A local AI consumption risk assessment method and system

Abstract

This invention discloses a local AI-based consumer risk assessment method, system, and computer-readable storage medium, belonging to the fields of artificial intelligence and consumer risk control technology. It loads an AI model and rule base locally on the user's terminal, collects publicly available consumer data in a non-intrusive manner, performs multi-dimensional risk analysis such as evaluating authenticity and price reasonableness, and outputs real-time warning information. All data is stored locally with encryption and is not uploaded to the cloud. The various modules of the system work collaboratively to achieve a complete risk assessment process. This invention addresses the pain points of existing technologies such as privacy leaks, poor compliance, and response delays. It has advantages such as full-scenario adaptability and low power consumption, complies with domestic and international data security regulations, has high technical barriers, can support global deployment, and balances risk control effectiveness with user privacy protection.

Description

Technical Field

[0001] This invention belongs to the fields of artificial intelligence, data security and consumer risk control technology. Specifically, it relates to an AI-based consumer risk identification method and system based on local computing on the user terminal. It is applicable to all consumption scenarios such as online shopping, food delivery, instant retail, and in-store consumption. It solves the pain points of risk identification, privacy protection and rights protection assistance in the consumption process. Moreover, it does not rely on the cloud and does not leak user privacy throughout the process, and complies with domestic and international data security regulations. Background Technology

[0002] With the widespread adoption of online and offline consumption scenarios such as e-commerce, food delivery, instant retail, and in-store shopping, users face numerous risks and inconveniences during the consumption process, mainly in the following aspects: First, false information, such as fake reviews and order-brushing practices on e-commerce platforms, makes it difficult for users to distinguish the true quality of goods, making them vulnerable to being scammed and suffering losses; Second, privacy risks, as existing consumer risk control tools mostly rely on cloud computing, requiring the uploading of sensitive information such as user consumption data and browsing history, posing a risk of data leakage; Third, insufficient compliance, as some tools obtain internal data from third-party platforms through web scraping, easily triggering the platform's anti-scraping mechanism and facing compliance risks; Fourth, high barriers to rights protection, as users lack professional and compliant communication skills when encountering consumer disputes, and the rights protection process is cumbersome; Fifth, existing risk control tools are mostly deployed in the cloud, resulting in high response latency and inability to adapt to the low-power, low-memory operating requirements of terminals, leading to a poor user experience. Existing consumer risk control solutions largely rely on cloud servers for data processing and analysis. This not only poses risks of privacy breaches but is also limited by network conditions, resulting in slow response times and an inability to provide real-time risk warnings. While some solutions attempt local processing, they suffer from incomplete technical solutions, limited risk identification dimensions, and inadequate privacy protection, failing to comprehensively address various pain points in the consumer process. Furthermore, existing solutions lack systematic local computing and privacy protection designs, failing to balance risk control effectiveness with user privacy and security, and are ill-suited to the diverse needs of various consumption scenarios, thus failing to form a complete closed loop for consumer risk protection. To address the shortcomings of existing technologies, there is an urgent need for an AI-powered consumer risk assessment method and system that is fully localized, independent of the cloud, does not leak privacy, is compliant and non-intrusive, and can cover all consumer scenarios. This system should be able to achieve real-time risk identification and early warning, protect user privacy and security, lower the threshold for users to protect their rights, and solve the core problem that existing technologies cannot simultaneously address the issues of "risk control effectiveness, privacy and security, compliance, and user experience". Summary of the Invention

[0003] (a) Technical problems to be solved The purpose of this invention is to overcome the shortcomings of existing technologies and provide a local AI consumer risk identification method and system, solving the following core problems existing in the prior art: 1. Existing risk control tools rely on cloud computing, which requires uploading user consumption data and browsing history, posing a risk of privacy leaks; 2. Relying on web crawlers to obtain data from third-party platforms can easily trigger the platforms' anti-crawling mechanisms, resulting in poor compliance and potential restrictions on its use. 3. High response latency prevents real-time risk warnings, making it easy for users to miss opportunities to mitigate risks; 4. Risk identification is based on a single dimension and cannot cover all consumption scenarios such as online shopping, food delivery, instant retail, and in-store consumption; 5. The privacy protection technology is vaguely described, making it impossible to achieve "usable but invisible" local data processing, resulting in insufficient control over user data; 6. The system and methods are not strongly correlated, the module collaboration logic is unclear, and it is difficult to efficiently achieve a closed loop in the risk control process; 7. Poor adaptability, unable to meet the low power consumption and low memory requirements of terminals, affecting the user experience. (II) Technical Solution To achieve the above objectives, this invention provides a local AI-based consumer risk identification method and system. The entire process is based on local computation on the user's terminal, without relying on cloud servers or uploading any sensitive user data. It balances risk control effectiveness, privacy and security, compliance, and user experience. The specific technical solution is as follows: 1. A local AI-based method for identifying consumer risks and pitfalls. A local AI-based method for identifying consumer risks and pitfalls, characterized by the following steps: S1. Local Model Loading: The pre-trained NLP risk identification model, price rule engine, advertising compliance detection engine, and lightweight privacy computing module are loaded locally on the user terminal. The model parameters and rule base are stored locally on the user terminal, without relying on cloud synchronization, and support offline updates. Model loading and updating operations can be completed without a network connection, adapting to the low power consumption requirements of the terminal. S2. Local Collection of Consumption Data: By monitoring the user terminal interface and extracting text from publicly accessible pages authorized by the user, at least one type of consumption data is obtained, including product title, product description, user reviews, product price, and product parameters. The consumption data is not uploaded to the cloud server throughout the process, and no data from third-party platforms is obtained through web scraping. Only publicly visible text information is collected to ensure compliance and non-intrusiveness. S3. Multi-dimensional local risk analysis: Based on the locally loaded model and modules, perform local analysis on the collected consumption data. The analysis includes at least one of the following: (1) Evaluation authenticity analysis: Local semantic recognition, template detection, and abnormal feature identification are performed on user evaluation texts to output evaluation authenticity scores, helping users to identify fake evaluations and fraudulent evaluations; (2) Price rationality analysis: Based on the price benchmark library maintained locally on the user terminal, calculate the interval percentile value of the current commodity price, identify price anomaly traps such as "price increase followed by price decrease" and "falsely marked original price", and output the price risk level; (3) Compliance analysis of advertising: detect illegal expressions such as absolute terms, false promises, and exaggerated effects in product titles and descriptions, and output compliance reminders to prevent users from being misled by false advertising; (4) Privacy and security verification: The collected consumption data is locally desensitized, and sensitive information such as mobile phone number and delivery address is blurred to ensure that sensitive user information is not leaked. S4. Risk Scoring and Early Warning Output: Based on the results of the multi-dimensional local risk analysis, a comprehensive risk score of 0-100 is generated, and early warning information is output in real time on the user terminal interface in at least one of the following forms: floating ball, pop-up window, interface marker; the early warning information includes risk type, risk level, and consumption suggestion measures, so that users can quickly understand the risk and avoid it. S5. Local Data Closed Loop: The multi-dimensional local risk analysis results and user-marked product data are encrypted and stored locally on the user's terminal without being uploaded to any cloud server; users can export, delete, and back up the data locally on their terminal, enabling users to have complete control over their personal consumption data and ensuring data privacy and security. Furthermore, the evaluation authenticity analysis described in step S3 specifically includes: performing local word vector matching, templated sentence detection, and identification of fraudulent order feature keywords on user evaluation text, and outputting an evaluation authenticity score of 0-100 points; the fraudulent order feature keywords include at least one of "cashback for positive reviews" and "gifts for follow-up reviews", and potential fraudulent orders are quickly identified through such keywords. Furthermore, the price reasonableness analysis described in step S3 specifically includes: based on the locally maintained price benchmark library, calculating the interval quantile value of the current commodity price in the price benchmark library, identifying at least one price trap such as "price increase followed by price decrease" or "falsely labeled original price", and outputting one of the three price risk levels: normal, slightly high, and artificially high, to assist users in judging the reasonableness of the price. Furthermore, the compliance analysis of the advertising mentioned in step S3 specifically includes: detecting at least one type of illegal content in the product title and product description, such as absolute terms, false promises, and exaggerated efficacy claims, and generating a compliance reminder; the absolute terms include at least one of "most effective", "first", and "never", and the false promises include at least one of "money-back guarantee" and "lifetime warranty". Furthermore, the pre-trained NLP risk identification model mentioned in step S1 is a lightweight semantic recognition model. The lightweight privacy computing module adopts at least one lightweight implementation algorithm among differential privacy and homomorphic encryption to adapt to the low power consumption and low memory operation requirements of user terminals, avoid occupying too many terminal resources, and not affect the user's normal use of other terminal functions. Furthermore, the calculation logic for the comprehensive risk score in step S4 is as follows: evaluation authenticity accounts for 40%, price reasonableness accounts for 30%, and advertising compliance accounts for 30%. The warning information can be viewed through the floating ball on the user terminal to see a detailed risk description, making it convenient for users to quickly understand the risk details and avoidance measures. 2. A local AI-based consumer risk assessment system A local AI-based consumer risk assessment system, characterized by comprising: The local model management module is used to load pre-trained NLP risk identification models, price rule engines, advertising compliance detection engines, and lightweight privacy computing modules locally on the user terminal. It is also responsible for the local updates and version management of the models and modules, supports offline updates, does not rely on the network, and adapts to the low-power operation requirements of the terminal. The local data acquisition module is used to acquire consumption data through user terminal interface monitoring and public text extraction. It does not crawl or upload the consumption data, but only processes it locally on the user terminal to ensure data privacy and security. The multi-dimensional risk analysis module includes sub-modules for evaluating authenticity, pricing reasonableness, advertising compliance, and privacy and security. It is used to perform multi-dimensional risk analysis on locally collected consumer data and output the analysis results and risk levels for each dimension. The risk warning interaction module is used to generate a comprehensive risk score of 0-100 based on the results of multi-dimensional risk analysis. It outputs warning information to the user terminal in real time through at least one of the following forms: floating ball, pop-up window, and interface marker. It supports users to customize warning thresholds and display methods to adapt to different user operating habits. The local data storage module uses AES-256 encryption to store model parameters, rule base, risk analysis results, and user-tagged data. It supports local data export, deletion, and backup without uploading to any cloud server, thus enabling users to have complete control over their personal data. The modules work together to implement all the steps of the aforementioned local AI consumer risk identification method, without relying on the cloud, leaking user privacy, or illegally obtaining data from third-party platforms, thus balancing risk control effectiveness and compliance. Furthermore, the evaluation authenticity submodule uses word vector matching, template matching, and feature engineering to identify fake evaluations locally and output an evaluation authenticity score; the price reasonableness submodule is used to maintain a local price benchmark library, realize price trend analysis and price trap identification, and output a price risk level; the privacy and security submodule is used to perform local de-identification and differential privacy processing on the collected consumer data to ensure that sensitive user information is not leaked. Furthermore, the local model management module supports incremental model updates without reloading all model parameters, reducing the power consumption of the user terminal and avoiding impacting normal user operation; the risk warning interaction module supports dual-modal interaction of voice and text, adapting to the operating habits of users of different ages and lowering the operating threshold. 3. A computer-readable storage medium A computer-readable storage medium is characterized in that it stores a computer program, which, when executed by a processor, implements all the steps of the aforementioned local AI consumer risk identification method; the computer-readable storage medium includes at least one of an SD card, a USB flash drive, a solid-state drive, a portable hard drive, and cloud-encrypted storage, wherein the cloud-encrypted storage is only used for user-initiated data backup and is not used for model calculations or data processing, thus ensuring data privacy and security. (III) Beneficial Effects Compared with the prior art, the present invention has the following advantages: 1. Ultimate Privacy and Security Protection: The entire process is based on local computing on the user's terminal, without relying on cloud servers. It does not upload any sensitive information such as user consumption data, browsing history, or tag data. It adopts AES-256 encrypted storage and lightweight privacy computing algorithms to eliminate the risk of privacy leakage at the source. It complies with the requirements of the Personal Information Protection Law, the Measures for Supervision and Administration of Online Transactions, and overseas regulations such as GDPR, and can be safely deployed overseas. 2. Strong compliance and no usage risk: It adopts a non-intrusive design, does not crawl, does not call third-party platform interfaces, and does not interfere with the platform's transaction process. It only collects publicly visible text information, avoids triggering the platform's anti-crawling mechanism, solves the problem of insufficient compliance of existing tools, and can be used stably for a long time. 3. High real-time performance and rapid response: All model loading, data processing, risk analysis, and early warning output are completed locally on the terminal, with a response time of less than 3 seconds. It does not rely on the network and can realize real-time risk warning, solving the pain point of response delay in existing cloud tools. 4. Full-scenario adaptation and high practicality: It covers all consumption scenarios such as online shopping, food delivery, instant retail, and in-store consumption. Multi-dimensional risk analysis can comprehensively solve various problems encountered by users in the consumption process. At the same time, it is adapted to the low power consumption and low memory operation requirements of terminals, so that elderly and novice users can get started quickly. 5. High technical barriers that are difficult to bypass: The combination of "local AI model + lightweight privacy computing + multi-dimensional risk analysis + module collaboration" forms a unique technical barrier, making it difficult for competitors to circumvent infringement by simply adjusting parameters or modifying the solution, which is in line with Huawei's ironclad rule of "attack and defense" in patent layout. 6. Excellent commercial adaptability and global deployment: The technical solution is adapted to overseas consumer scenarios, complies with overseas data privacy regulations, supports PCT global deployment and overseas licensing monetization, and maximizes both "patent protection and commercial benefits", which is in line with Huawei's patent logic of "global deployment to control the market". Detailed Implementation

[0004] The present invention will be further described in detail below with reference to specific embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention. Example 1: Specific Implementation of Local AI-Based Consumer Risk Detection Method A local AI-based method for identifying consumer risks includes the following steps: S1. Local Model Loading: After the user launches the terminal application, the local model management module automatically loads the pre-trained lightweight NLP semantic recognition model, price rule engine, advertising compliance detection engine, and lightweight differential privacy computing module locally on the user's terminal (mobile terminal such as mobile phone / iPad). The model parameters and rule base are stored in the local storage unit of the terminal, without relying on cloud synchronization, and support offline updates. When updating, only the incremental model package needs to be downloaded, without reloading the entire model, thus reducing the terminal power consumption. S2. Local Collection of Consumer Data: When users browse e-commerce product pages, food delivery merchant pages, fresh food supermarket product pages, or view relevant product information before making a purchase in-store, the local data collection module listens through the terminal interface and extracts publicly visible consumer data such as product titles, product descriptions, user reviews, product prices, and product parameters. During the collection process, no third-party platform interfaces are called, no web crawlers are used to obtain internal platform data, only publicly displayed text information is extracted, and no data is uploaded to the cloud throughout the entire process. S3. Multi-dimensional Local Risk Analysis: The multi-dimensional risk analysis module performs local analysis on the collected consumption data, specifically including: (1) Evaluation authenticity analysis: The evaluation authenticity submodule performs local word vector matching on user evaluation text, detects fraudulent keywords such as "cashback for positive reviews" and "gifts for follow-up reviews", identifies templated evaluation sentences, outputs an evaluation authenticity score of 80 points, and reminds users that "the evaluation is suspected of being fraudulent and should be used with caution". (2) Price rationality analysis: The price rationality submodule calls the locally maintained price benchmark library (stores price data of the same category of goods for the past 180 days), calculates the interval percentile value of the current commodity price in the benchmark library, identifies the commodity as having a "price trap of rising first and then falling", and outputs the "too high" price risk level; (3) Advertising compliance analysis: The advertising compliance submodule detects the absolute word "most effective" in the product title and outputs a compliance reminder "Absolute advertising exists, be cautious when placing an order"; (4) Privacy and security verification: The privacy and security submodule performs local obfuscation processing on the user's delivery address contained in the collected data, hiding sensitive information such as building number and house number, to ensure that sensitive information is not leaked. S4. Risk Scoring and Early Warning Output: The risk warning interaction module integrates the above multi-dimensional analysis results and calculates a comprehensive risk score of 75 points based on the weights of "evaluation authenticity 40%, price reasonableness 30%, and advertising compliance 30%". The warning information is output in real time through the terminal's floating ball: "Risk score 75 / 100, order with caution; suspected fake reviews, price is too high, and there is absolute advertising". Users can click on the floating ball to view detailed risk explanations. S5. Local Data Closed Loop: The local data storage module uses AES-256 encryption to store the results of this risk analysis and the "Buy with Caution" label marked by the user locally, without uploading them to the cloud; users can export, delete or back up this data in the terminal settings, and have complete control over their personal data. Example 2: Specific Implementation of a Local AI-Powered Consumer Risk Detection System A local AI-based consumer risk identification system includes the following modules, which work together to implement all the steps of the method described in Example 1: 1. Local Model Management Module: Deployed locally on the user terminal, it is responsible for loading the lightweight NLP semantic recognition model, price rule engine, advertising compliance detection engine, and lightweight differential privacy computing module; it supports incremental model updates without reloading all parameters, reducing terminal power consumption; it maintains model version management to ensure that the model and rule base are updated synchronously, adapting to new consumer traps. 2. Local Data Acquisition Module: Employing interface monitoring and public text extraction technologies, this module only collects publicly visible consumer data from the terminal interface. It does not crawl, call third-party platform interfaces, or upload any data, ensuring compliance and non-intrusiveness. The collected data includes product titles, descriptions, reviews, prices, parameters, etc., providing data support for subsequent risk analysis. 3. Multi-dimensional risk analysis module: It includes four sub-modules, which respectively realize the identification of evaluation authenticity, price reasonableness analysis, publicity compliance detection, and privacy and security verification. Each sub-module is based on local models and rule bases, completes analysis independently and outputs results, without relying on cloud computing, and has a fast response speed. 4. Risk Warning Interaction Module: Supports three warning formats: floating ball, pop-up window, and interface marker. Users can customize warning thresholds (e.g., trigger a warning when the risk score is ≥70). Supports dual-modal interaction of voice and text. Users can quickly obtain warning information and avoidance suggestions by using the voice command "View Risk". 5. Local Data Storage Module: Utilizes AES-256 encrypted storage to store model parameters, rule bases, risk analysis results, user-tagged data, etc.; supports local data export, deletion, and backup, without uploading to any cloud server, ensuring user data privacy and security, and complying with domestic and international data protection regulations. Example 3: Specific Implementation of a Computer-Readable Storage Medium A computer-readable storage medium, in the form of an SD card, stores a computer program. When executed by a terminal processor, the program implements all the steps of the local AI consumer risk identification method described in Example 1. Users can insert the SD card into terminals such as mobile phones and tablets to launch the application without an internet connection, and complete operations such as risk identification, early warning, and data storage. It also supports cloud-encrypted backup, and the backup data is only used for user self-recovery and is not used for model calculation and data processing, further ensuring data security.

Claims

1. A local AI-based method for identifying consumer risks and pitfalls, characterized in that, Includes the following steps: S1. Local loading of models and engines: The risk identification model, price rule engine, publicity compliance detection engine and privacy computing module are loaded locally on the user terminal; the parameters and rule base of the models and engines are stored locally on the user terminal, supporting offline incremental updates and not relying on cloud synchronization; S2. Local collection of consumption data: Consumption-related data is obtained through listening to user terminal interfaces and extracting text from publicly authorized pages; the collected data is not uploaded to cloud servers and no data from third-party platforms is obtained through web scraping. S3. Multi-dimensional Local Risk Analysis: Based on the model and modules loaded in step S1, local analysis is performed on the collected consumption data. The analysis includes at least one of the following: (1) Evaluation authenticity analysis: Local semantic recognition and feature detection are performed on consumer-related evaluation texts to output evaluation authenticity index; (2) Price rationality analysis: Based on locally maintained price benchmark data, the current consumer prices are compared within a range and anomalies are identified, and price risk indicators are output; (3) Compliance analysis of advertising: Detect illegal statements in consumer-related advertising materials and output compliance reminders; (4) Privacy and security verification: Local desensitization and privacy protection processing are performed on the collected consumption data to prevent the leakage of sensitive user information; S4. Risk Warning Output: Based on the multi-dimensional analysis results of step S3, generate comprehensive risk indicators and output warning information and consumption suggestions in real time on the user terminal interface in a preset format. S5. Local Data Closed-Loop Management: Risk analysis results and user-tagged data are stored locally in an encrypted manner, allowing users to export, delete, and back up the data on their local terminals without transmitting data to any cloud servers or third parties.

2. The local AI consumer risk identification method according to claim 1, characterized in that, The evaluation authenticity analysis described in step S3 specifically includes: performing local semantic matching, template detection, and abnormal feature identification on the evaluation text, and outputting quantifiable indicators of evaluation authenticity; the abnormal features include features related to fraudulent orders and features of repeated evaluations.

3. The local AI consumer risk identification method according to claim 1, characterized in that, The price rationality analysis described in step S3 specifically includes: calculating the interval quantile of the current consumer price based on local price benchmark data, identifying abnormal price fluctuations and false pricing behavior, and outputting the price risk level; the price benchmark data is maintained locally and updated in real time by the user terminal.

4. The local AI consumer risk identification method according to claim 1, characterized in that, The compliance analysis of advertising mentioned in step S3 specifically includes: detecting illegal statements in consumer-related advertising texts, wherein the illegal statements include at least one of absolute terms, false promises, and exaggerated efficacy statements, and generating targeted compliance reminders.

5. The local AI consumer risk identification method according to claim 1, characterized in that, The risk identification model mentioned in step S1 is a lightweight semantic recognition model, and the privacy computing module adopts a lightweight privacy protection algorithm to meet the low power consumption and low memory operation requirements of user terminals.

6. The local AI consumer risk identification method according to claim 1, characterized in that, The output format of the warning information in step S4 includes at least one of the following: a floating ball, an interface marker, and a lightweight pop-up window; the comprehensive risk index adopts a quantitative scoring format and supports user-defined warning thresholds.

7. A local AI-based consumer risk identification system, characterized in that, include: The local model and engine management module is used to load and update the risk identification model, price rule engine, advertising compliance detection engine and privacy computing module locally on the user terminal, and supports offline updates. The local data acquisition module is used to acquire consumption-related data through interface monitoring and public text extraction. It does not crawl or upload data, and only processes it locally on the user terminal. The multi-dimensional risk analysis module is used to perform multi-dimensional local analysis on the collected consumer data based on local models and engines to evaluate its authenticity, price reasonableness, advertising compliance, and privacy security, and output the analysis results for each dimension. The risk warning and interaction module is used to generate comprehensive risk indicators and warning information by integrating multi-dimensional analysis results, outputting them locally on the user terminal in real time, and providing consumption suggestions. The local data storage module is used to encrypt and store model parameters, rule base, risk analysis results, and user-tagged data locally. It supports users to export, delete, and back up locally, without transmitting any data to the cloud. The modules work together to implement all the steps of the local AI consumer risk identification method according to any one of claims 1 to 6.

8. The local AI consumer risk identification system according to claim 7, characterized in that, The multi-dimensional risk analysis module includes an evaluation authenticity analysis submodule, a price reasonableness analysis submodule, a publicity compliance analysis submodule, and a privacy and security verification submodule; each submodule runs independently and collaboratively outputs analysis results, all of which are calculated locally on the user's terminal.

9. The local AI consumer risk identification system according to claim 7, characterized in that, The local model and engine management module supports incremental model updates without reloading all parameters, reducing the power consumption of the user terminal; the local data storage module uses encrypted storage to ensure data security.

10. The local AI consumer risk identification system according to claim 7, characterized in that, The risk warning interaction module supports dual-modal interaction of voice and text, adapting to different user operating habits; the local data acquisition module is configured with an anti-crawler avoidance mechanism to avoid triggering third-party platform restrictions.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements all the steps of the local AI consumer risk identification method according to any one of claims 1 to 6; the computer-readable storage medium includes at least one of a removable storage medium and a terminal local storage unit, and supports user-independent data backup and deletion.

12. The computer-readable storage medium according to claim 11, characterized in that, The storage medium uses encrypted storage, allowing only local access from user terminals, and does not transmit any stored data to cloud servers or third-party platforms; the storage medium supports multi-terminal compatibility and can realize cross-terminal data synchronization (the synchronization process is completed locally and does not upload to the cloud).