Automatic retrieval and comparison system based on multi-platform e-commerce integration of online sales products

Abstract

The invention belongs to the technical field of e-commerce platforms, and discloses an automatic retrieval and comparison system based on multi-platform e-commerce integration of online sales products, which comprises a data acquisition module, a commodity information integration module, a retrieval module, a comparison and analysis module, a recommendation module, a user interaction module, a feedback module and a real-time updating module, and the data acquisition module is responsible for capturing related information of online sales products from the multi-e-commerce platform. According to the method, multi-platform e-commerce online sales product data islanding can be broken through API calling and web crawler technologies, meanwhile, commodity feature similarity calculation is performed through a TF-I DF + cosine similarity combined method, commodity text similarity can be quantified, multi-modal fusion introduction can further improve accuracy, repeated recommendation is avoided, and the method is suitable for large-scale popularization and application. And the comparative analysis module can evaluate the online sales products more comprehensively and objectively in combination with dimension data, so that the prejudice of a single index is avoided, and the problem that the comparison dimension of the online sales products is single is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of e-commerce platforms, and specifically relates to an automatic retrieval and comparison system based on multi-platform e-commerce integration of online sales products. Background Art

[0002] As the core carrier of digital commerce, e-commerce platforms have become a key hub connecting supply and demand by integrating multi-dimensional data and technical support. Their core functional architecture includes product and transaction infrastructure, user operation system and data-driven decision-making. They can be divided into B2C (business-to-consumer), B2B (business-to-business) and C2C (consumer-to-consumer) according to transaction entities; and can be divided into self-operated platforms and open platforms according to operating models.

[0003] With the development of science and technology and the progress of society, e-commerce platforms have penetrated into everyone's daily life. More and more people are shopping through e-commerce platforms. Therefore, an automatic retrieval and comparison system integrating multi-platform e-commerce has emerged. However, traditional automatic retrieval and comparison systems are mostly suitable for e-commerce platforms with public API interfaces. Some e-commerce platforms without public API interfaces still have data silo problems. At the same time, the online sales products recommended by traditional automatic retrieval and comparison systems are highly repetitive. In addition, due to the single comparison dimension, it is impossible to evaluate products more objectively and is prone to single indicator bias. Therefore, this needs to be improved. Summary of the Invention

[0004] The purpose of the present invention is to provide an automatic retrieval and comparison system based on the integration of multiple e-commerce platforms for online sales products to solve the problems raised in the above background technology.

[0005] In order to achieve the above-mentioned purpose, the present invention provides the following technical solution: an automatic retrieval and comparison system based on the integration of multi-platform e-commerce for online sales products, comprising: a data collection module, a product information integration module, a retrieval module, a comparison and analysis module, a recommendation module, a user interaction module, a feedback module, and a real-time update module, wherein:

[0006] The data collection module is responsible for capturing relevant information of online sales products from multiple e-commerce platforms and providing raw data support for subsequent processing;

[0007] The product information integration module integrates product information collected from different e-commerce platforms, eliminates data differences and duplications, and forms a unified and standardized product information database, providing a high-quality data foundation for subsequent retrieval, comparison, recommendation and other functions;

[0008] The search module is used for users to search the integrated product information according to keywords, categories, and price ranges to quickly find products that meet their needs;

[0009] The comparative analysis module performs multi-dimensional comparative analysis on the retrieved multiple products, helping users to fully understand the advantages and disadvantages of each product and make more informed purchasing decisions;

[0010] The recommendation module provides users with personalized product recommendations based on their historical behavior, preferences, and currently retrieved product information, thereby increasing the probability of users finding their desired products.

[0011] The user interaction module provides users with a friendly operation interface, making it convenient for users to enter search conditions, view search results, compare products and interact with recommendations;

[0012] The feedback module is used to collect user feedback on search results, comparative analysis, recommended products, and the overall experience of the system, in order to optimize the performance of the system and the effectiveness of recommendations;

[0013] The real-time update module is used to update product information and user behavior data in real time, so that the system can promptly reflect the latest developments of the e-commerce platform and changes in user needs.

[0014] As a preferred technical solution of the present invention, the data acquisition module includes API interface calls, web crawler technology, and data storage. The API interface calls send requests to the e-commerce platform server through the public API interfaces provided by major e-commerce platforms to obtain product data in JSON or XML format. During the call, the request rate is controlled by the token bucket algorithm or the leaky bucket algorithm. The web crawler technology simulates browser behavior, sends HTTP requests to the e-commerce platform website, obtains the HTML code of the web page, and then extracts the required product information by parsing the HTML. In actual use, dynamic User-Agent pools, IP proxy pools, and request interval randomization are used to avoid being blocked:

[0015] Delay=Base delay ×(1+Random(0,1))

[0016] In the formula, Base delay For the base delay, Random(0,1) is a random number between 0 and 1;

[0017] The data storage is divided into temporary storage and persistent storage. Temporary storage stores the collected data in a temporary database Redis for subsequent rapid processing and deduplication; persistent storage stores the data that has been preliminarily cleaned and deduplicated in a relational database MySQL or a non-relational database MongoDB to facilitate subsequent query and analysis.

[0018] As a preferred technical solution of the present invention, the product information integration module includes data cleaning, data standardization and data deduplication. The data cleaning is divided into removing noise data and processing missing values; the data standardization includes unit unification and format specification; the data deduplication includes similarity calculation based on product features and establishing a unique identifier.

[0019] As a preferred technical solution of the present invention, the product feature similarity calculation adopts a hybrid method of cosine similarity calculation combined with TF-IDF text vectorization, integrating the product title, description and image hash value features to improve the accuracy of duplicate product identification. The specific method is as follows:

[0020] The TF-IDF weight calculation formula is:

[0021] TF_IDF(t,d)=TF(t,d)×IDF(t)

[0022] Where, Used to measure the frequency of a word in a document. Used to measure the discriminative power of words;

[0023] The cosine similarity calculation formula is:

[0024]

[0025] At the same time, the product image features and text features are combined to perform multimodal fusion through the weighted fusion formula:

[0026] Final_Similarity=w1×Text_Sim+w2×Imag_Sim

[0027] Among them, w1 and w2 are weight coefficients, satisfying w1+w2=1, and are dynamically adjusted in real time based on user survey feedback.

[0028] As a preferred technical solution of the present invention, the retrieval module includes text retrieval and image retrieval. The text retrieval optimizes the query term weight distribution based on the BM25 algorithm, combines with MySQL full-text index to achieve millisecond-level response, and introduces Word2Vec word vectors to expand synonym retrieval capabilities; the image retrieval uses ResNet50 to extract the deep features of the product main image, combines with cosine similarity calculation to realize the "image search" function, and compresses the feature vector dimension to 256 dimensions to balance accuracy and efficiency.

[0029] As a preferred technical solution of the present invention, the comparative analysis module includes price trend analysis, comprehensive cost-performance rating and price trend prediction. The price trend analysis adopts the moving average method to predict the price trend of commodities:

[0030]

[0031] Where, MA n is the n-day moving average price, P t is the price on day t;

[0032] The comprehensive cost-effectiveness rating is based on multiple dimensions including price, sales volume, and ratings:

[0033]

[0034] In the formula, w1, w2, w3 are weights, obtained based on user preference surveys, and Price min The lowest price among similar products. Log(Sales+1) is used to suppress the weight of products with excessively high sales volume to prevent them from dominating the score.

[0035] The price trend prediction adopts LightGBM regression model:

[0036]

[0037] Where, is the damage function, used to measure the predicted price y i and the real price The square error ensures that the model fits the historical data accurately. It is a regularization term that avoids overfitting of the model by penalizing the absolute value of the coefficient (j is the feature weight).

[0038] As a preferred technical solution of the present invention, the recommendation module includes user behavior-based recommendations and content-based recommendations. The user behavior-based recommendations take into account the user's behavior sequence and predict the products that the user may be interested in next; the content-based recommendations construct a user interest feature vector based on the user's historical browsing, purchasing and other behaviors.

[0039] As a preferred technical solution of the present invention, the feedback module includes sentiment analysis and topic mining. The sentiment analysis adopts the BERT classification model and optimizes the model through cross entropy loss:

[0040]

[0041] Where N is the number of samples, C is the number of categories, and y i,c is the true label, p i,c is the predicted probability;

[0042] The topic mining uses the LDA model to extract topic distribution from the evaluation:

[0043]

[0044] Where z is the topic, d is the document, and w is the word.

[0045] As an optimal technical solution of the present invention, the real-time update module includes data timing update, real-time event monitoring and cache update strategy. The data timing update sets the update cycle of different types of data according to the data update frequency and system requirements of the e-commerce platform, adopts the incremental update method, and only obtains the data that has changed since the last update, reducing the data transmission volume and processing time; the real-time event monitoring uses the Kafka message queue to monitor the real-time events of the e-commerce platform, and at the same time uses WebSocket technology to establish a long connection between the client and the server to achieve real-time data synchronization; the cache update strategy is divided into cache invalidation mechanism and cache update notification. For the cache invalidation mechanism: when the cache expires, the system automatically obtains the latest data from the database and updates the cache; for the cache update notification: when the data source is updated, the cache system is notified to update through the publish-subscribe model.

[0046] The beneficial effects of the present invention are as follows:

[0047] Through API interface calls and web crawler technology, the present invention can break the data silos of online sales products on multiple e-commerce platforms. At the same time, it calculates the similarity of product features through a method combining TF-IDF+cosine similarity, which can quantify the similarity of product texts. The introduction of multimodal fusion can further improve accuracy and avoid duplicate recommendations. The comparative analysis module can combine dimensional data to evaluate online sales products more comprehensively and objectively, avoid the bias of a single indicator, and solve the problem of a single comparison dimension for online sales products. In addition, it can also predict the price trend of online sales products, making it convenient for users to choose the best price. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a system block diagram of the present invention. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0050] like Figure 1 As shown, the embodiment of the present invention provides an automatic retrieval and comparison system based on the integration of multiple e-commerce platforms for online sales products, including: a data collection module, a product information integration module, a retrieval module, a comparison and analysis module, a recommendation module, a user interaction module, a feedback module, and a real-time update module, wherein:

[0051] The data collection module is responsible for capturing relevant information about online sales products from multiple e-commerce platforms and providing raw data support for subsequent processing;

[0052] The product information integration module integrates product information collected from different e-commerce platforms, eliminating data discrepancies and duplications to form a unified and standardized product information database, providing a high-quality data foundation for subsequent search, comparison, recommendation and other functions;

[0053] The search module allows users to search the integrated product information based on keywords, categories, and price ranges to quickly find products that meet their needs;

[0054] The comparative analysis module conducts multi-dimensional comparative analysis on multiple retrieved products, helping users fully understand the advantages and disadvantages of each product and make more informed purchasing decisions;

[0055] The recommendation module provides users with personalized product recommendations based on their historical behavior, preferences, and currently retrieved product information, increasing the probability of users discovering their desired products.

[0056] The user interaction module provides users with a friendly operation interface, making it convenient for users to enter search conditions, view search results, compare products, and interact with recommendations;

[0057] The feedback module is used to collect user feedback on search results, comparative analysis, recommended products, and the overall system experience, in order to optimize system performance and recommendation results;

[0058] The real-time update module is used to update product information and user behavior data in real time, so that the system can promptly reflect the latest developments of the e-commerce platform and changes in user needs.

[0059] The data collection module and the product information integration module can obtain sufficient data sources by capturing and integrating online product information from multiple e-commerce platforms. Through the cooperation of the retrieval module and the user interaction module, users can check and search for corresponding products according to their own needs. The cooperation of the comparison and analysis module and the recommendation module can recommend suitable online products for users to choose. In addition, the cooperation of the feedback module and the real-time update module can ensure the timeliness and effectiveness of the system's information acquisition.

[0060] The data collection module includes API interface calls, web crawler technology, and data storage. API interface calls send requests to the e-commerce platform server through the public API interfaces provided by major e-commerce platforms to obtain product data in JSON or XML format. The request rate is controlled by the token bucket algorithm or leaky bucket algorithm during the call. The web crawler technology simulates browser behavior, sends HTTP requests to the e-commerce platform website, obtains the HTML code of the webpage, and then extracts the required product information by parsing the HTML. In actual use, dynamic User-Agent pools, IP proxy pools, and request interval randomization are used to avoid being banned:

[0061] Delay=Base delay ×(1+Random(0,1))

[0062] In the formula, Base delay For the base delay, Random(0,1) is a random number between 0 and 1;

[0063] Data storage is divided into temporary storage and persistent storage. Temporary storage stores the collected data in the temporary database Redis for subsequent rapid processing and deduplication; persistent storage stores the data that has undergone preliminary cleaning and deduplication in the relational database MySQL or the non-relational database MongoDB to facilitate subsequent query and analysis.

[0064] API interface calls have the advantages of stable and standardized data acquisition, and can also obtain more detailed product information, such as product title, price, inventory, sales volume, number of reviews, etc.; web crawler technology is used to obtain data from e-commerce platforms that do not provide API interfaces or have limited API interface functions. It is highly flexible and can obtain all information displayed on the web page without being restricted by the API interface.

[0065] Among them, the product information integration module includes data cleaning, data standardization and data deduplication. Data cleaning is divided into removing noise data and processing missing values; data standardization includes unit unification and format specification; data deduplication includes similarity calculation based on product characteristics and establishing unique identification.

[0066] Noise data is removed by deleting invalid characters, advertising information, repeated spaces, etc. from the collected data. Missing values ​​are estimated and filled based on the historical sales data of the product and the average inventory level of similar products. Data standardization uses a combination of Min-Max normalization and Z-Score normalization strategies to unify the scales of different dimensional features (such as price and sales volume). Unit unification is used to unify the units of product price, weight, size and other information on different e-commerce platforms. Format specification is used to unify the format of text information such as product titles and descriptions, such as removing redundant line breaks and tabs, and standardizing the use of punctuation marks.

[0067] The product feature similarity calculation uses a hybrid method combining cosine similarity calculation with TF-IDF text vectorization, integrating the product title, description, and image hash value features to improve the accuracy of duplicate product identification. The specific method is as follows:

[0068] The TF-IDF weight calculation formula is:

[0069] TF_IDF(t,d)=TF(t,d)×IDF(t)

[0070] Where, Used to measure the frequency of a word in a document. Used to measure the discriminative power of words;

[0071] The cosine similarity calculation formula is:

[0072]

[0073] At the same time, the product image features and text features are combined to perform multimodal fusion through the weighted fusion formula:

[0074] Final_Similarity=w1×Text_Sim+w2×Imag_Sim

[0075] Among them, w1 and w2 are weight coefficients, satisfying w1+w2=1, and are dynamically adjusted in real time based on user survey feedback.

[0076] Cosine similarity measures similarity by the angle between vectors. Its value range is [0, 1], and the closer it is to 1, the more similar the products are. TF-IDF+cosine similarity quantifies the similarity between product texts using statistical methods. It is efficient and interpretable, and the introduction of multimodal fusion can further improve the effect.

[0077] Among them, the retrieval module includes text retrieval and image retrieval. Text retrieval optimizes the query term weight distribution based on the BM25 algorithm, combines with MySQL full-text index to achieve millisecond-level response, and introduces Word2Vec word vector to expand synonym retrieval capabilities; image retrieval uses ResNet50 to extract the deep features of the product main image, combines with cosine similarity calculation to realize the "image search" function, and compresses the feature vector dimension to 256 dimensions to balance accuracy and efficiency.

[0078] The BM25 algorithm is developed based on a probabilistic retrieval model and is based on rigorous mathematical theory. By quantifying the relevance between documents and queries, it effectively improves the accuracy of retrieval results and the quality of relevance ranking. Users can adjust parameters based on specific dataset characteristics (such as document length distribution and query term distribution) to optimize retrieval results.

[0079] The comparative analysis module includes price trend analysis, comprehensive cost-performance rating, and price trend forecasting. The price trend analysis uses the moving average method to predict commodity price trends:

[0080]

[0081] Where, MA n is the n-day moving average price, P t is the price on day t;

[0082] The comprehensive cost-effectiveness rating is based on multiple dimensions including price, sales volume, and ratings:

[0083]

[0084] In the formula, w1, w2, w3 are weights, obtained based on user preference surveys, and Price min The lowest price among similar products. Log(Sales+1) is used to suppress the weight of products with excessively high sales volume to prevent them from dominating the score.

[0085] Price trend prediction uses LightGBM regression model:

[0086]

[0087] Where, is the damage function, used to measure the predicted price y i and the real price The square error ensures that the model fits the historical data accurately. It is a regularization term that avoids overfitting of the model by penalizing the absolute value of the coefficient (j is the feature weight).

[0088] The moving average method is a classic time series analysis tool that smooths short-term fluctuations by calculating the average of price data over consecutive time periods, thereby revealing long-term trends. Its core advantages lie in its simplicity, efficiency, strong anti-interference ability, and adaptability. The comprehensive cost-effectiveness score integrates multi-dimensional data such as price, sales volume, and user ratings to more comprehensively and objectively evaluate the value of a product, avoiding the bias of a single indicator. Price trend forecasts can predict future price changes.

[0089] Among them, the recommendation module includes user behavior-based recommendations and content-based recommendations. User behavior-based recommendations consider the user's behavior sequence and predict the products that the user may be interested in next; content-based recommendations construct user interest feature vectors based on the user's historical browsing, purchasing and other behaviors.

[0090] First, content-based recommendations are used to provide users with a preliminary list of product recommendations, and then collaborative filtering recommendations are used to optimize and supplement the list. By organically combining user behavior-based recommendations with content-based recommendations, the advantages of each method can be fully utilized to improve the accuracy and diversity of recommendations.

[0091] The feedback module includes sentiment analysis and topic mining. Sentiment analysis uses the BERT classification model, which is optimized through cross-entropy loss:

[0092]

[0093] Where N is the number of samples, C is the number of categories, and y i,c is the true label, p i,c is the predicted probability;

[0094] Topic mining uses the LDA model to extract topic distribution from reviews:

[0095]

[0096] Where z is the topic, d is the document, and w is the word.

[0097] Sentiment analysis analyzes the sentiment tendency of user evaluation and feedback texts to determine whether the user's evaluation is positive, negative or neutral; topic mining uses the LDA model to cluster user feedback topics and extract key issues and improvement suggestions that users are concerned about.

[0098] Among them, the real-time update module includes data scheduled update, real-time event monitoring and cache update strategy. Data scheduled update sets the update cycle of different types of data according to the data update frequency and system requirements of the e-commerce platform, adopts incremental update method, and only obtains data that has changed since the last update, reducing data transmission volume and processing time; real-time event monitoring uses Kafka message queue to monitor real-time events of the e-commerce platform, and uses WebSocket technology to establish a long connection between the client and the server to achieve real-time data synchronization; cache update strategy is divided into cache invalidation mechanism and cache update notification. For cache invalidation mechanism: when the cache expires, the system automatically obtains the latest data from the database and updates the cache; for cache update notification: when the data source is updated, the cache system is notified to update through the publish-subscribe model.

[0099] By timely updating the real-time update module, the timeliness and validity of the data can be ensured, thus helping users obtain the latest product information.

[0100] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0101] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An automatic search and comparison system based on the integration of multiple e-commerce platforms for online sales products, characterized by: It includes: data collection module, product information integration module, retrieval module, comparison and analysis module, recommendation module, user interaction module, feedback module, and real-time update module. The data collection module is responsible for capturing relevant information of online sales products from multiple e-commerce platforms and providing raw data support for subsequent processing; The product information integration module integrates product information collected from different e-commerce platforms, eliminates data differences and duplications, and forms a unified and standardized product information database, providing a high-quality data foundation for subsequent retrieval, comparison, recommendation and other functions; The search module is used for users to search the integrated product information according to keywords, categories, and price ranges to quickly find products that meet their needs; The comparative analysis module performs multi-dimensional comparative analysis on the retrieved multiple products, helping users to fully understand the advantages and disadvantages of each product and make more informed purchasing decisions; The recommendation module provides users with personalized product recommendations based on their historical behavior, preferences, and currently retrieved product information, thereby increasing the probability of users finding their desired products. The user interaction module provides users with a friendly operation interface, making it convenient for users to enter search conditions, view search results, compare products and interact with recommendations; The feedback module is used to collect user feedback on search results, comparative analysis, recommended products, and the overall experience of the system, in order to optimize the performance of the system and the effectiveness of recommendations; The real-time update module is used to update product information and user behavior data in real time, so that the system can promptly reflect the latest developments of the e-commerce platform and changes in user needs.

2. The automatic search and comparison system based on multi-platform e-commerce integration of online sales products according to claim 1 is characterized by: The data collection module includes API interface calls, web crawler technology, and data storage. The API interface calls send requests to the e-commerce platform server through the public API interfaces provided by major e-commerce platforms to obtain product data in JSON or XML format. The request rate is controlled by the token bucket algorithm or leaky bucket algorithm during the call. The web crawler technology simulates browser behavior, sends HTTP requests to the e-commerce platform website, obtains the HTML code of the web page, and then extracts the required product information by parsing the HTML. In actual use, dynamic User-Agent pools, IP proxy pools, and request interval randomization are used to avoid being blocked: Delay=Base delay ×(1+Random(0,1)) In the formula, Base delay For the base delay, Random(0,1) is a random number between 0 and 1; The data storage is divided into temporary storage and persistent storage. Temporary storage stores the collected data in a temporary database Redis for subsequent rapid processing and deduplication; persistent storage stores the data that has been preliminarily cleaned and deduplicated in a relational database MySQL or a non-relational database MongoDB to facilitate subsequent query and analysis.

3. The automatic search and comparison system based on multi-platform e-commerce integration of online sales products according to claim 1 is characterized by: The product information integration module includes data cleaning, data standardization and data deduplication. The data cleaning is divided into removing noise data and processing missing values; the data standardization includes unit unification and format specification; the data deduplication includes similarity calculation based on product features and establishing a unique identifier.

4. The automatic search and comparison system based on multi-platform e-commerce integration of online sales products according to claim 3 is characterized by: The product feature similarity calculation uses a hybrid method of cosine similarity calculation combined with TF-IDF text vectorization, integrating the product title, description, and image hash value features to improve the accuracy of duplicate product identification. The specific method is as follows: The TF-IDF weight calculation formula is: TF_IDF(t,d)=TF(t,d)×IDF(t) Where, Used to measure the frequency of a word in a document. Used to measure the discriminative power of words; The cosine similarity calculation formula is: At the same time, the product image features and text features are combined to perform multimodal fusion through the weighted fusion formula: Final_Similarity=w1×Text_Sim+w2×Imag_Sim Among them, w1 and w2 are weight coefficients, satisfying w1+w2=1, and are dynamically adjusted in real time based on user survey feedback.

5. The automatic search and comparison system based on multi-platform e-commerce integration of online sales products according to claim 1 is characterized by: The retrieval module includes text retrieval and image retrieval. The text retrieval optimizes the query word weight distribution based on the BM25 algorithm, combines with MySQL full-text index to achieve millisecond-level response, and introduces Word2Vec word vectors to expand synonym retrieval capabilities; the image retrieval uses ResNet50 to extract the deep features of the product main image, combines with cosine similarity calculation to realize the "image search" function, and compresses the feature vector dimension to 256 dimensions to balance accuracy and efficiency.

6. The automatic search and comparison system based on multi-platform e-commerce integration of online sales products according to claim 1, characterized in that: The comparative analysis module includes price trend analysis, comprehensive cost-performance rating, and price trend forecasting. The price trend analysis uses the moving average method to predict commodity price trends: Where, MA n is the n-day moving average price, P t is the price on day t; The comprehensive cost-effectiveness rating is based on multiple dimensions including price, sales volume, and ratings: In the formula, w1, w2, w3 are weights, obtained based on user preference surveys, and Price min The lowest price among similar products. Log(Sales+1) is used to suppress the weight of products with excessively high sales volume to prevent them from dominating the score. The price trend prediction adopts LightGBM regression model: Where, is the damage function, used to measure the predicted price y i and the real price The square error ensures that the model fits the historical data accurately. It is a regularization term that avoids overfitting of the model by penalizing the absolute value of the coefficient (j is the feature weight).

7. The automatic search and comparison system based on multi-platform e-commerce integration of online sales products according to claim 1 is characterized by: The recommendation module includes user behavior-based recommendations and content-based recommendations. The user behavior-based recommendations consider the user's behavior sequence and predict the products that the user may be interested in next; the content-based recommendations construct a user interest feature vector based on the user's historical browsing, purchasing and other behaviors.

8. The automatic search and comparison system based on multi-platform e-commerce integration of online sales products according to claim 1 is characterized by: The feedback module includes sentiment analysis and topic mining. The sentiment analysis uses the BERT classification model and the model is optimized through cross entropy loss: Where N is the number of samples, C is the number of categories, and y i,c is the true label, p i,c is the predicted probability; The topic mining uses the LDA model to extract topic distribution from the evaluation: Where z is the topic, d is the document, and w is the word.

9. The automatic search and comparison system based on multi-platform e-commerce integration of online sales products according to claim 1 is characterized by: The real-time update module includes data scheduled update, real-time event monitoring and cache update strategy. The data scheduled update sets the update cycle of different types of data according to the data update frequency and system requirements of the e-commerce platform, adopts the incremental update method, and only obtains the data that has changed since the last update, reducing the data transmission volume and processing time; the real-time event monitoring uses the Kafka message queue to monitor the real-time events of the e-commerce platform, and adopts WebSocket technology to establish a long connection between the client and the server to achieve real-time data synchronization; the cache update strategy is divided into cache invalidation mechanism and cache update notification. For the cache invalidation mechanism: when the cache expires, the system automatically obtains the latest data from the database and updates the cache; for the cache update notification: when the data source is updated, the cache system is notified to update through the publish-subscribe model.

Images