Personalized e-commerce commodity display method based on AI recommendation algorithm
By combining multimodal data fusion and dynamic recommendation mechanisms with deep learning and collaborative filtering algorithms, the cold start problem of AI recommendation algorithms for new users and new products is solved, achieving diversity and privacy protection, and improving user experience and platform efficiency.
Patent Information
- Application Number
- CN202510753739.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-06
- Publication Date
- 2025-10-17
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing AI recommendation algorithms are ineffective in recommending new users and new products, suffer from cold start problems, offer limited recommendations, pose a high risk of privacy leaks, require significant computing resources, and struggle to respond to user changes in real time, thus impacting user experience and platform stickiness.
It employs multimodal data fusion, dynamic recommendation mechanisms, privacy protection technologies, and blockchain storage, combined with deep learning and collaborative filtering algorithms. It addresses cold start through social network analysis and natural language processing, introduces an information entropy weighting mechanism, and uses differential privacy and blockchain to ensure data security.
It effectively solves the cold start problem for new users and new products, improves the diversity and accuracy of recommendations, protects user privacy, enhances user experience and platform stickiness, and increases product exposure and sales revenue.
Abstract
Description
Technical Field
[0001] The present invention relates to the field of Internet e-commerce technology, and specifically to a personalized e-commerce product display method based on an AI recommendation algorithm. Background Art
[0002] Personalized e-commerce product display methods based on AI recommendation algorithms analyze users' historical behavior, interests, preferences, and purchasing habits, leveraging machine learning and data mining techniques to tailor product display content to each user. This approach typically combines collaborative filtering, content recommendation, and deep learning algorithms to extract features from multi-dimensional information such as users' browsing history, search behavior, and purchase data to predict products that users may be interested in. This approach helps e-commerce platforms increase product exposure, improve user conversion rates, and enhance customer satisfaction, while reducing information overload and optimizing the shopping experience.
[0003] Although the personalized e-commerce product display method based on AI recommendation algorithm can effectively improve user experience and platform sales, the current technology still has some shortcomings and challenges: for new users or new products, the recommendation system is difficult to generate accurate recommendations due to lack of sufficient data. This problem is called the cold start problem. New users do not have enough historical behavior data, and new products are not widely browsed and purchased, resulting in poor performance of the recommendation system in these situations; AI recommendation algorithm often relies on user's historical behavior and preference for recommendation, which is prone to "information cocoon" phenomenon, that is, the recommended products are highly concentrated in the categories that the user has ever purchased or browsed, resulting in that the user can only see the products similar to his interests, lacking diversity and freshness, and may miss some potential interest points or products; AI recommendation algorithm needs a large amount of personal data to make accurate recommendations, such as browsing records, purchase history, location, etc., which may cause users' concern about privacy leakage. The platform needs to balance the relationship between personalized recommendation and user privacy, and avoid data abuse or leakage; the recommendation system may produce certain bias due to unbalanced training data or design defects of the algorithm itself. For example, if a certain type of product has been clicked or purchased more, the system may over-push this type of product, resulting in recommendation bias towards popular products, ignoring niche products or brands; AI recommendation system mainly relies on users' past behavior to make predictions, which may not be able to capture users' changes and new interest points in real time. For example, some users may change their interests or purchase habits at a certain time, and traditional recommendation systems may not be able to reflect these changes in time, resulting in that the recommended results do not match the current needs of the users; modern AI recommendation algorithm (especially deep learning model) usually needs a large amount of computing resources and data processing capability, which may bring high cost to some e-commerce platforms. Especially in the case of large number of users and products, the computing and storage requirements of real-time recommendation system are more demanding; when the recommendation system is too accurate, it may lead to users' over-reliance on the platform's recommendation, reducing their motivation to explore independently. This may reduce users' stickiness to the platform, affect their long-term activity, and even lead to user loss.
[0004] Therefore, we propose a personalized e-commerce product display method based on AI recommendation algorithm. SUMMARY
[0005] To achieve the above purpose, the present application provides the following technical scheme: a personalized e-commerce product display method based on AI recommendation algorithm, comprising the following steps:
[0006] S1 data collection and preprocessing:
[0007] User behavior data is collected from user interaction behavior, including but not limited to browsing records, search keywords, click records, purchase history, etc.
[0008] Collect attribute data of commodities, including categories, prices, brands, descriptions, ratings, etc.
[0009] Standardize and clean all collected data, remove irrelevant or redundant data, fill in missing values, and ensure data quality.
[0010] S2 User Profile Construction:
[0011] Based on user behavior data, use deep learning models (such as convolutional neural networks CNN, long short-term memory networks LSTM, etc.) to generate high-dimensional user profiles.
[0012] User profiles include but are not limited to interest vectors, purchase preferences, price sensitivity, active time periods, and other multi-dimensional information, and are updated in real time through incremental learning models.
[0013] At the same time, combined with user social data (such as social media activities, likes, shares, comments, etc.) and location information, further refine the user profile and improve the accuracy of recommendations.
[0014] S3 Cold Start Problem Processing:
[0015] For new users, use a recommendation method based on user social networks to infer their potential interest points using friend or follow relationships in social networks.
[0016] For new commodities, use natural language processing techniques to analyze the text description, user reviews, and labels of the commodity, and combine the commodity's popular trends (such as search frequency, hot sales list, etc.) to make preliminary recommendations.
[0017] Further adjust the recommendation strategy through user first interaction feedback (such as click rate, browsing time, etc.).
[0018] S4 Dynamic Recommendation Mechanism:
[0019] Based on user real-time behavior data (such as recently viewed commodities, current search keywords, etc.), use multi-level collaborative filtering combined with deep learning models to generate real-time recommendations.
[0020] In the recommendation process, consider the time factor and optimize the time series-based recommendation mechanism to ensure that the recommended results meet the user's current interests and demand changes.
[0021] Through ensemble learning methods, integrate the output results of multiple recommendation algorithms and give a comprehensive weight to give the final recommendation.
[0022] S5 Recommendation Result Diversification:
[0023] A diversity metric is introduced to balance factors such as information entropy, product category diversity, and brand distribution to ensure that recommended content is not limited to popular products in users' historical behavior.
[0024] Using a diversity-weighted algorithm, the recommended diversity score is calculated based on product attributes (such as brand, price range, category, etc.) and introduced into the final recommendation ranking.
[0025] S6 privacy protection mechanism:
[0026] During the user data processing process, differential privacy algorithms are used to protect user privacy information, and random perturbation and noise injection techniques are used to ensure that data cannot be reversed.
[0027] Blockchain technology is used to store all data interaction records during the recommendation process to ensure that the data cannot be tampered with. Smart contracts are used to control data access rights to ensure that user data can only be used with user authorization.
[0028] Preferably, the user portrait is constructed by combining a deep learning model, using a convolutional neural network to process the user's image data and product attributes, and combining a recurrent neural network to model the user behavior sequence to generate a multi-dimensional, high-precision user interest vector.
[0029] Preferably, the cold start problem is handled by combining social media data mining and content-based recommendation algorithms, using social network analysis technology to determine user potential interests, and combining product similarity calculation to generate recommendations for new products.
[0030] Preferably, the diversity of recommendation results is achieved through an information entropy weighting mechanism, which calculates a comprehensive score and balances the diversity and personalized relevance of recommendations based on multiple factors such as the popularity, rating, category, and diversity of user historical behaviors of the products.
[0031] Preferably, the privacy protection mechanism records detailed logs of all user data processing and recommendation data exchanges through blockchain-based smart contracts, ensuring the transparency and immutability of data usage, while protecting user identity and behavior data through de-identification and differential privacy.
[0032] Compared with the existing technology, the present invention provides a personalized e-commerce product display method based on AI recommendation algorithm, which has the following beneficial effects:
[0033] 1、The personalized e-commerce product display method based on AI recommendation algorithm can effectively solve the cold start problem of new users and new products by multi-modal data fusion (social data, product description, trend data, etc.). New users do not need historical behavior data, and their potential interests can be accurately predicted through social relationship and social network analysis, ensuring the accuracy of initial recommendation.
[0034] 2、The personalized e-commerce product display method based on AI recommendation algorithm can avoid single recommendation and ensure the diversity of product recommendation by introducing an information entropy weighting mechanism, so that users will not always see products similar to their interests, but through balancing diversity and relevance, improving user discovery, increasing product exposure and sales opportunities.
[0035] 3、The personalized e-commerce product display method based on AI recommendation algorithm uses differential privacy technology and blockchain technology to protect user privacy, ensuring the accuracy of personalized recommendation and fully protecting user data security. In the scope of user authorization, access and use data, and can trace every step of operation in the recommendation process, improve user trust.
[0036] 4、The personalized e-commerce product display method based on AI recommendation algorithm dynamically adjusts the recommendation mechanism based on real-time user behavior, quickly responds to changes in user interest, improves the accuracy of recommendation results, avoids recommending outdated or irrelevant products, and enhances the user shopping experience.
[0037] 5、The personalized e-commerce product display method based on AI recommendation algorithm can help users easily find products of interest through personalized recommendation and diversified product display, enhancing the attractiveness and stickiness of the platform. At the same time, accurate recommendation can also improve purchase conversion rate and increase platform sales revenue. DETAILED DESCRIPTION
[0038] The technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0039] EMBODIMENT
[0040] Embodiment of personalized e-commerce product display method based on AI recommendation algorithm
[0041] The personalized e-commerce product display method based on AI recommendation algorithm comprises the following steps:
[0042] S1 data acquisition and preprocessing:
[0043] User behavior data is collected from user interactions, including but not limited to browsing history, search keywords, click records, purchase history, etc.
[0044] Attribute data of the goods is collected, including the category, price, brand, description, rating, etc.
[0045] All collected data is standardized and cleaned, irrelevant or redundant data is removed, missing values are filled, and data quality is ensured.
[0046] S2 User Profile Construction:
[0047] Based on user behavior data, deep learning models such as convolutional neural networks (CNN) and long short-term memory networks (LSTM) are used to generate high-dimensional user profiles.
[0048] User profiles include but are not limited to interest vectors, purchase preferences, price sensitivity, active time periods, and other multi-dimensional information, and are updated in real time through incremental learning models.
[0049] At the same time, combined with user social data (such as social media activity, likes, shares, comments, etc.) and location information, the user profile is further refined to improve the accuracy of recommendations.
[0050] S3 Cold Start Problem Processing:
[0051] For new users, a recommendation method based on user social networks is used to infer their potential interest points using friend or follow relationships in social networks.
[0052] For new goods, natural language processing technology is used to analyze the text description, user reviews, tags, etc. of the goods, combined with the hot trends of the goods (such as search frequency, hot sales list, etc.) to make preliminary recommendations.
[0053] Through user first interaction feedback (such as click rate, browsing time, etc.), the recommendation strategy is further adjusted.
[0054] S4 Dynamic Recommendation Mechanism:
[0055] Based on real-time behavior data of users (such as recently viewed goods, current search keywords, etc.), multi-level collaborative filtering and deep learning models are used to generate real-time recommendations.
[0056] In the recommendation process, time factors are considered to optimize time series-based recommendation mechanisms to ensure that the recommended results meet the current interests and demand changes of users.
[0057] Through ensemble learning methods, the output results of multiple recommendation algorithms are integrated to give the final recommendation with comprehensive weights.
[0058] S5 Recommendation result diversification:
[0059] Introduce diversity metrics, balance through information entropy, product category diversity, brand distribution, etc. to ensure that the recommended content is not limited to popular products in user historical behavior.
[0060] Use diversity weighting algorithm to calculate the diversity score of the recommended products according to product attributes (such as brand, price range, category, etc.), and introduce it into the final recommendation ranking.
[0061] S6 Privacy protection mechanism:
[0062] In the process of user data processing, use differential privacy algorithm to protect user privacy information, and through random disturbance and noise injection technology to ensure that data cannot be inversely calculated.
[0063] Use blockchain technology to store all data interaction records in the recommendation process to ensure data tamper-proof. And through the smart contract to control data access permissions, ensure that user data can only be used under user authorization.
[0064] Specifically, the construction of user portrait is through the joint deep learning model, using convolutional neural network to process user image data and product attributes, and combining recurrent neural network to model user behavior sequence, generating multi-dimensional, high-precision user interest vector.
[0065] Specifically, the processing of cold start problem is combined with social media data mining and content-based recommendation algorithm, using social network analysis technology to determine user's potential interest, and combining product similarity calculation to generate recommendations for new products.
[0066] Specifically, the diversity of the recommended results is realized through the information entropy weighting mechanism, which calculates a comprehensive score according to the heat, rating, category of the product, and the diversity of user historical behavior, and balances the diversity and individuality of the recommended results.
[0067] Specifically, the privacy protection mechanism records all user data processing, recommendation data exchange details through the smart contract based on blockchain, ensures the transparency and tamper-proof of data use, and protects user identity and behavior data through de-identification and differential privacy.
[0068] Through the technical solution, in the application, through multi-modal data fusion (social data, commodity description, trend data, etc.), the application can effectively solve the cold start problem of new users and new commodities. New users do not need historical behavior data, and their potential interests can be accurately predicted through social relationship and social network analysis, ensuring the accuracy of initial recommendation. By introducing an information entropy weighting mechanism, the application can avoid single recommendation and ensure the diversity of commodity recommendation, so that users will not always see commodities similar to their interests, but through the balance of diversity and relevance, the discoverability of users is improved, and the exposure rate and sales opportunities of commodities are increased. Using differential privacy technology and blockchain technology to protect user privacy ensures the accuracy of personalized recommendation and fully protects the data security of users. Within the scope of user authorization, access and use data, and be able to trace each step of operation in the recommendation process, improve user trust. The dynamic recommendation mechanism is adjusted based on the real-time behavior of users, which can quickly respond to changes in user interests, improve the accuracy of recommendation results, avoid recommending outdated or irrelevant commodities, and enhance the shopping experience of users. Through personalized recommendation and diversified commodity display, users can more easily find commodities of their interest, enhancing the attractiveness and stickiness of the platform. At the same time, accurate recommendation can also improve the purchase conversion rate and increase the sales revenue of the platform.
[0069] User portrait construction
[0070] Data collection:
[0071] The e-commerce platform collects user click data (browsing commodities, search keywords), transaction data (purchasing commodities, payment amount), and social data (user likes and sharing behavior on social media).
[0072] Deep learning modeling:
[0073] Convolutional neural networks are used to extract features of user commodity browsing behavior, and LSTM is used to model user historical behavior sequences (such as time sequences of each purchase, search, and browse) to generate user interest vectors.
[0074] User portrait includes multi-dimensional user interests, such as purchase categories (clothing, electronic products, food, etc.), brand preferences, and price sensitivity (users tend to high-priced or low-priced commodities).
[0075] Social data fusion:
[0076] For user social media behavior, natural language processing techniques (such as sentiment analysis and emotion analysis) are used to extract user interest tags, which are combined with user purchase history to further refine user portrait.
[0077] Social network analysis:
[0078] For new users, their potential interests are inferred by analyzing their friend or follow relationships on social media. For example, if a new user's friends like certain products or brands, the system can recommend these products to the new user.
[0079] Content-based recommendations:
[0080] For new products, the relevance of the product to the user's interests is evaluated using techniques such as TF-IDF (Term Frequency-Inverse Document Frequency) by analyzing the product's description, keywords, reviews, etc. New products are recommended to groups that match the user's interests.
[0081] Differential privacy:
[0082] When collecting and processing user behavior data, differential privacy techniques are applied to ensure that user data is not leaked or tracked by adding noise and random perturbations to the data.
[0083] Blockchain data storage:
[0084] All behavior related to recommendations (data access, recommendation generation, user feedback, etc.) is recorded through blockchain technology to ensure data transparency and tamper resistance. Each data access is managed through a smart contract to ensure that data processing and use are only within the scope of user authorization.
[0085] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.
Claims
1. A personalized e-commerce product display method based on an AI recommendation algorithm, characterized by: The following steps are involved: S1 data collection and preprocessing: Collect user behavior data from user interaction behaviors, including but not limited to browsing history, search keywords, click history, purchase history, etc. Collect product attribute data, including product category, price, brand, description, rating, etc. All collected data are standardized and cleaned, irrelevant or redundant data are removed, missing values are filled, and data quality is ensured. S2 user portrait construction: Based on user behavior data, high-dimensional user portraits are generated using deep learning models (such as convolutional neural networks (CNN), long short-term memory networks (LSTM), etc.). User portraits include but are not limited to multi-dimensional information such as interest vectors, purchasing preferences, price sensitivity, active time periods, etc., and are updated in real time through incremental learning models. At the same time, by combining users' social data (such as social media activities, likes, shares, comments, etc.) and location information, we can further refine user portraits and improve the accuracy of recommendations. S3 cold start problem handling: For new users, a recommendation method based on the user's social network is adopted, which uses the friends or follow-up relationships in the social network to infer their potential interests. For new products, we use natural language processing technology to analyze the product's text description, user reviews, tags, etc., and make preliminary recommendations based on the product's popular trends (such as search frequency, best-selling lists, etc.). Further adjust the recommendation strategy through user's first interaction feedback (such as click rate, browsing time, etc.). S4 dynamic recommendation mechanism: Based on users' real-time behavior data (such as recently browsed products, currently searched keywords, etc.), multi-level collaborative filtering is combined with deep learning models to generate real-time recommendations. During the recommendation process, time factors are taken into consideration and the time series-based recommendation mechanism is optimized to ensure that the recommendation results are in line with the user's current interests and demand changes. Through ensemble learning methods, the output results of multiple recommendation algorithms are integrated, and the final recommendation is given based on the comprehensive weights. S5 recommendation results are diverse: A diversity metric is introduced to balance factors such as information entropy, product category diversity, and brand distribution to ensure that recommended content is not limited to popular products in users' historical behavior. Using a diversity-weighted algorithm, the recommended diversity score is calculated based on product attributes (such as brand, price range, category, etc.) and introduced into the final recommendation ranking. S6 privacy protection mechanism: During the user data processing process, differential privacy algorithms are used to protect user privacy information, and random perturbation and noise injection techniques are used to ensure that data cannot be reversed. Blockchain technology is used to store all data interaction records during the recommendation process to ensure that the data cannot be tampered with. Smart contracts are used to control data access rights to ensure that user data can only be used with user authorization.
2. The personalized e-commerce product display method based on the AI recommendation algorithm according to claim 1, characterized in that: in, The construction of user portraits is achieved through a joint deep learning model, which uses convolutional neural networks to process user image data and product attributes, and combines recurrent neural networks to model user behavior sequences to generate multi-dimensional, high-precision user interest vectors.
3. The personalized e-commerce product display method based on the AI recommendation algorithm according to claim 1, characterized in that: in, The cold start problem is addressed by combining social media data mining with content-based recommendation algorithms, using social network analysis techniques to determine users' potential interests and combining product similarity calculations to generate recommendations for new products.
4. The personalized e-commerce product display method based on the AI recommendation algorithm according to claim 1, characterized in that: in, The diversity of recommendation results is achieved through an information entropy weighting mechanism, which calculates a comprehensive score and balances the diversity and personalized relevance of recommendations based on multiple factors such as the popularity, rating, category, and diversity of user historical behavior of the product.
5. The personalized e-commerce product display method based on the AI recommendation algorithm according to claim 1, characterized in that: in, The privacy protection mechanism uses blockchain-based smart contracts to record detailed logs of all user data processing and recommendation data exchanges, ensuring the transparency and immutability of data usage, while protecting user identity and behavioral data through de-identification and differential privacy.