional personality feature vectors of the user through a deep learning model; Virtual adaptation module: used to generate the virtual human interaction parameters that best match the user's personality communication and provide emotional value to the user; Commodity feature value generation module: used to construct an emotional demand value vector matrix of the commodity library, and the matrix contains multi-dimensional feature vectors of the functional attributes, usage scenarios and emotional values of the commodity; Commodity matching engine module: used to calculate the matching degree between the user demand vector and the commodity feature vector, generate a personalized commodity recommendation sequence and output it to the user terminal.

10. An electronic device, characterized in that, Including: Memory, Processor, and A computer program stored on the memory and executable on the processor, and when the processor executes the computer program, it implements the virtual human marketing method based on AI personality analysis as described in any one of claims 1 to 9.

Images