A large language model wisdom marketing dialogue method, system, device and medium
By analyzing user intent using a large language model and dynamically adjusting promotional strategies, combined with proactive dialogue to optimize the e-commerce marketing interface, this approach solves the problems of misjudgment of user intent and waste of resources in existing technologies, achieving efficient marketing conversion and resource utilization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JIMENG COMPUTER (BEIJING) CO LTD
- Filing Date
- 2026-03-02
- Publication Date
- 2026-06-05
AI Technical Summary
Existing e-commerce marketing dialogue systems fail to combine user behavior time-series characteristics, historical profiles, and real-time status, resulting in a high rate of misjudgment of purchase intent, waste of preferential resources, low redemption rate, and low conversion efficiency.
We employ a large language model-based intelligent marketing dialogue approach, which analyzes user intent through multimodal input, processes users in a tiered manner, dynamically adjusts promotional strategies, optimizes product interfaces through proactive dialogue, and sets standard templates to prevent inappropriate expressions.
It improved product click-through rate and order conversion rate, reduced customer acquisition cost, increased discount redemption rate, reduced marketing waste, and increased sales volume.
Smart Images
Figure CN122152989A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of e-commerce marketing technology, specifically to a large language model-based intelligent marketing dialogue method, system, device, and medium. Background Technology
[0002] Existing e-commerce marketing dialogue systems largely rely on simple semantic matching to identify user needs, failing to incorporate user behavior time-series characteristics, historical profiles, and real-time status. This leads to a high rate of misjudgment of purchase intent, with current technologies generally achieving an accuracy rate below 85%. This results in a waste of resources, such as "not prioritizing high-intent users while over-marketing to low-intent users," directly lowering conversion efficiency. E-commerce industry promotional resources (discount coupons, direct price reductions, gifts, etc.) often adopt a "non-discriminatory distribution" model, failing to consider the dynamic balance between user purchase intentions, product gross profit, and discounted inventory. This results in high-value discounts being claimed but not redeemed by low-intent users, with current redemption rates generally below 45%. Meanwhile, high-intent users abandon purchases due to not receiving suitable discounts, leading to wasted marketing costs and lost conversion opportunities. The lack of proactive dialogue between products and users prevents the timely understanding of user needs, resulting in low marketing efficiency. Summary of the Invention
[0003] (a) Technical problems to be solved
[0004] To address the shortcomings of existing technologies, this invention provides a method, system, device, and medium for intelligent marketing dialogue based on a large language model, solving the problems mentioned in the background section.
[0005] (II) Technical Solution
[0006] To achieve the above objectives, the present invention provides the following technical solution: a large language model-based intelligent marketing dialogue method, specifically comprising the following steps:
[0007] Step 1: Obtain the multimodal input content of e-commerce users, analyze the content using an attention mechanism and a temporal LSTM network, and predict the purchase intention of the e-commerce users. The multimodal input specifically includes text input, voice input, and product screenshot input.
[0008] Step 2: Classify e-commerce users to link them with product discounts and dynamically adjust the product discounts. Specifically, the e-commerce user classification includes high-priority users and general-priority users.
[0009] Step 3: Obtain the browsing content of e-commerce users within a specific range, and initiate an active dialogue based on the browsing content. Adjust the product interface using the content of the active dialogue. The specific range is specifically a time range or a quantity range.
[0010] Step 4: Set a standard template on the product interface and use the standard template to define the compliance of the same product. The standard template specifically prohibits e-commerce violations such as "tenfold compensation for counterfeit products" and "permanent after-sales service".
[0011] Preferably, the e-commerce user purchase intent prediction algorithm in step one includes the following steps:
[0012] P1. Extract user static features, dynamic features, and product features. Static features include historical purchase frequency, average order value, and product category preference. Dynamic features include current session round, number of products viewed, and dwell time. Product features include category popularity, discount level, and inventory tightness. Then, use an LSTM network to perform time-series modeling of user dynamic features to capture the behavioral sequence patterns of browsing, inquiry, and adding to cart.
[0013] P2. A multi-head attention mechanism is introduced to give higher attention to high-weight features, specifically the shopping cart addition action and price inquiry content. The purchase intention confidence and conversion time prediction interval are output through a fully connected layer, and the cross-entropy loss function is used for model training.
[0014] Preferably, the dynamic adjustment of the product discount in step two includes the following steps:
[0015] S1: Calculate the priority of discount allocation based on "purchase intention confidence level × product gross profit × remaining discount inventory". The specific formula is: P = α × C + β × M - γ × (S / S0), where α, β, and γ are weighting coefficients, C is the purchase intention confidence level, M is the product gross profit, S is the current discount inventory, and S0 is the initial discount inventory.
[0016] S2: Then, users with P≥0.7 are positioned as high-priority users and given high-value benefits, such as discounts or direct price reductions. Users with P<0.7 are positioned as low-priority users and given lightweight benefits, such as coupons or free gifts.
[0017] Preferably, the dynamic adjustment of product discounts also includes:
[0018] Dynamic alert for discounted inventory: When discounted inventory is less than or equal to a threshold (specifically 10% of the initial inventory), the priority of discount allocation for low-value users is automatically reduced to prevent discounts from being exhausted.
[0019] Real-time gross profit matching: Automatically adjust the weight coefficient t for low-gross-profit products. Low gross profit is defined as gross profit ≤ 10%. Reduce the allocation ratio of high-value discounts and control marketing costs.
[0020] Dynamic time-based adjustment: Prioritize the allocation of discounts based on the remaining time of the promotion to promote short-term conversion. Specifically, the remaining time is ≤1 hour countdown.
[0021] Preferably, step three, which involves adjusting the product interface using the content of the proactive dialogue, includes the following steps:
[0022] G1: Extract the browsing content i of e-commerce users, where i represents different types of products. After an e-commerce user browses the same product i for more than a set time t, the product interface will send an active dialogue interface, prompting whether to push the same or similar product types. After receiving the dialogue, push or not push, and the active dialogue interface ends.
[0023] G2: After a conversation, the product interface displays the same and similar product types as the pushed product i. After the number of views exceeds the set number x, the system identifies the price range of the product type and adjusts the price range of the pushed product.
[0024] Preferably, in step G1, when the interface prompts whether to push the same or similar product types, the selectable area will only display an "Yes" or "No" operation interface.
[0025] Preferably, the product interface in G1 will initiate a dialogue interface, which will stop initiating dialogue within a time range after continuously performing "No" operations.
[0026] This invention also discloses a large language model intelligent marketing dialogue system for implementing any of the methods described herein, comprising:
[0027] Intent prediction module: Acquires the multimodal input content of e-commerce users, analyzes the content using an attention mechanism and a temporal LSTM network, and predicts the purchase intention of the e-commerce users;
[0028] Tiered processing module: This module tiers e-commerce users, linking them to product discounts and dynamically adjusting the discounts. Specifically, the tiered e-commerce user system includes high-priority users and general-priority users.
[0029] Proactive dialogue module: Acquires browsing content of e-commerce users within a specific range, initiates proactive dialogue based on the browsing content, and adjusts the product interface using the content of the proactive dialogue. The specific range is specifically a time range or a quantity range.
[0030] Standard Definition Module: Set a standard template on the product interface and use the standard template to define the compliance of the same product. The standard template specifically prohibits e-commerce violations such as "tenfold compensation for counterfeit products" and "permanent after-sales service".
[0031] The present invention also discloses an electronic device, including a processor, a memory, and a high-speed communication interface, wherein the memory stores a computer program, and the processor executes the program to implement any of the methods described herein.
[0032] The present invention also discloses a computer-readable storage medium storing a computer program that, when executed by a processor, implements any of the methods described herein.
[0033] (III) Beneficial Effects
[0034] This invention provides a method, system, device, and medium for intelligent marketing dialogue based on a large language model. Compared with existing technologies, it has the following advantages:
[0035] This large language model-based intelligent marketing dialogue method, system, equipment, and medium, through accurate prediction of purchase intent and short-path conversion guidance, increases product click-through conversion rate by 40%-60% and order conversion rate by 35%-55%, reducing e-commerce customer acquisition costs; it also achieves optimal resource utilization efficiency: the dynamic discount scheduling algorithm increases discount redemption rate by more than 50%, avoids discount waste, reduces marketing costs by 25%-40%, and, combined with proactive dialogue content to adjust the product interface, greatly increases product sales. Attached Figure Description
[0036] Figure 1 This is a flowchart illustrating a smart marketing dialogue method using a large language model, as shown in an embodiment of this application.
[0037] Figure 2 This is a block diagram of a large language model-based intelligent marketing dialogue system, as shown in an embodiment of this application. Detailed Implementation
[0038] 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 embodiments of the present invention, and not all embodiments. 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.
[0039] Please see Figures 1-2 The present invention provides a technical solution:
[0040] A smart marketing dialogue method based on a large language model includes the following steps:
[0041] Step 1: Obtain the multimodal input content of e-commerce users, analyze the content using an attention mechanism and a temporal LSTM network, and predict the purchase intention of e-commerce users. The multimodal input specifically includes text input, voice input, and product screenshot input.
[0042] The e-commerce user purchase intent prediction algorithm in step one includes the following steps:
[0043] P1. Extract user static features, dynamic features, and product features. Static features include historical purchase frequency, average order value, and product category preference. Dynamic features include current session round, number of products viewed, and dwell time. Product features include category popularity, discount level, and inventory tightness. Then, use an LSTM network to perform time-series modeling of user dynamic features to capture the behavioral sequence patterns of browsing, inquiry, and adding to cart.
[0044] P2. A multi-head attention mechanism is introduced to give higher attention to high-weight features, specifically the shopping cart addition action and price inquiry content. The purchase intention confidence and conversion time prediction interval are output through a fully connected layer, and the cross-entropy loss function is used for model training.
[0045] Step 2: Classify e-commerce users to link them with product discounts and dynamically adjust product discounts. Specifically, the e-commerce user classification includes high-priority users and general-priority users.
[0046] Step two involves dynamically adjusting product discounts, including the following steps:
[0047] S1: Calculate the priority of discount allocation based on "purchase intention confidence level × product gross profit × remaining discount inventory". The specific formula is: P = α × C + β × M - γ × (S / S0), where α, β, and γ are weighting coefficients, C is the purchase intention confidence level, M is the product gross profit, S is the current discount inventory, and S0 is the initial discount inventory.
[0048] S2: Then, users with P≥0.7 are positioned as high-priority users and given high-value benefits, such as discounts or direct price reductions. Users with P<0.7 are positioned as low-priority users and given lightweight benefits, such as coupons or free gifts.
[0049] Dynamic adjustments to product discounts also include:
[0050] Dynamic alert for discounted inventory: When discounted inventory is less than or equal to a threshold (specifically 10% of the initial inventory), the priority of discount allocation for low-value users is automatically reduced to prevent discounts from being exhausted.
[0051] Real-time gross profit matching: Automatically adjust the weight coefficient t for low-gross-profit products. Low gross profit is defined as gross profit ≤ 10%. Reduce the allocation ratio of high-value discounts and control marketing costs.
[0052] Dynamic time-based adjustment: Prioritize the allocation of discounts based on the remaining time of the promotion to promote short-term conversion. Specifically, the remaining time is ≤1 hour countdown.
[0053] Step 3: Obtain the browsing content of e-commerce users within a specific range, and initiate proactive dialogue based on the browsing content. Use the content of the proactive dialogue to adjust the product interface. The specific range is specifically a time range or a quantity range.
[0054] Step three involves adjusting the product interface using the content of the proactive dialogue, including the following steps:
[0055] G1: Extract the browsing content i of e-commerce users, where i represents different types of products. After an e-commerce user browses the same product i for more than a set time t, the product interface will send an active dialogue interface, prompting whether to push the same or similar product types. After receiving the dialogue, push or not push, and the active dialogue interface ends.
[0056] G2: After a conversation, the product interface displays the same and similar product types as the pushed product i. After the number of views exceeds the set number x, the system identifies the price range of the product type and adjusts the price range of the pushed product.
[0057] When the interface prompts whether to push the same or similar product types in step G1, the selectable area will only show the "Yes" or "No" operation interface.
[0058] In G1, the product interface will send out an active dialogue interface. After the "No" operation is performed continuously, specifically after 2-3 consecutive appearances, the active dialogue will stop within a time range, specifically the next 1-2 hours.
[0059] Step 4: Set a standard template on the product interface and use the standard template to define the compliance of the same products. The standard template specifically prohibits e-commerce violations such as "tenfold compensation for fakes" and "permanent after-sales service".
[0060] Intent prediction module: Acquires multimodal input content from e-commerce users, analyzes the content using an attention mechanism and a temporal LSTM network, and predicts the purchase intent of e-commerce users;
[0061] Tiered processing module: This module categorizes e-commerce users, enabling a linkage between users and product discounts, and dynamically adjusting product discounts. Specifically, the e-commerce user tiers include high-priority users and general-priority users.
[0062] Proactive dialogue module: Obtains browsing content of e-commerce users within a specific range, initiates proactive dialogue based on the browsing content, and adjusts the product interface using the content of the proactive dialogue. The specific range is specifically a time range or a quantity range.
[0063] Standard Definition Module: Set a standard template on the product interface and use the standard template to define the compliance of the same product. The standard template specifically prohibits e-commerce violations such as "tenfold compensation for counterfeit products" and "permanent after-sales service".
[0064] This invention also discloses a large language model-based intelligent marketing dialogue system, comprising:
[0065] Intent prediction module: Acquires the multimodal input content of e-commerce users, analyzes the content using an attention mechanism and a temporal LSTM network, and predicts the purchase intention of the e-commerce users;
[0066] Tiered processing module: This module tiers e-commerce users, linking them to product discounts and dynamically adjusting the discounts. Specifically, the tiered e-commerce user system includes high-priority users and general-priority users.
[0067] Proactive dialogue module: Acquires browsing content of e-commerce users within a specific range, initiates proactive dialogue based on the browsing content, and adjusts the product interface using the content of the proactive dialogue. The specific range is specifically a time range or a quantity range.
[0068] Standard Definition Module: This module sets a standard template on the product interface. This template is used to define the compliance of a uniform product. Specifically, the standard template prohibits e-commerce violations such as "tenfold compensation for counterfeit goods" and "permanent after-sales service."
[0069] The present invention also discloses an electronic device, including a processor, a memory, and a high-speed communication interface, wherein the memory stores a computer program, and the processor executes the program to implement any one of the methods.
[0070] The present invention also discloses a computer-readable storage medium storing a computer program that, when executed by a processor, implements any one of the methods.
[0071] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.
[0072] Case Study: Payment Reminder Scenarios During Major E-commerce Sales Events (618 Shopping Cart Payment Reminders)
[0073] Scenario characteristics: The user has items in their shopping cart but has not placed an order, the discount is about to expire, or there is an intention to compare products with competitors;
[0074] Application of the method of this invention:
[0075] Purchase intent prediction algorithm: Identify "hesitant users" (confidence level 0.6-0.8) and trigger "inventory alert + limited-time offer" payment reminder;
[0076] Dynamic discount scheduling algorithm: Based on the remaining time of the promotion (≤1 hour), the priority of discount allocation is increased, and "stacked coupons" are issued;
[0077] Cross-channel context alignment algorithm: synchronizes the shopping cart status of users across apps, mini-programs, and web pages, and unifies payment reminders;
[0078] Implementation results: Shopping cart redemption rate increased by 58%, the number of rounds of payment reminder conversations was reduced from 5 to 2, and the rate of negative user feedback decreased by 70%.
[0079] With accurate purchase intent prediction and short-path conversion guidance, product click-through rate increases by 40%-60%, order conversion rate increases by 35%-55%, and e-commerce customer acquisition costs are reduced; resource utilization efficiency is optimized: dynamic discount scheduling algorithm increases discount redemption rate by more than 50%, avoids discount waste, reduces marketing costs by 25%-40%, and adjusts product interface with proactive dialogue content, greatly increasing product sales.
[0080] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
[0082] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems and methods according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0083] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.
Claims
1. A large language model-based intelligent marketing dialogue method, characterized in that, Specifically, the following steps are included: Step 1: Obtain the multimodal input content of e-commerce users, analyze the content using an attention mechanism and a temporal LSTM network, and predict the purchase intention of the e-commerce users. The multimodal input specifically includes text input, voice input, and product screenshot input. Step 2: Classify e-commerce users to link them with product discounts and dynamically adjust the product discounts. Specifically, the e-commerce user classification includes high-priority users and general-priority users. Step 3: Obtain the browsing content of e-commerce users within a specific range, and initiate an active dialogue based on the browsing content. Adjust the product interface using the content of the active dialogue. The specific range is specifically a time range or a quantity range. Step 4: Set a standard template on the product interface and use the standard template to define the compliance of the same product. The standard template specifically prohibits e-commerce violations such as "tenfold compensation for counterfeit products" and "permanent after-sales service".
2. The intelligent marketing dialogue method based on a large language model according to claim 1, characterized in that: The e-commerce user purchase intent prediction algorithm in step one includes the following steps: P1. Extract user static features, dynamic features, and product features. Static features include historical purchase frequency, average order value, and product category preference. Dynamic features include current session round, number of products viewed, and dwell time. Product features include category popularity, discount level, and inventory tightness. Then, use an LSTM network to perform time-series modeling of user dynamic features to capture the behavioral sequence patterns of browsing, inquiry, and adding to cart. P2. A multi-head attention mechanism is introduced to give higher attention to high-weight features, specifically the shopping cart addition action and price inquiry content. The purchase intention confidence and conversion time prediction interval are output through a fully connected layer, and the cross-entropy loss function is used for model training.
3. The intelligent marketing dialogue method based on a large language model according to claim 1, characterized in that: Step two, which involves dynamically adjusting the product discounts, includes the following steps: S1: Calculate the priority of discount allocation based on "purchase intention confidence level × product gross profit × remaining discount inventory". The specific formula is: P = α × C + β × M - γ × (S / S0), where α, β, and γ are weighting coefficients, C is the purchase intention confidence level, M is the product gross profit, S is the current discount inventory, and S0 is the initial discount inventory. S2: Then, users with P≥0.7 are positioned as high-priority users and given high-value benefits, such as discounts or direct price reductions. Users with P<0.7 are positioned as low-priority users and given lightweight benefits, such as coupons or free gifts.
4. The intelligent marketing dialogue method based on a large language model according to claim 3, characterized in that: The dynamic adjustment of product discounts also includes: Dynamic alert for discounted inventory: When discounted inventory is less than or equal to a threshold (specifically 10% of the initial inventory), the priority of discount allocation for low-value users is automatically reduced to prevent discounts from being exhausted. Real-time gross profit matching: Automatically adjust the weight coefficient t for low-gross-profit products. Low gross profit is defined as gross profit ≤ 10%. Reduce the allocation ratio of high-value discounts and control marketing costs. Dynamic time-based adjustment: Prioritize the allocation of discounts based on the remaining time of the promotion to promote short-term conversion. Specifically, the remaining time is ≤1 hour countdown.
5. The intelligent marketing dialogue method based on a large language model according to claim 1, characterized in that: Step three, which involves adjusting the product interface using the content of the proactive dialogue, includes the following steps: G1: Extract the browsing content i of e-commerce users, where i represents different types of products. After an e-commerce user browses the same product i for more than a set time t, the product interface will send an active dialogue interface, prompting whether to push the same or similar product types. After receiving the dialogue, push or not push, and the active dialogue interface ends. G2: After a conversation, the product interface displays the same and similar product types as the pushed product i. After the number of views exceeds the set number x, the system identifies the price range of the product type and adjusts the price range of the pushed product.
6. The intelligent marketing dialogue method based on a large language model according to claim 5, characterized in that: In step G1, when the interface prompts whether to push the same or similar product types, the selectable area will only display an "Yes" or "No" operation interface.
7. The intelligent marketing dialogue method based on a large language model according to claim 5, characterized in that: The product interface in G1 will initiate a dialogue interface. After continuously performing the "No" operation, the interface will stop initiating the dialogue within a certain time range.
8. A large language model-based intelligent marketing dialogue system, characterized in that: To implement the method according to any one of claims 1-7, comprising: Intent prediction module: Acquires the multimodal input content of e-commerce users, analyzes the content using an attention mechanism and a temporal LSTM network, and predicts the purchase intention of the e-commerce users; Tiered processing module: This module tiers e-commerce users, linking them to product discounts and dynamically adjusting the discounts. Specifically, the tiered e-commerce user system includes high-priority users and general-priority users. Proactive dialogue module: Acquires browsing content of e-commerce users within a specific range, initiates proactive dialogue based on the browsing content, and adjusts the product interface using the content of the proactive dialogue. The specific range is specifically a time range or a quantity range. Standard Definition Module: Set a standard template on the product interface and use the standard template to define the compliance of the same product. The standard template specifically prohibits e-commerce violations such as "tenfold compensation for counterfeit products" and "permanent after-sales service".
9. An electronic device, characterized in that, The device includes a processor, a memory, and a high-speed communication interface. The memory stores a computer program, and the processor executes the program to implement the method according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The device contains a computer program that, when executed by a processor, implements the method according to any one of claims 1-7.