User preference analysis system based on shopping cart data

Abstract

The user preference analysis system based on shopping cart data of the present application, the system comprises a data acquisition module for collecting the interactive operation of the user and the shopping cart in real time; a behavior sequence modeling module for receiving the operation event signal, encoding the user operation sequence through a pre-trained time sequence analysis model, and generating a behavior feature vector signal; a preference analysis module for receiving the behavior feature vector signal, generating a dynamic preference signal reflecting the real-time interest tendency and decision state of the user through a multi-dimensional preference calculation model; a strategy generation module for receiving the dynamic preference signal, generating a strategy control signal containing personalized recommendation or marketing instructions according to pre-defined business rules; and a user portrait updating module connected to the data acquisition module and the strategy generation module respectively, receiving the operation event signal and the strategy control signal. The user preference analysis system based on shopping cart data of the present application can solve the problem of shallow, static and one-sided use of user shopping cart data in the prior art.

Description

Technical Field

[0001] This invention relates to the field of e-commerce data processing technology, and more specifically to a user preference analysis system based on shopping cart data. Background Technology

[0002] In e-commerce, a deep understanding of user preferences is crucial for achieving precise marketing, personalized recommendations, and improved conversion rates. Traditional user preference analysis systems largely rely on explicit data such as users' final purchase behavior, browsing history, or search queries. While these methods have achieved some success, they have significant limitations. They often overlook a treasure trove of user intent information—the shopping cart. The data in the shopping cart not only contains the set of products the user explicitly expressed interest in, but also records their hesitation, comparisons, and trade-offs during the decision-making process. Current technologies for utilizing shopping cart data typically remain at a basic level, such as performing simple association rule analysis to discover which products are frequently purchased together. This analysis is static and retrospective, failing to capture the temporal characteristics of user actions and the deeper motivations behind them.

[0003] For example, a user adding an item to their shopping cart and then deleting it could stem from price changes, comparisons with other products, or simply a temporary saving for later reference. Traditional systems cannot effectively distinguish these scenarios, thus missing valuable negative feedback information. Furthermore, existing systems lack the ability to respond to real-time user behavior, failing to intervene promptly and provide incentives such as personalized discounts at crucial decision-making moments, such as when a user shows strong interest in an item but hesitates due to price, thus missing opportunities to improve conversion rates. The user profile update mechanism also often relies on positive purchasing behavior, resulting in a one-sided and outdated profile that fails to reflect the full picture of a user's current interests. Therefore, there is an urgent need in this field for an intelligent system capable of deeply mining dynamic shopping cart behavior data, analyzing real-time changes in user preferences, and instantly transforming the analysis results into operational strategies to overcome the shortcomings of existing technologies and comprehensively improve the intelligent service level of e-commerce platforms. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the purpose of this invention is to provide a user preference analysis system based on shopping cart data, which solves the problem that the existing technology uses user shopping cart data superficially, statically, and one-sidedly, failing to deeply mine users' true intentions and dynamic preferences. This invention generates operation event signals by collecting multi-dimensional interactive operations between users and their shopping carts in real time; it encodes the operation sequences using a time-series analysis model to generate feature vectors representing dynamic behavioral patterns; based on this, it generates dynamic preference signals reflecting real-time interests and decision-making states by calculating the competitive and complementary relationships of goods and users' price sensitivity; it triggers personalized marketing strategies based on these signals; and it innovatively uses non-purchase behavior to negatively calibrate user profiles, thereby constructing a comprehensive, accurate, and dynamically updated user preference analysis system, realizing a transformation from shallow statistics to deep intention mining.

[0005] This invention provides a user preference analysis system based on shopping cart data, comprising:

[0006] The data acquisition module collects user interactions with the shopping cart in real time, generating operation event signals that include operation type, product identifier, and timestamp.

[0007] The behavior sequence modeling module receives operation event signals, encodes user operation sequences through a pre-trained time series analysis model, and generates behavior feature vector signals that represent the user's dynamic behavior patterns.

[0008] The preference analysis module receives behavioral feature vector signals and generates dynamic preference signals that reflect users' real-time interests and decision-making status through a multi-dimensional preference calculation model.

[0009] The strategy generation module receives dynamic preference signals and generates strategy control signals containing personalized recommendations or marketing instructions based on predefined business rules.

[0010] The user profile update module is connected to both the data acquisition module and the strategy generation module. It receives operation event signals and strategy control signals, performs negative calibration and positive enhancement on the user profile based on the user's incomplete transaction behavior and system feedback behavior, generates updated user profile signals, and stores them in the user database.

[0011] In one embodiment of the present invention, the operation types in the operation event signals collected by the data acquisition module further include product addition operation, product deletion operation, product quantity adjustment operation, and product details page jump operation. The data acquisition module captures the above operations by listening to the event bus of the user interface and parsing the payload of the application interface call, thereby forming a complete underlying data source that records the user's intention change process, laying the foundation for subsequent in-depth analysis of user behavior patterns.

[0012] In one embodiment of the present invention, the pre-trained temporal analysis model in the behavior sequence modeling module is a deep neural network model based on an attention mechanism. This model models the long-term and short-term dependencies between user operations by converting the timestamps in the operation event signals into relative time interval embedding vectors and fusing them with the product identifier embedding vectors. Finally, it outputs a fixed-dimensional behavior feature vector signal that incorporates the contextual information of the user's entire session history.

[0013] In one embodiment of the present invention, the multi-dimensional preference calculation model in the preference analysis module includes a product competition relationship calculation unit and a product complementarity relationship calculation unit. The competition relationship calculation unit identifies potential substitution relationships by analyzing product pairs that are frequently added to but not purchased in different users' shopping carts, while the complementarity relationship calculation unit strengthens the complementary product association by analyzing product combinations that are jointly added, jointly retained, and ultimately purchased. The resulting dynamic preference signal encodes the aforementioned complex product relationships.

[0014] In one embodiment of the present invention, the preference analysis module further includes a price sensitivity assessment unit. This unit calculates the user's price sensitivity coefficient for a specific category or a specific product by analyzing the temporal correlation between the user's deletion operation and the product price change events recorded by the system. This coefficient is then incorporated as an important dimension into the dynamic preference signal, thereby enabling the system to determine whether the user's decision is mainly driven by price factors.

[0015] In one embodiment of the present invention, the strategy generation module predefines business rules including real-time discount trigger rules, inventory shortage reminder rules, and personalized product recommendation rules. The module analyzes the user decision-making state and price sensitivity contained in the dynamic preference signal, matches the most suitable business rules, and generates a highly timely and targeted strategy control signal to guide the user to complete the purchase decision.

[0016] In one embodiment of the present invention, the process of generating policy control signals by the policy generation module is a dynamic feedback loop. It will adjust the content and intensity of the current output policy in real time according to the user's subsequent interaction with the previous policy. If the user performs a positive operation after the policy is executed, the policy will be maintained or fine-tuned. If the user performs a negative operation or ignores the policy, a backup policy scheme will be activated, thereby achieving adaptive optimization of the policy.

[0017] In one embodiment of the present invention, when the user profile update module performs negative calibration on the user profile, it will focus on analyzing the operation event signal of the user actively deleting the product, and determine the reason for deletion based on the dynamic preference signal output by the preference analysis module. If it is due to price sensitivity, the weight of the price preference tag in the user profile will be adjusted accordingly. If it is due to clear dislike, the weight of the relevant interest tag in the user profile will be reduced.

[0018] In one embodiment of the present invention, when the user profile update module positively enhances the user profile, it includes not only the final purchased products, but also high-intent products that the user has kept in the shopping cart for a long time but has repeatedly viewed, even if not purchased immediately. The system will assign higher weights to the user interest tags corresponding to such products, thereby constructing a dynamic profile that better reflects the user's potential needs and long-term preferences.

[0019] In one embodiment of the present invention, the behavior sequence modeling module, preference analysis module, and strategy generation module are deployed in a microservice architecture and receive signals output by upstream modules asynchronously through a message queue. This design enables the system to have high scalability and fault tolerance. The processing capacity bottleneck or temporary failure of a single module will not cause the entire analysis process to be interrupted, thereby ensuring the reliability of the system in large-scale high-concurrency scenarios.

[0020] The user preference analysis system based on shopping cart data provided by this invention generates operation event signals by collecting multi-dimensional interactive operations between users and shopping carts in real time; it encodes the operation sequences using a time series analysis model to generate feature vectors representing dynamic behavior patterns; based on this, it generates dynamic preference signals reflecting real-time interests and decision-making states by calculating the competitive and complementary relationships of products and users' price sensitivity; it triggers personalized marketing strategies based on these signals; and it innovatively uses non-purchase behavior to negatively calibrate user profiles, thereby constructing a comprehensive, accurate, and dynamically updated user preference analysis system, realizing a transformation from shallow statistics to deep intent mining. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a system architecture diagram for a user preference analysis system based on shopping cart data. Detailed Implementation

[0023] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0024] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0025] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0026] Please see Figure 1 The user preference analysis system based on shopping cart data of the present invention includes: a data acquisition module, which collects user interaction operations with the shopping cart in real time, forming operation event signals including operation type, product identifier, and timestamp; a behavior sequence modeling module, which receives the operation event signals and encodes the user operation sequence through a pre-trained time series analysis model to generate behavior feature vector signals representing the user's dynamic behavior pattern; a preference analysis module, which receives the behavior feature vector signals and generates dynamic preference signals reflecting the user's real-time interest tendencies and decision-making status through a multi-dimensional preference calculation model; a strategy generation module, which receives the dynamic preference signals and generates strategy control signals including personalized recommendations or marketing instructions according to predefined business rules; and a user profile update module, which is connected to the data acquisition module and the strategy generation module respectively, receives the operation event signals and strategy control signals, performs negative calibration and positive enhancement on the user profile based on the user's incomplete transaction behavior and system feedback behavior, generates an updated user profile signal, and stores it in the user database.

[0027] like Figure 1As shown, its core lies in using a series of collaborative technical modules to perform in-depth, dynamic, and sequential analysis of user interactions in the shopping cart of an e-commerce platform. This surpasses traditional static analysis methods, yielding more accurate and real-time insights into user preferences. The system first relies on a data acquisition module, which acts as the sensing end of the entire system. Its responsibility is to continuously and automatically capture all micro-interactions between the user and the shopping cart interface. These interactions go far beyond simple product additions or final purchases; they encompass a complete behavioral spectrum filled with user intent signals. Specifically, this module programmatically listens to the event bus of the user's browser or mobile application front-end interface, intercepting and parsing every application interface call for the shopping cart function. Whenever a user performs an action, such as clicking the "Add to Cart" button, adjusting the product quantity selector, clicking the "Delete" icon, or even navigating back to a product's details page from the shopping cart page, the module instantly captures this event. Subsequently, it extracts key metadata from the event payload. This metadata is structured into operation event signals, whose core elements must include a classification code for the operation type, a unique product identifier of the target product, and a high-precision timestamp. This timestamp is crucial, as it allows a series of discrete operational events to be reorganized by subsequent modules in chronological order, thus forming a meaningful narrative of user behavior. This raw signal stream forms the cornerstone of the entire system analysis, faithfully recording every subtle hesitation, comparison, and change in the user's decision-making process, making it possible to uncover their deeper preferences.

[0028] Furthermore, the operation event signals are then transmitted to the behavior sequence modeling module, which is the technical hub for intelligent analysis in this system. Its task is to transform the original, time-ordered sequence of operation events into a mathematical expression that can condense and represent complex user behavior patterns. The core of this module is a time-series analysis model pre-trained using a large amount of historical user behavior data. This model is not a simple statistical model, but a deep neural network model based on an attention mechanism, such as a variant of the Transformer, which is particularly suitable for this type of task due to its powerful ability to capture long-range dependencies. The module's processing flow begins with the deep encoding of the operation event signals. First, it processes timestamp information, converting the occurrence time of each event into a relative time interval between it and the previous event, and mapping this interval value to a high-dimensional real-number vector, i.e., the time interval embedding vector, through a mathematical embedding function. This aims to allow the model to perceive the pace of user operations. For example, rapidly and continuously adding multiple items may mean that the user is bulk-collecting or comparing, while a product added without being used for a long time may mean that the user is in a state of hesitation or forgetfulness. Simultaneously, the product identifier is transformed into a product embedding vector through another embedding layer. This vector learns to encode the semantic information of the product in the vector space during training. Next, a crucial data fusion step occurs, where the time interval embedding vector is concatenated or added to the product embedding vector to generate a novel fusion vector that simultaneously contains both temporal context and product information to represent each operation event.

[0029] Specifically, this sequence of fused vectors is then fed into the multi-layered structure of the time-series analysis model. The model's self-attention mechanism kicks in, allowing each action event in the sequence (i.e., each fused vector) to interact and be weighted with all other events in the sequence to determine which events are important for understanding the user's current state and which are relatively minor throughout the behavioral history. For example, a user's recent deletion of an item might have a strong attentional association with adding a competitor's item a long time ago, suggesting a comparative decision. Through this mechanism, the model successfully models the complex long-term and short-term dependencies between user actions, understanding the context behind the behavior. Finally, the model encodes (or summarizes) the entire variable-length behavioral sequence into a fixed-dimensional, dense real-number vector—the behavioral feature vector signal. This vector is a powerful distributed representation, like a "behavioral fingerprint." While humans cannot directly interpret it, the values ​​in each dimension encode certain abstract features of the user's behavioral patterns, such as browsing depth, decision-making decisiveness, brand loyalty, and price sensitivity. The output of this signal marks the system's transition from processing raw data to understanding high-level abstract features.

[0030] like Figure 1 As shown, the behavioral feature vector signal is then transmitted to the preference analysis module. This module acts as the "brain" of the system, responsible for interpreting abstract behavioral patterns into concrete and actionable business insights. Internally, it runs a multi-dimensional preference calculation model. This model receives the behavioral feature vector and generates a dynamic preference signal through a series of forward calculations. This signal is a structured output designed to comprehensively reflect the user's real-time interests and decision-making state. Its calculation dimensions include, but are not limited to: the probability distribution of the user's interest intensity in various items in the current shopping cart; the prediction of the user's most likely next action (such as continuing to browse, deleting items, or checking out); and special indicators calculated by internal sub-units. These sub-units include a product competition relationship calculation unit and a product complementarity relationship calculation unit. The function of the competition relationship calculation unit is to identify which products are substitutable in the user's mind. Its algorithm identifies this pattern by analyzing a large number of product pairs in users' shopping carts that are frequently added together but rarely purchased simultaneously. For example, if a user frequently adds the same model of mobile phone from different brands to their shopping cart but only buys one, the system will determine that there is a strong competitive relationship between these products. The complementary relationship calculation unit, on the other hand, seeks out product combinations that are frequently added together, retained together, and ultimately settled together, such as mobile phones and protective cases, to strengthen the association between them. All this information—abstract behavioral characteristics and specific product relationships—is fused and encoded in the final generated dynamic preference signal. This signal is no longer a black-box vector, but a clear set of instructions that can be parsed by downstream modules, indicating "what" the user "wants," "why" they want it, and "how much" they want it, providing a solid data foundation for ultimately achieving precise business intervention. Through the close collaboration of these three core modules, the entire system has achieved a transformation from raw data collection to deep intelligent insights, realizing a revolutionary analysis of user preferences from the surface to the depths, and from static to dynamic.

[0031] Furthermore, the multi-dimensional preference calculation model in the preference analysis module is a key component for in-depth analysis based on behavioral feature vector signals. The core innovation of this model lies in its inclusion of two specialized calculation units: a product competition relationship calculation unit and a product complementarity relationship calculation unit. The competition relationship calculation unit is designed to understand the trade-offs and choices made by users during the selection process. It is not based on simple product attribute similarity, but rather on machine learning algorithms to mine frequently occurring patterns in large-scale historical shopping cart data. Specifically, it statistically analyzes combinations of products that are frequently added to different users' shopping carts but ultimately not purchased together. For example, when many users frequently add smartphones from brand A and brand B to the same shopping cart but only purchase one, this unit can automatically identify a strong competitive or substitutive relationship between the two products and quantify the strength of this relationship as a competition index. Conversely, the complementarity relationship calculation unit focuses on discovering products with synergistic effects. It identifies this pattern by analyzing product combinations that users jointly add to their shopping carts, jointly retain (i.e., never delete), and ultimately checkout together, such as tents and sleeping bags, printers and ink cartridges. This unit calculates a complementary confidence level to quantify the strength of this association. Ultimately, all these complex relationship calculations—which products are competitors, which are partners, and how strong the relationships are—are integrated and encoded as an important dimension into the final dynamic preference signal. This makes the signal not only contain information about what users "might like," but also deep insights into "liking this item over that item" and "liking this item also makes them highly likely to like that item," providing an unprecedentedly rich context for subsequently developing highly targeted strategies. At the system implementation level, this ensures that the above complex analysis process can run stably and efficiently in a large-scale production environment. It specifies that the behavior sequence modeling module, preference analysis module, and strategy generation module are deployed using a microservice architecture, and asynchronous communication between modules is achieved through message queues. This is a highly decoupled and scalable distributed system design. Each core functional module is encapsulated as an independent, self-contained microservice with its own independent computing resources, storage resources, and lifecycle. Collaboration between them is not achieved through direct function calls or synchronous interface requests, but rather through sending and receiving messages to and from a central message queue. For example, after completing the calculation, the behavior sequence modeling module publishes the generated behavior feature vector signal as a message to a specified message topic. The preference analysis module, as a subscriber to this topic, retrieves the message from its assigned queue and processes it. This asynchronous communication mechanism brings two core advantages: high scalability and fault tolerance.In high-concurrency scenarios, if the processing speed of preference analysis becomes a bottleneck, operations personnel can easily launch more instances of the microservice to consume messages in the message queue in parallel, thereby achieving horizontal scaling without modifying any business code or restarting other modules. Regarding fault tolerance, temporary failures of a single module (such as crashes due to insufficient resources) will not interrupt the entire analysis process. Unprocessed messages will remain in the message queue, waiting to be processed after the failed module recovers, thus ensuring no data loss and that the analysis task is eventually completed. This architectural design enables the system of this invention to handle the massive traffic surges faced by e-commerce platforms during promotional periods, ensuring the reliability and continuity of the analysis service and providing a solid technical foundation for the implementation of large-scale commercial applications.

[0032] like Figure 1As shown, the preference analysis module also includes a price sensitivity assessment unit, specifically designed to understand the role of price factors in user decision-making. This unit employs a time-series event correlation analysis method to achieve this goal. It continuously monitors the operation event flow from the data acquisition module, focusing particularly on the key action of "product deletion." Simultaneously, it accesses the system's product price log, which records all product price change events. The core algorithm of this unit calculates the time-series correlation between user deletion operations and specific product price change events. For example, if the system observes that a user repeatedly deletes a product after a price discount ends or after a slight price increase, it determines that the user has a high price sensitivity to this type of product. This analysis can be refined to the granularity of product categories or even individual products, thereby calculating refined price sensitivity coefficients. This coefficient is a highly valuable value, incorporated as a core feature into the dynamic preference signal. In this way, the dynamic preference signal can clearly indicate the user's decision-making state: for example, a user has a strong interest in a laptop in their shopping cart but hesitates because the current price exceeds their psychological expectations. This judgment fundamentally changed the basis of the system's interaction with users, evolving from guessing what users "want" to understanding "why users haven't bought yet," thus providing direct and reliable data support for triggering precise price incentive strategies. The strategy generation module then begins to fulfill its responsibility of transforming data analysis into business action. This module predefines a set of configurable business rules, which directly correspond to different operational objectives. Common rules include real-time discount trigger rules, which generate a personalized coupon instruction immediately when dynamic preference signals indicate that a user has a strong intention for a certain product but is hindered by price sensitivity; inventory shortage reminder rules, which promptly push reminder signals to users when the inventory of a user's preferred product drops below a critical value, using the scarcity effect to promote decision-making; and personalized product recommendation rules, which recommend substitutes, complements, or highly related products that users are more likely to like based on the competitive and complementary relationships parsed from the dynamic preference signals. The module's workflow involves parsing the received dynamic preference signals, extracting key dimensions (such as interest intensity, decision-making stage, price sensitivity, and product relationships), matching them with conditions in the rule base, and finally executing the rule with the highest matching degree to generate corresponding policy control signals. This process ensures that every marketing message pushed to a user or every product recommendation displayed to a user is triggered at a critical moment in their decision-making process, based on their deepest preferences and concerns, thereby achieving a dual improvement in marketing campaign efficiency and user experience.

[0033] Specifically, the strategy generation process was optimized, upgrading it from a static command transmitter to a dynamic feedback loop system with learning capabilities. This mechanism recognizes that the strategy initially presented to the user is not always the optimal solution, and the system needs to be able to adjust in real time based on user feedback. This module continuously monitors a series of subsequent user interactions after the strategy control signal is issued, treating these behaviors as feedback on the strategy's effectiveness. For example, if a user immediately uses a discount coupon and completes payment after receiving it, this is a strong positive feedback signal, and the system will determine that the strategy was successful, maintaining or fine-tuning such strategies in similar future scenarios. Conversely, if the user ignores the discount coupon, or even worse, deletes the product after receiving the recommendation, this is a negative or ineffective feedback signal. In this case, the system will not stick to the original plan but will activate a backup strategy, such as offering a larger discount on top of the original one, or changing the recommended product category. Through this continuous closed loop of perception, decision-making, execution, and feedback, the strategy generation module can continuously perform adaptive optimization, so that the control precision and effectiveness of its output strategy will continue to improve over time, ultimately achieving a perfect match with each user's unique personality and maximizing the probability of business conversion.

[0034] Furthermore, the specific implementation logic of negative calibration in the user profile update module is the core manifestation of this invention's transcendence of traditional methods relying solely on positive feedback. This module receives raw operation event signals from the data acquisition module and policy control signals from the policy generation module. Its calibration process begins with a deep analysis of the key behavior of the user actively deleting products. This module does not simply equate deletion with aversion, but initiates a sophisticated reasoning process. It first cross-compares the timestamp of the deletion operation, product identifier, and other information with the dynamic preference signal generated by the preference analysis module. The rich context contained in the dynamic preference signal, such as the user's price sensitivity coefficient, historical interest intensity in this product category, and patterns in the operation sequence, becomes the core basis for judging the reason for deletion. For example, if the system records that the price discount of the product has just ended before the deletion operation, and the dynamic preference signal shows that the user has high price sensitivity to this type of product, the module will determine with high confidence that the deletion is due to price factors. Conversely, if there is no price change before and after the deletion operation, and the user lacks in-depth browsing behavior of the product, it may be judged as clear dislike. Based on this reasoning, the module will differentiate the tag weights in the central user profile database. If the reason is price sensitivity, the weight of the "price sensitive" tag in the user profile will be strengthened, while the interest weight of the product category or brand may be slightly reduced, but not completely eliminated, as the user's actual need may still exist. If the reason is clear dislike, the weight of interest tags related to the specific attributes of the product (such as brand, specific functions, model) in the user profile will be significantly reduced, ensuring that the subsequent recommendation process can effectively avoid similar products, thereby improving recommendation accuracy and user satisfaction. This negative calibration mechanism based on causal inference makes the user profile no longer a set of interests that only increases and decreases, but a sophisticated dashboard that can dynamically reflect user likes and dislikes and understand the motivations behind them. The positive enhancement mechanism of the user profile update module is innovative in that it also includes strong intention signals that users have not ultimately converted into purchases in the scope of profile construction, thereby capturing users' potential needs and long-term preferences. Traditional user profiles rely almost entirely on final purchase data to strengthen interest tags, but this method ignores a large amount of high-value information in the user's decision-making process. This module's positive reinforcement criteria significantly broaden its data sources, focusing particularly on items that users repeatedly view in their shopping carts, even if not immediately purchased. These items remain in the cart for significantly longer than average, and users repeatedly return to their product details pages by clicking on product images or names. This series of behavioral patterns constitutes a powerful chain of high intent signals.The system interprets this behavior as indicating a sustained and strong user interest in the product. The reason for the failure to purchase may not be a lack of desire, but rather external factors such as insufficient budget, waiting for a larger discount, or the need for more time to consider the decision. Therefore, the user profile update module assigns a higher weight increment to the user interest tags corresponding to such high-intent products. This means that even if the user doesn't purchase this time, the system has already mentally marked it as a highly relevant item for that user. In subsequent browsing sessions, highly similar products, complementary items, or more competitive offers will receive higher recommendation priority. This mechanism greatly optimizes the cold start and interest exploration problems, providing highly relevant content at the beginning of a new session, effectively shortening the user's decision-making path, and building a forward-looking dynamic profile that not only reflects what the user has bought in the past but also predicts what the user wants to buy in the future.

[0035] This invention presents a user preference analysis system based on shopping cart data. It generates operation event signals by collecting multi-dimensional interactions between users and their shopping carts in real time. The system then encodes these operation sequences using a time-series analysis model to generate feature vectors representing dynamic behavioral patterns. Based on this, it calculates the competitive and complementary relationships between products and the user's price sensitivity to generate dynamic preference signals reflecting real-time interests and decision-making states. These signals trigger personalized marketing strategies. Furthermore, the system innovatively uses non-purchase behavior to negatively calibrate user profiles, thereby constructing a comprehensive, accurate, and dynamically updated user preference analysis system, achieving a shift from superficial statistics to deep intent mining.

[0036] Therefore, the user preference analysis system based on shopping cart data of the present invention solves the problem that the existing technology uses user shopping cart data superficially, statically, and one-sidedly, which cannot deeply explore the user's true intentions and dynamic preferences, resulting in inaccurate personalized services and low marketing efficiency.

[0037] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A user preference analysis system based on shopping cart data, characterized in that, include: The data acquisition module collects user interactions with the shopping cart in real time, generating operation event signals that include operation type, product identifier, and timestamp. The behavior sequence modeling module receives operation event signals, encodes user operation sequences through a pre-trained time series analysis model, and generates behavior feature vector signals that characterize the user's dynamic behavior patterns. The pre-trained temporal analysis model in the behavior sequence modeling module is a deep neural network model based on the attention mechanism. This model converts the timestamps in the operation event signal into relative time interval embedding vectors and fuses them with the product identifier embedding vectors to model the long-term and short-term dependencies between user operations. Finally, it outputs a fixed-dimensional behavior feature vector signal that integrates the context information of the user's entire session history. The preference analysis module receives behavioral feature vector signals and generates dynamic preference signals reflecting users' real-time interests and decision-making states through a multi-dimensional preference calculation model. The multi-dimensional preference calculation model in the preference analysis module includes a product competition relationship calculation unit and a product complementarity relationship calculation unit. The competition relationship calculation unit identifies potential substitution relationships by analyzing product pairs that are frequently added to but not purchased in different users' shopping carts, while the complementarity relationship calculation unit strengthens the complementary product association by analyzing product combinations that are jointly added, jointly retained, and ultimately purchased. The dynamic preference signal generated in the end encodes the above-mentioned complex product relationships. The strategy generation module receives dynamic preference signals and generates strategy control signals containing personalized recommendations or marketing instructions based on predefined business rules. The predefined business rules include real-time discount trigger rules, inventory shortage reminder rules, and personalized product recommendation rules. This module analyzes the user's decision-making state and price sensitivity contained in the dynamic preference signals, matches the most suitable business rules, and generates highly timely and targeted strategy control signals to guide users to complete their purchase decisions. The user profile update module is connected to the data acquisition module and the strategy generation module respectively. It receives operation event signals and strategy control signals, performs negative calibration and positive enhancement on the user profile based on the user's incomplete transaction behavior and system feedback behavior, generates an updated user profile signal and stores it in the user database.

2. The user preference analysis system based on shopping cart data according to claim 1, characterized in that, The operation types in the operation event signals collected by the data acquisition module further include product addition operations, product deletion operations, product quantity adjustment operations, and product details page jump operations. The data acquisition module captures the above operations by listening to the event bus of the user interface and parsing the payload of the application interface call, thereby forming a complete underlying data source that records the user's intention change process, laying the foundation for subsequent in-depth analysis of user behavior patterns.

3. The user preference analysis system based on shopping cart data according to claim 1, characterized in that, The preference analysis module also includes a price sensitivity assessment unit. This unit calculates the user's price sensitivity coefficient for a specific category or product by analyzing the temporal correlation between the user's deletion operation and the product price change events recorded by the system. This coefficient is then incorporated as an important dimension into the dynamic preference signal, enabling the system to determine whether the user's decision is primarily driven by price factors.

4. The user preference analysis system based on shopping cart data according to claim 1, characterized in that, The process of generating policy control signals by the policy generation module is a dynamic feedback loop. It will adjust the content and intensity of the current output policy in real time according to the user's subsequent interaction with the previous policy. If the user performs a positive operation after the policy is executed, the policy will be maintained or fine-tuned. If the user performs a negative operation or ignores the policy, a backup policy scheme will be activated, thereby achieving adaptive optimization of the policy.

5. The user preference analysis system based on shopping cart data according to claim 1, characterized in that, When the user profile update module performs negative calibration on the user profile, it focuses on analyzing the operation event signal of the user actively deleting the product, and judges the reason for deletion based on the dynamic preference signal output by the preference analysis module. If it is due to price sensitivity, the weight of the price preference tag in the user profile is adjusted accordingly. If it is due to clear dislike, the weight of the relevant interest tag in the user profile is reduced.

6. The user preference analysis system based on shopping cart data according to claim 1, characterized in that, When the user profile update module positively enhances the user profile, it considers not only the final purchased products, but also high-intent products that users keep in their shopping carts for a long time but repeatedly view, even if they do not buy them immediately. The system will assign higher weights to the user interest tags corresponding to such products, thereby constructing a dynamic profile that better reflects the user's potential needs and long-term preferences.

7. The user preference analysis system based on shopping cart data according to claim 1, characterized in that, The behavior sequence modeling module, preference analysis module, and strategy generation module are deployed in a microservice architecture and receive signals from upstream modules asynchronously through message queues. This design enables the system to have high scalability and fault tolerance. The processing capacity bottleneck or temporary failure of a single module will not cause the entire analysis process to be interrupted, thereby ensuring the reliability of the system in large-scale high-concurrency scenarios.

Images