E-commerce AI interaction method and system fusing user intention and dynamic portrait

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an e-commerce AI interaction method and system fusing user intention and dynamic portray.The method comprises the steps that user query information and explicit and implicit feedback information are obtained at first, then the user query information is input into a preset intention detection and guide module, a user demand feature vector is generated, and then the user demand feature vector is input into a dynamic portrayal database; the method comprises the following steps: acquiring a user demand feature vector, inputting the display-implicit feedback information into a preset display-implicit feedback channel, then carrying out weighted analysis to obtain a user portrait, and finally, responding to purchase confirmation of the user query information, inputting the user demand feature vector and the user portrait into a preset humanistic care generation module, and outputting a care response statement. Compared with the prior art, the method has the advantages that the more comprehensive user portraits can be utilized to generate the care type response statements with the better estrus sharing effect, and therefore the technical problem that in the prior art, the AI assistant interaction experience is poor is solved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology. More specifically, this invention relates to an e-commerce AI interaction method and system that integrates user intent and dynamic profiles. Background Technology

[0002] With the advent of the artificial intelligence era, e-commerce shopping has gradually adopted AI assistants to replace human service providers. These AI assistants typically possess multiple functions such as intelligent product recommendations, user care, and intelligent dialogue. Most existing AI assistants use algorithms such as TF-IDF (Word Frequency-Inverse Document Frequency) or BM25 (Best Match 25th Generation) to intelligently match and recommend products based on user input, and utilize NLP (Neural Linguistic Programming) technology to identify user emotional states, dynamically adjust communication strategies, and provide regular user care services. However, limited by the underlying logic of their algorithms / models, these technologies have the following drawbacks: (1) Existing AI assistants can only respond to users’ explicitly stated needs and have difficulty supporting fuzzy queries. This can lead to the loss of key dimensions during the query process, resulting in recommendation results that do not meet customer expectations. For example, when a user says “I want a piece of clothing suitable for summer,” the aforementioned AI assistants usually cannot proactively ask about the budget, user preferences, and usage scenarios, often resulting in broad and inaccurate recommendation results.

[0003] (2) Existing AI assistants rely too much on users’ historical explicit behaviors to build user profiles, ignoring the deep needs and occasional interests implied in user conversations. They cannot distinguish between users’ “temporary needs” and “long-term preferences”, which can easily lead to the recommendation results being trapped in an “information cocoon” and users having fewer opportunities to discover new products.

[0004] (3) Existing AI assistants automatically stop interacting after the user completes the transaction, lacking understanding of the user's purchasing motivation and emotional response. For example, after purchasing birthday gifts, souvenirs and other emotionally valuable items, users do not receive empathetic feedback, making it difficult to build brand loyalty and repeat purchase intentions.

[0005] (4) For some sensitive products, such as medical supplies or funeral supplies, existing AI assistants lack special processing mechanisms, which may trigger inappropriate smart replies, violate user privacy expectations and damage trust in the product supplier brand.

[0006] In summary, existing AI assistants suffer from poor interactive experiences due to the aforementioned technical deficiencies. Summary of the Invention

[0007] To address the aforementioned technical problem of poor interactive experience in existing AI assistants, this invention discloses an e-commerce AI interaction method and system that integrates user intent and dynamic profiles.

[0008] In a first aspect, this invention discloses an e-commerce AI interaction method that integrates user intent and dynamic profiles, comprising: Obtain user query information and explicit / implicit feedback information; Input the user's query information into the preset intent detection and guidance module to generate a user demand feature vector; After inputting explicit and implicit feedback information into preset explicit and implicit feedback channels, weighted analysis is performed to obtain user profiles; In response to a user's purchase confirmation based on their query information, the user's demand feature vector and user profile are input into a preset humanistic care generation module, which then outputs a caring response statement. Beneficial Effects: The method of this invention first generates a user demand feature vector by inputting user query information into a preset intent detection and guidance module, thereby achieving vector conversion of the user's natural language to support subsequent quantitative analysis. Next, the method performs weighted analysis by inputting explicit and implicit feedback information into preset explicit and implicit feedback channels, resulting in a more comprehensive and accurate user profile. Finally, when the user confirms a purchase, the user demand feature vector and user profile are input into a preset humanistic care generation module, outputting a caring response statement. Compared to existing technologies, the method of this invention can utilize a more comprehensive user profile to generate more empathetic and caring response statements, thus solving the technical problem of poor interactive experience in existing AI assistants.

[0009] Preferably, the intent detection and guidance module has a built-in semantic encoding function based on Transformer; the user query information is input into the preset intent detection and guidance module to generate a user demand feature vector, including: Break down user query information into a sequence of user input; The user input sequence is fed into the semantic encoding function to generate a user demand feature vector.

[0010] Preferably, the intent detection and guidance module also incorporates a confidence algorithm; after generating the user demand feature vector, the method of the present invention further includes: The high-dimensional numerical representation of the user demand feature vector is input into the confidence algorithm for calculation to obtain the confidence score; Determine if the confidence level is below the confidence threshold. If so, output guiding questions for missing items in the user demand feature vector.

[0011] Preferably, explicit and implicit feedback information includes explicit feedback information and implicit feedback information; explicit and implicit feedback channels include explicit feedback channels and implicit feedback channels; after the explicit and implicit feedback information is input into the preset explicit and implicit feedback channels, a weighted analysis is performed to obtain a user profile, including: The explicit feedback information is input into the explicit feedback channel for feature extraction to obtain the explicit feedback feature vector; The implicit feedback information is input into the implicit feedback channel for feature extraction to obtain the implicit feedback feature vector; The explicit feedback feature vector and the implicit feedback feature vector are input into a preset weighted fusion algorithm to calculate the user profile.

[0012] The preferred weighted fusion algorithm is as follows:

[0013] In the formula, express User profile vector at any given moment; express User profile vector at any given moment; Indicates having Explicit feedback feature vectors in each dimension; Indicates having Implicit feedback feature vectors in 10 dimensions; This represents the image smoothing parameter, and its value ranges from 0 to 1; Indicates the first The explicit feedback weighting coefficients for each dimension; Indicates the first Implicit feedback weighting coefficients for each dimension.

[0014] Preferably, before confirming the purchase in response to the user's query information, the method of the present invention further includes: Input the user demand feature vector and user profile into the preset product recommendation model to calculate the user's recommendation score for multiple candidate products; The candidate products with the highest recommendation scores are pushed to the user's device.

[0015] Preferably, the humanistic care generation module has a built-in emotional text generation model; by inputting the user demand feature vector and user profile into the preset humanistic care generation module, it outputs caring response statements, including: Based on the user demand feature vector, the product attributes of the final ordered product, and the contextual semantic features of the product attributes, define the user's purchase motivation tags; The purchase motivation tags, their corresponding contextual semantic features, and user profiles are input into the emotional text generation model for parsing and reconstruction, and the caring response statements are output.

[0016] Preferably, after outputting the caring response statement, the method of the present invention further includes: Determine whether the emotional intensity of the caring response statement exceeds the emotional intensity threshold. If so, replace the caring response statement with neutral language.

[0017] Preferably, after outputting the caring response statement, the method of the present invention further includes: Input the caring response statement into the preset compliance and ethics control module, determine whether the caring response statement meets ethical constraints, and if so, use the caring response statement as the final output.

[0018] Secondly, the present invention also discloses an e-commerce AI interaction system that integrates user intent and dynamic profile, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the e-commerce AI interaction method that integrates user intent and dynamic profile described in the first aspect is implemented.

[0019] The beneficial effects of this invention are as follows: (1) Compared with the prior art, the method of the present invention can use a more comprehensive user profile to generate more empathetic and caring response statements, thereby solving the technical problem of poor AI assistant interaction experience in the prior art.

[0020] (2) Compared with the prior art, the method of the present invention can make intelligent recommendations for products by using a more comprehensive user profile, and can explore the deep needs and occasional interests of users, thereby eliminating the "information cocoon".

[0021] (3) Compared with existing technologies, this method can provide more specific feedback questions to address the lack of key information in user query information, thereby improving the customer experience of "product recommendation" and "caring response". Attached Figure Description

[0022] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1 This is a flowchart of the e-commerce AI interaction method that integrates user intent and dynamic profile in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the structure of the e-commerce AI interaction system that integrates user intent and dynamic profile in Embodiment 2 of this method. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0025] Example 1 like Figure 1 As shown, this embodiment discloses an e-commerce AI interaction method that integrates user intent and dynamic profile, including: S10: Obtain user query information and explicit / implicit feedback information.

[0026] In this embodiment, the aforementioned user query information mainly refers to the natural language entered by the user into the search bar or interactive interface of the relevant product application. Explicit and implicit feedback information includes both explicit and implicit feedback information. Explicit feedback information refers to user interaction records in the product platform application, such as likes, ratings, or dislikes, which are directly reflected in the application and are known to the public. Implicit feedback information mainly refers to the user's real-time interactive operation information, including product browsing time, product skipping frequency, and dialogue / search interaction keywords.

[0027] It should be noted that the acquisition and processing of the aforementioned information is limited to the product platform application. When a user logs into the relevant product platform application, the user will be asked to provide authorization for the relevant information, and the aforementioned information will be kept confidential in accordance with the relevant confidentiality agreement. Approximately, the AI ​​assistant obtained using the method of this embodiment can be regarded as a user-specific agent. It is an independently running software module, and the data processing results it generates are not directly integrated into the network environment. Compared with traditional artificial intelligence models, it offers stronger security and confidentiality.

[0028] Preferably, to make the acquisition of explicit feedback information more accurate, the method in this embodiment is also configured with a local differential protection mechanism on the client side: Define genuine user feedback on product categories as The disturbance feedback generated by the device must meet the following conditions:

[0029] In the formula, This represents the actual feedback value; This indicates the reported feedback value; Represents a probability function; Indicate privacy budget and It needs to meet the k-anonymity requirements of the GDPR (General Data Protection Regulation) and Maximum loss; Represents natural numbers.

[0030] S20: Input the user query information into the preset intent detection and guidance module to generate a user demand feature vector.

[0031] In this embodiment, the intent detection and guidance module is a program encapsulation module that has built-in semantic encoding functions and confidence algorithms based on Transformer. Step S20 includes: S21: Break down the user query information into a sequence of user inputs.

[0032] Specifically, in order to encode and represent the natural language input of users through a semantic understanding model, the user query information needs to be broken down into the following user input sequence:

[0033] In the above encoding representation, This represents the user input sequence. Indicates the first Each word.

[0034] S22: Input the user input sequence into the semantic encoding function to generate a user demand feature vector.

[0035] Specifically, the function that converts the above user input sequence into a user demand feature vector is expressed as follows:

[0036] In the formula, Represents the user demand feature vector. This represents a semantic encoding function.

[0037] Through the above steps S21-S22, the user demand feature vector can be specifically quantified to reflect the user's current real purchase expectations.

[0038] Furthermore, considering the possibility of missing key information, after step S22 above, the method of this embodiment further includes: S23: Input the high-dimensional numerical representation of the user demand feature vector into the confidence algorithm to calculate the confidence score.

[0039] Specifically, to implement the above steps, it is necessary to predefine a set of dimensions in the relevant system:

[0040] The aforementioned set of dimensions serves as the confidence criterion for the user demand feature vector. To achieve this high-dimensional numerical representation, the following example illustrates this: For the "budget" item, if a "budget"-related encoding item exists in the user demand feature vector, then the "budget" item is assigned the value "1"; otherwise, it is assigned "0". Correspondingly, the numerical assignments for other dimensions can also be implemented using the same method.

[0041] More specifically, the confidence algorithm described above is as follows:

[0042] In the formula, Indicates the first The confidence level for each dimension ranges from 0 to 1. This represents the Sigmoid activation function; Indicates the first The weight coefficients for each dimension of classification, which can be obtained based on model pre-training or assigned based on expert experience; A high-dimensional numerical representation of the feature vector of user needs; Indicates the first Bias terms for each dimension of classification.

[0043] S24: Determine if the confidence level is lower than the confidence threshold. If so, output guiding questions for missing items in the user demand feature vector.

[0044] It should be explained that in this embodiment, a text mapping is predefined for each item in the user demand feature vector. When it is determined that the confidence level of one of the items is lower than the confidence level threshold, the mapping is used to automatically bring up the corresponding guiding question and provide feedback to the interactive interface.

[0045] Through the above steps S23-S24, when user intent is missing, the method of this embodiment can identify missing intent and provide proactive feedback so that the user can actively clarify their purchase intent, providing more reliable evidence for subsequent output from caring users.

[0046] S30: Input the explicit and implicit feedback information into the preset explicit and implicit feedback channels and perform weighted analysis to obtain the user profile.

[0047] In this embodiment, the explicit and implicit feedback channels include an explicit feedback channel and a implicit feedback channel, which are two different information processing channels that operate in parallel during information processing. Further, step S30 includes: S31: Input the explicit feedback information into the explicit feedback channel for feature extraction to obtain the explicit feedback feature vector.

[0048] S32: Input the implicit feedback information into the implicit feedback channel for feature extraction to obtain the implicit feedback feature vector.

[0049] S33: Input the explicit feedback feature vector and the implicit feedback feature vector into the preset weighted fusion algorithm to calculate the user profile.

[0050] The weighted fusion algorithm mentioned above is specifically as follows:

[0051] In the formula, express User profile vector at any given moment; express User profile vector at any given moment; Indicates having Explicit feedback feature vectors in each dimension; Indicates having Implicit feedback feature vectors in 10 dimensions; This represents the profile smoothing parameter, which ranges from 0 to 1, with an initial value of 0 for new users. Indicates the first The explicit feedback weighting coefficients for each dimension; Indicates the first The implicit feedback weighting coefficients are calculated for each dimension. The weight matrix relating the explicit and implicit feedback weighting coefficients can be generated through pre-training of an artificial intelligence model.

[0052] Through steps S31-S33 above, the method of this embodiment can obtain a more comprehensive and accurate user profile by inputting explicit and implicit feedback information into a preset explicit and implicit feedback channel and then performing weighted analysis.

[0053] After generating the aforementioned user demand feature vector and user profile, intelligent product recommendation can be performed. The specific process for implementing this function is as follows: S400: Input the user demand feature vector and user profile into the preset product recommendation model to calculate the user's recommendation score for multiple candidate products.

[0054] The core algorithm of the aforementioned product recommendation model is:

[0055] In the formula, Indicates user For goods Recommendation score; Represents the cosine similarity function; This represents the product feature vector, whose vector dimensions and dimensional attributes are consistent with the user profile vector. Indicates motivational moderating factors; This represents the similarity calculation function; Indicates user The deep motivation embedding representation, such as "gift" or "personal use", is based on the parsing of the semantic understanding submodule in the product recommendation model; Indicates goods The deep motivational embedding representation, whose source of acquisition is the same as the above. The similarities are not elaborated upon here.

[0056] S401: Push the candidate product with the highest recommendation score to the user.

[0057] Through the above steps S400-S401, the method of this embodiment can obtain more suitable product recommendations based on user profiles, thereby improving user experience.

[0058] S40: In response to the user's purchase confirmation based on the query information, input the user's demand feature vector and user profile into the preset humanistic care generation module, and output a caring response statement. Step S40 above includes: S41: Define user purchase motivation tags based on user demand feature vectors, product attributes of the final ordered goods, and contextual semantic features of product attributes.

[0059] Specifically, after a user completes browsing the products and places an order, the AI ​​assistant in this embodiment will use the user's demand feature vector as a template to construct a corresponding behavioral sequence:

[0060] In the above expression, Represents a sequence of actions. Indicates the first The item sequence is broken down into sub-items. The above behavioral sequence is used to quantify the user's actual purchasing behavior. Its transformation process is similar to steps S21-S22, the difference being that the above... Used to characterize a user's actual purchasing behavior, Used to characterize a user's potential purchasing behavior.

[0061] After constructing the above behavioral sequence, the feature vectors of the behavioral sequence, product attributes, and contextual semantic features need to be input into the motivation evaluation model:

[0062] In the formula, Labels indicating purchase motivation; The function is used to find the conditional probability. Motivational label for finding the maximum value Other variables , and ; Represents a set of motivation categories; Represents a sequence of actions; It represents contextual semantic features, including feature encoding such as time, festival name, and sentiment words.

[0063] It should be explained that the textual representation of the above motivation category set can be {gift-giving, personal use, ..., private}, and these textual representations can be encoded according to the number of motivation category sets. When performing probability calculations, motivation tags can be calculated. The final probability value is obtained by summing the encoding matching degrees between the data and the multidimensional data.

[0064] S42: Input the purchase motivation tag and its corresponding contextual semantic features and user profile into the emotional text generation model for parsing and reconstruction, and output the caring response statement.

[0065] Specifically, the motivation label and other variables with the highest probability are found using the aforementioned motivation assessment model. and Then, these purchase motivation tags, contextual semantic features, and user profiles are input into the sentiment generation model. In the process, a caring response statement is generated, and it is determined whether the emotional intensity of the caring response statement exceeds the emotional intensity threshold. If so, the caring response statement is replaced with a neutral terminology. The expression for step S42 above is:

[0066] In the formula, Statements that express concern or care. This represents an emotion generation model, which can be an existing BERT model, as mentioned above. , and This method automatically guides the model to generate text. Through steps S10-S40, the method of the present invention can utilize a more comprehensive user profile to generate more empathetic and caring response statements, thereby solving the technical problem of poor interactive experience of AI assistants in the prior art.

[0067] Furthermore, after step S40 above, the method of this embodiment further includes: S50: Input the caring response statement into the preset compliance and ethics control module, determine whether the caring response statement meets ethical constraints, and if so, use the caring response statement as the final output.

[0068] Step S50 above relies on an ethical constraint rule engine, the specific expression of which is:

[0069] In the formula, Represents the ethical suppression function; Indicates the actual category of goods purchased; when the above sensitive conditions are triggered, the corresponding emotional output is automatically disabled; This represents a collection of sensitive product categories, which may include medical supplies, funeral supplies, debt services, and psychological counseling; len represents a function for calculating character length. Indicates the character length threshold; This represents the sentiment score calculated using the RoBERTa model, which ranges from -1 to 1; when If the value is 0, execute step S50 above; otherwise, replace the caring response statement with a neutral statement output.

[0070] For neutral statement output, it can be done through a manual interface or by calling standard statements from a pre-defined neutral language library.

[0071] Furthermore, on the server side, the method in this embodiment also aggregates N disturbance feedbacks. For product categories True interest weight Perform unbiased estimation:

[0072] In the formula, This represents the total number of aggregated user feedback items. Indicates the first Disturbance feedback values ​​reported by individual users; Indicates the category of goods The estimated weights; Indicates based on The calculated perturbation probability parameters.

[0073] Based on the aforementioned estimated weights, dynamic updates to user profiles can be supported:

[0074] In the formula, This indicates that at time t+1, the user is in the product category Interest value on This indicates that at time t, the user is in the product category Interest value on; This represents the decaying learning rate. ; This represents a truncation function, used to force a value to be limited to the range of 0-1 to prevent data overflow.

[0075] Based on the above technical description, the method of this embodiment differs from the prior art in the following ways:

[0076] Based on the above differences, from the perspective of improving the user's personal experience, its beneficial effects are mainly reflected in the following four aspects: Firstly, it can reduce invalid searches by proactively guiding users to accurately match real needs, thereby improving the efficiency of product discovery for users.

[0077] Secondly, it can make cross-category recommendations based on motivation rather than simply historical behavior, thereby broadening users' consumption horizons and breaking the information cocoon.

[0078] Third, it can improve the effect of humanistic care and enhance users' emotional connection.

[0079] Fourth, it is applicable to user-defined agents, which are centered on individual users and effectively avoid over-selling and emotional manipulation.

[0080] Example 2 like Figure 2 As shown, this embodiment discloses an e-commerce AI interaction system that integrates user intent and dynamic profile, including a processor and a memory. The memory stores computer program instructions. When the computer program instructions are executed by the processor, the e-commerce AI interaction method that integrates user intent and dynamic profile described in Embodiment 1 is implemented.

[0081] The system in this embodiment also includes other components well known to those skilled in the art, such as communication interfaces. Their settings and functions are known in the art, and therefore will not be described in detail here.

[0082] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions that can be stored or otherwise maintained by such a computer-readable medium.

[0083] In the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.

[0084] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. An e-commerce AI interaction method that integrates user intent and dynamic profiles, characterized in that, include: Obtain user query information and explicit / implicit feedback information; The user query information is input into a preset intent detection and guidance module to generate a user demand feature vector; The explicit and implicit feedback information is input into a preset explicit and implicit feedback channel and then subjected to weighted analysis to obtain a user profile. In response to the purchase confirmation of the user's query information, the user's demand feature vector and the user profile are input into a preset humanistic care generation module, and a caring response statement is output.

2. The e-commerce AI interaction method integrating user intent and dynamic profile as described in claim 1, characterized in that, The intent detection and guidance module has a built-in semantic encoding function based on Transformer; the user query information is input into the preset intent detection and guidance module to generate a user demand feature vector, including: The user query information is broken down into a user input sequence; The user input sequence is input into the semantic encoding function to generate the user demand feature vector.

3. The e-commerce AI interaction method integrating user intent and dynamic profile as described in claim 2, characterized in that, The intent detection and guidance module also includes a built-in confidence algorithm; after generating the user demand feature vector, the method further includes: The high-dimensional numerical representation of the user demand feature vector is input into the confidence algorithm for calculation to obtain the confidence score; Determine whether the confidence level is lower than the confidence threshold. If so, output a guiding question about the missing item in the user demand feature vector.

4. The e-commerce AI interaction method integrating user intent and dynamic profile as described in claim 1, characterized in that, The explicit and implicit feedback information includes explicit feedback information and implicit feedback information; the explicit and implicit feedback channels include explicit feedback channels and implicit feedback channels; The explicit and implicit feedback information is input into a preset explicit and implicit feedback channel and then subjected to weighted analysis to obtain a user profile, including: The explicit feedback information is input into the explicit feedback channel for feature extraction to obtain the explicit feedback feature vector; The implicit feedback information is input into the implicit feedback channel for feature extraction to obtain the implicit feedback feature vector; The explicit feedback feature vector and the implicit feedback feature vector are input into a preset weighted fusion algorithm to calculate the user profile.

5. The e-commerce AI interaction method integrating user intent and dynamic profile according to claim 4, characterized in that, The weighted fusion algorithm is specifically as follows: In the formula, express User profile vector at any given moment; express User profile vector at any given moment; Indicates having Explicit feedback feature vectors in each dimension; Indicates having Implicit feedback feature vectors in 10 dimensions; This represents the image smoothing parameter, and its value ranges from 0 to 1; Indicates the first The explicit feedback weighting coefficients for each dimension; Indicates the first Implicit feedback weighting coefficients for each dimension.

6. The e-commerce AI interaction method integrating user intent and dynamic profile according to claim 1, characterized in that, Prior to purchase confirmation in response to the user query information, the method further includes: The user demand feature vector and the user profile are input into a preset product recommendation model to calculate the user's recommendation score for multiple candidate products. The candidate product with the highest recommendation score is pushed to the user's device.

7. The e-commerce AI interaction method integrating user intent and dynamic profile according to claim 1, characterized in that, The humanistic care generation module has a built-in emotional text generation model; the user demand feature vector and the user profile are input into the preset humanistic care generation module, and the output is a caring response statement, including: Based on the user demand feature vector, the product attributes of the final ordered product, and the contextual semantic features of the product attributes, define the user's purchase motivation tags; The purchase motivation tags, their corresponding contextual semantic features, and user profiles are input into the emotional text generation model for parsing and reconstruction, and the caring response statements are output.

8. The e-commerce AI interaction method integrating user intent and dynamic profile according to claim 1, characterized in that, After outputting the caring response statement, the method further includes: Determine whether the emotional intensity of the caring response statement is higher than the emotional intensity threshold. If so, replace the caring response statement with neutral language.

9. The e-commerce AI interaction method integrating user intent and dynamic profile according to claim 1, characterized in that, After outputting caring response statements, the method further includes: The caring response statement is input into a preset compliance and ethics control module to determine whether the caring response statement meets ethical constraints. If so, the caring response statement is used as the final output.

10. An e-commerce AI interaction system that integrates user intent and dynamic profiles, characterized in that, It includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the e-commerce AI interaction method that integrates user intent and dynamic profile as described in any one of claims 1-9 is implemented.

Images