Commodity management system of network marketing platform

Abstract

The invention relates to the technical field of marketing platform management, and discloses a commodity management system of a network marketing platform, which comprises a multi-source information fusion module, a dynamic attribute tag module, an association recommendation module, an inventory and market linkage module and a feedback optimization module. Through cooperation of multi-source information fusion and a dynamic attribute label system, full-dimension integration and real-time updating of commodity information are realized, limitation of single environmental protection attribute or user behavior analysis is broken through, commodity cross-sales efficiency is improved through scene-based association recommendation, and commodity cross-sales efficiency is improved. Static attribute presentation of commodity management is upgraded to dynamic market response, and the control ability and marketing accuracy of the platform for the whole view of the commodity market are remarkably enhanced.

Description

Technical Field

[0001] This invention relates to the field of marketing platform management technology, and more specifically discloses a product management system for an online marketing platform. Background Technology

[0002] Online marketing platforms are tools that use the internet as a foundation to systematically promote products or services through websites and other media. In a broad sense, they encompass forms such as search engine promotion and social media marketing, while in a narrow sense, they refer to online business activities.

[0003] The prior art patent document with authorization announcement number CN120013563A discloses "a computer Internet marketing management system and method", which relates to the field of Internet marketing management technology. This application includes an information acquisition module, an information analysis module, a marketing plan formulation module and a marketing plan optimization module.

[0004] The patent document with authorization announcement number CN119228417A discloses "a marketing platform management system for internet promotion", which includes a product information analysis module, a customer information analysis module, a promotion information analysis module, and a sales information analysis module. By collecting product information of target companies, it obtains the environmental information of each product. Then, by obtaining the environmental information of each product of other companies, it obtains each threatening product of the target company. Based on the environmental browsing information of each promotion platform, it obtains the green platform. Based on the advertising volume, effective customer volume, and potential customer volume of each threatening product of the green platform, it obtains each preset promotion platform.

[0005] While existing technologies have laid a certain foundation in terms of marketing precision and user experience optimization—for example, by focusing on the environmental attributes of products to improve the efficiency of green promotion and by realizing personalized recommendations based on user behavior data—and by adjusting strategies based on data feedback to reduce costs and ensure the flexibility and completeness of marketing plans, existing technologies rely too heavily on a single environmental dimension, neglecting other key product attributes and full lifecycle management. They also focus on user-side analysis, with weak support for dynamic product management (such as inventory and category updates) and lack integration of real-time market trend data, resulting in fragmented functions and making it difficult to form a closed-loop management capability for products from information integration to market response. Summary of the Invention

[0006] The main technical problem solved by this invention is to provide a product management system for an online marketing platform, which can solve the problems mentioned in the background art above.

[0007] To solve the above-mentioned technical problems, according to one aspect of the present invention, more specifically, a product management system for an online marketing platform, comprising: The multi-source information fusion module integrates basic product information, cross-platform data, and real-time market data to construct a dynamic product information database. The dynamic attribute tagging module uses NLP technology to analyze user reviews and social discussion content, generating and updating dynamic product tags in real time. The association recommendation module combines dynamic product tags with user behavior data to construct a three-dimensional association model of "product-scenario-user". The inventory and market linkage module uses the LSTM neural network algorithm for sales forecasting and, combined with supply chain data, automatically triggers replenishment and promotion strategies when inventory is abnormal or market popularity changes. The feedback optimization module monitors user behavior and negative reviews on the product display page, automatically adjusts the focus of product descriptions, and synchronizes optimization actions with supply chain quality inspection.

[0008] Furthermore, the multi-source information fusion module includes: a data acquisition module, a preprocessing module, and a dynamic storage module; Data acquisition module: It uses distributed web crawlers to collect basic product information from e-commerce platforms, product mentions on social media, and user discussions. At the same time, it connects to third-party evaluation platforms and supply chain systems through API interfaces to obtain product quality scores, inventory data, and production cycle information. Preprocessing module: Cleans and standardizes the collected data; Dynamic storage module: It uses a time-series database to store processed data, maintains a dynamic product information database with an hourly incremental update mechanism, and supports multi-dimensional retrieval.

[0009] Furthermore, the dynamic attribute label module includes: an information parsing module, a label generation and weight calculation module, and a label update module; Information parsing module: Based on NLP technology, it parses user reviews and social discussion texts, and extracts keywords describing product features through word segmentation and semantic recognition; Tag generation and weight calculation module: Combines the sentiment analysis results of the BERT model to generate dynamic tags for products, and calculates tag weights based on keyword frequency and sentiment intensity; Tag update module: The tag library is updated based on new data every 24 hours, and low-weight tags that have not been mentioned for 7 consecutive days are automatically removed.

[0010] Furthermore, the association recommendation module includes: an association mining module, an association model construction module, and a recommendation module; Association Mining Module: Utilizes the Apriori algorithm to analyze sales data from the past 30 days and mine association rules for scenario-based product combinations; Association Model Construction Module: Based on the mined association rules, a three-dimensional association model of "product-scenario-user" is constructed, and product co-occurrence frequency, scenario matching degree and user preference tags are entered; Recommendation module: Receives user behavior trigger signals, calls 3D correlation model data to generate a list of related product recommendations, filters out products with insufficient stock or those already purchased by the user, and outputs the final recommendation result.

[0011] Furthermore, the inventory and market linkage module includes: a sales forecasting module, an early warning and replenishment module, and a market demand response module; Sales forecasting module: Based on LSTM neural network, it takes historical sales data, holiday factors and competitor price fluctuation data as input to generate a 7-day short-term sales forecast curve; Early warning and replenishment module: Compares predicted sales with current inventory. When inventory is lower than the predicted value, a replenishment reminder is triggered and pushed to the supply chain system, and the procurement priority is updated synchronously. Market popularity response module: Real-time monitoring of the trending search rankings of product-related keywords. When the popularity reaches a preset threshold, it automatically activates homepage banner display and limited-time promotional strategies such as issuing discount coupons.

[0012] Furthermore, the feedback optimization module includes: a user behavior monitoring module, a negative review keyword extraction module, and a strategy adjustment module; User behavior monitoring module: Through page tracking technology, it collects user behavior data in real time, such as dwell time, click location, and bounce rate on product display pages, and generates a user attention heatmap; Negative review keyword extraction module: This module segments and counts the frequency of negative product reviews, extracts high-frequency negative keywords, and tags the corresponding product issues. Strategy Adjustment Module: Optimizes the display logic of product pages based on user behavior data, and updates the focus of product descriptions based on negative review keywords.

[0013] Furthermore, the data cleaning involves removing redundant data based on a rule engine; the standardization process involves adopting standardized field formats for commodity data.

[0014] Furthermore, the page tracking technology involves embedding lightweight monitoring code in key interactive areas of the product display page, and collecting timestamps, click coordinates, and page scroll depth data of user behavior through asynchronous data transmission.

[0015] The beneficial effects of the product management system for an online marketing platform of the present invention are as follows: By integrating multi-source information and collaborating with a dynamic attribute tagging system, the platform achieves comprehensive integration and real-time updates of product information. This not only breaks through the limitations of analyzing single environmental attributes or user behavior, but also improves the efficiency of cross-selling products through scenario-based related recommendations. It upgrades product management from static attribute presentation to dynamic market response, significantly enhancing the platform's ability to control the overall product market and improve marketing accuracy.

[0016] Through the closed-loop design of the inventory and market linkage module and the feedback optimization module, the system can automatically adjust the supply chain strategy based on sales forecasts and real-time market trends. At the same time, it can dynamically optimize the product display logic by combining user behavior and negative reviews. This not only ensures the accurate matching of product supply and market demand, but also improves product competitiveness through continuous iteration, ultimately achieving the dual goals of cost reduction and efficiency improvement for the online marketing platform and enhanced user satisfaction. Attached Figure Description

[0017] The present invention will now be described in further detail with reference to the accompanying drawings and specific implementation methods.

[0018] Figure 1 This is a schematic diagram of the system principle. Detailed Implementation

[0019] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.

[0020] According to one aspect of the invention, such as Figure 1 As shown, a product management system for an online marketing platform is provided, including: The multi-source information fusion module integrates basic product information (such as specifications and origin), cross-platform data (such as social media mentions and third-party reviews), and real-time market data (such as competitor price fluctuations and trending topics) to build a dynamic product information database. This provides comprehensive and dynamic data support for product management, enabling the platform to accurately grasp the overall picture of the product market. This module includes: Data acquisition module: It uses distributed crawlers to collect basic product information (such as specifications and place of origin) from e-commerce platforms, product mentions on social media and user discussions. At the same time, it connects to third-party evaluation platforms and supply chain systems through API interfaces to obtain product quality scores, inventory data and production cycle information. Among them, the distributed crawler uses the Scrapy-Redis framework to achieve multi-node collaborative crawling, avoids anti-crawling mechanisms through IP proxy pool, and enables Selenium to simulate browser operations for JavaScript dynamically rendered pages. The API interface uses the OAuth2.0 authentication mechanism and develops adaptable parsers for different platform data formats to ensure consistent conversion of data in formats such as JSON and XML. In addition, the supply chain system is integrated with the ERP system in real time using a RESTful API, focusing on collecting information such as current inventory, production schedule, and logistics transportation nodes. Data transmission is secured using the HTTPS encryption protocol.

[0021] Preprocessing module: Cleans and standardizes the collected data; Among them, data cleaning is the removal of redundant data based on the rule engine. Specifically, the Drools rule engine is used to build a cleaning rule library. Redundancy is determined by calculating the cosine similarity of text for duplicate product information. Regular expressions are used to filter user comments containing invalid characters (such as consecutive special symbols or garbled characters). Data missing key fields (such as price and specifications) are marked as pending verification and a manual review process is triggered. The standardization process involves adopting a unified field format based on commodity data specifications. Specifically, a field mapping table is established based on commodity data standards, and the "specification" field is uniformly converted into the "length × width × height (cm)" format. The "place of origin" field is matched to the municipal administrative unit through the Gaode Map API.

[0022] Dynamic storage module: It uses a time-series database to store the processed data, maintains a dynamic product information database with an hourly incremental update mechanism, and supports multi-dimensional retrieval; The time series database uses InfluxDB, with a primary key index built by "product ID-timestamp". Basic product information, real-time sales and public opinion data are stored separately, and each data record retains the most recent 30-day historical version for trend analysis. In addition, multi-dimensional retrieval is implemented through Elasticsearch, which builds inverted indexes for fields such as product ID, category, tag, and timestamp, supports fuzzy queries (such as "summer + short sleeves") and range queries (such as "price > 100 yuan"), and the retrieval response time is controlled within 100ms.

[0023] This setup enables multi-dimensional real-time integration and efficient retrieval of product information, providing comprehensive and dynamic data support for the platform to accurately grasp the overall picture of the product market.

[0024] The dynamic attribute tagging module uses NLP technology to analyze user reviews and social discussions, generating and updating dynamic product tags in real time. This ensures product attributes align with real-time market feedback, enhancing product marketing visibility. This module includes: Information parsing module: Based on NLP technology, it parses user reviews and social discussion texts, and extracts keywords describing product features through word segmentation and semantic recognition; Among them, the word segmentation uses the Jieba word segmentation tool combined with a custom product domain dictionary to improve the accuracy of word segmentation. For example, "waterproof fabric" is accurately segmented into "waterproof" and "fabric". Part-of-speech tagging uses the NLTK library's part-of-speech tagger to mark nouns, verbs, adjectives, and other parts of speech to assist in subsequent attribute extraction. Finally, syntactic analysis uses the Stanford CoreNLP tool to construct a dependency syntax tree, identify the grammatical relationships between sentence components, and accurately locate the words that describe product attributes.

[0025] Tag generation and weight calculation module: Combines the sentiment analysis results of the BERT model to generate dynamic tags for products, and calculates tag weights based on keyword frequency and sentiment intensity; Among them, the BERT model uses a pre-trained Chinese model, which is fine-tuned on the product review dataset to improve the accuracy of sentiment analysis and output the sentiment polarity (positive, negative, neutral) and sentiment intensity value of the reviews; At the same time, when generating tags, the keywords after text parsing are filtered. For example, tags containing positive emotional words such as "high quality" and "durable" are generated as "high quality" tags, while tags containing negative words such as "fault" and "defect" are generated as "quality potential" tags. Finally, the weight calculation uses the TF-IDF algorithm combined with sentiment intensity weighting. TF-IDF measures the importance of the label in the text, and sentiment intensity adjusts the weight.

[0026] Tag update module: The tag library is updated based on new data every 24 hours, and low-weight tags that have not been mentioned for 7 consecutive days are automatically removed; Among them, new data acquisition is achieved by using distributed crawlers to capture the latest user reviews and social discussions on a daily schedule, and incrementally updating the text dataset; When iterating through the tag library, the TF-IDF values ​​and sentiment intensity weights of all tags are recalculated and sorted in descending order of weight. In addition, tags that have not appeared in new data for 7 consecutive days and whose weight is below a set threshold (e.g., 0.05) are automatically deleted from the tag library to ensure that the tag library reflects the latest market evaluation of products in real time.

[0027] The three components work together to enable the dynamic attribute tag module to achieve real-time dynamic updates and accurate presentation of product attributes. This allows it to closely follow market feedback, ensuring that product tags always reflect the latest user reviews and social discussion hotspots. This effectively enhances the product's recognizability and attractiveness during the marketing process, providing strong support for the platform's precision marketing.

[0028] The related recommendation module combines dynamic product tags with user behavior data to construct a three-dimensional "product-scenario-user" association model, enabling contextualized product recommendations and improving cross-selling efficiency to optimize the user shopping experience. This module includes: Association Mining Module: Utilizes the Apriori algorithm to analyze sales data from the past 30 days and mine association rules for scenario-based product combinations; Specifically, during the analysis, order data is grouped by product ID, and minimum support (e.g., minimum support is set to 5%, meaning that the combination appears more than 5% of the total number of orders) and minimum confidence (e.g., minimum confidence is set to 60%, meaning that the probability of purchasing B after purchasing A is more than 60%) are set. Frequent itemsets are generated through layer-by-layer iteration, and finally, scenario-based association rules such as "camping tent + sleeping bag + outdoor power supply" and "hot pot base + beef rolls" are extracted. Each rule is then labeled with a corresponding scenario tag (e.g., "camping" or "family dinner").

[0029] Association Model Construction Module: Based on the mined association rules, a three-dimensional association model of "product-scenario-user" is constructed, and product co-occurrence frequency, scenario matching degree and user preference tags are entered; The model is stored in a graph database (such as Neo4j), with product nodes associated with their dynamic tags, scene nodes associated with typical product combinations, and user nodes associated with historically purchased products and scene preferences. In addition, the co-occurrence frequency of products is represented by the weight value of the edge (for example, if A and B co-occur 1000 times, the weight is 1000), the scene matching degree is calculated by the cosine similarity between product tags and scene keywords (range 0-1), and user preference tags are extracted from user behavior data (for example, if you frequently buy baby products, you are labeled with the "parenting" tag).

[0030] Recommendation module: Receives user behavior trigger signals, calls 3D association model data to generate a list of related product recommendations, filters out products with insufficient stock or those already purchased by the user, and outputs the final recommendation result; Among them, the trigger signals include user clicks, adding to cart, or searching for a product. The system first locates the node of the product in the model, and then uses the shortest path algorithm (such as Dijkstra) to find the product with the highest relevance. Meanwhile, the filtering process is integrated with real-time data from the inventory management system (e.g., removing items with less than 5 units in stock and excluding items purchased by the user within the last 90 days). The final recommendation list is sorted in descending order of relevance score. For example, the top 3 items show scenario-based combinations (such as "You may also need to pair it with an outdoor power supply, suitable for camping scenarios").

[0031] This achieves the effect of precise recommendations based on specific scenarios, which not only increases the cross-selling rate of products, but also enhances the convenience and immersive experience of users' shopping.

[0032] The inventory and market linkage module uses an LSTM neural network algorithm for sales forecasting. Combined with supply chain data, it automatically triggers replenishment and promotional strategies when inventory levels are abnormal or market sentiment changes, constructing a closed loop of "sales-inventory-supply chain" to ensure that product supply matches market demand. This module includes: Sales forecasting module: Based on LSTM neural network, it takes historical sales data, holiday factors and competitor price fluctuation data as input to generate a 7-day short-term sales forecast curve; The LSTM neural network uses a 3-layer stacked structure (such as input layer + 2 hidden layers + output layer), with each hidden layer containing 64 neurons, and is trained using the Adam optimizer and MSE loss function. Historical sales data is taken from the daily sales of the past 180 days, and time series samples are generated by using a sliding window (window size is 7 days); Finally, the holiday factor is encoded using one-hot encoding (e.g., Spring Festival is marked as 1, and ordinary days are marked as 0). Competitor price fluctuation data is crawled hourly by a web crawler, and the difference rate between the daily average price and the average price of the previous 3 days is calculated as one of the input features.

[0033] Early warning and replenishment module: Compares predicted sales with current inventory. When inventory is lower than the predicted value, a replenishment reminder is triggered and pushed to the supply chain system, and the procurement priority is updated synchronously. The replenishment reminder is pushed to the supply chain ERP system via JSON format API. It includes product ID, suggested replenishment quantity (forecasted sales volume, current inventory), urgency level (graded according to the proportion of inventory gap to forecasted sales volume: for example, >50% is "urgent", 30%-50% is "priority", and <30% is "normal"). At the same time, the procurement priority is calculated based on the product profit margin and the impact of stockouts to generate a procurement ranking table.

[0034] Market popularity response module: Real-time monitoring of the trending search rankings of product-related keywords. When the popularity reaches a preset threshold, it automatically activates homepage banner display and limited-time promotional strategies such as issuing discount coupons. Among them, the hot search monitoring covers three major platforms: Weibo hot search, Douyin hot list, and Baidu index. Data is pulled every 15 minutes through API interface to extract the ranking and popularity value of product-related keywords (such as "outdoor tent" related to "camping equipment" and "camping artifacts").

[0035] This setup automates the entire process of sales forecasting, inventory alerts, and market response, avoiding the risks of stockouts or excess inventory while quickly seizing market opportunities and forming a dynamic closed loop of "forecasting-adjustment-feedback".

[0036] The feedback optimization module monitors user behavior and negative reviews on product display pages, automatically adjusts product descriptions, and synchronizes supply chain quality control optimizations to enhance product market competitiveness and user satisfaction, helping online marketing platforms reduce costs, increase efficiency, and achieve precise operations. This module includes: User behavior monitoring module: Through page tracking technology, it collects user behavior data in real time, such as dwell time, click location, and bounce rate on product display pages, and generates a user attention heatmap; Among them, the page buried point technology embeds lightweight monitoring codes in the key interaction areas of the product display page, and collects the timestamps, click coordinates, and page scroll depth data of user behaviors through asynchronous data transmission. The specific implementation method is to adopt a non-invasive buried point solution, add custom data attributes (such as data-track="price") in HTML tags to mark key areas such as price and specifications, capture user behaviors through JavaScript event listening (such as click, scroll, mouseleave), transmit the data to the backend in real time via WebSocket, and use a heat map generation tool (such as Hotjar) to visually display high-concern areas (for example, red indicates staying for more than 10 seconds, yellow indicates 5-10 seconds), and the bounce rate is determined by, for example, "no interaction within 10 seconds after entering the page".

[0037] Negative review keyword extraction module: Segment and count the frequencies of the product negative review texts, extract high-frequency negative keywords, and mark the corresponding product problem tags; Among them, for word segmentation, Jieba segmentation is combined with a stop word list (filtering meaningless words such as "de", "le", etc.), and then the frequency of keyword occurrences is counted using the collections library in Python. The proportion is calculated as "frequency / total number of negative reviews", and the words with a proportion exceeding a certain amount (for example, 15%) are extracted as high-frequency negative keywords (such as "slow logistics", "size does not match"); In addition, the problem tags adopt a hierarchical classification method, map the keywords to preset categories (for example, "slow logistics" → "delivery problem", "size does not match" → "specification error"), and associate them with the product ID for storage.

[0038] Strategy adjustment module: Optimize the product page display logic according to user behavior data, and update the key points of the product description in combination with negative review keywords; Among them, the optimization of the page display logic is based on heat map data,置顶 the area where users stay the longest (such as "material description"), and shorten the default display length of the high-bounce-rate area (such as a long brand story); For the update of the product description, the template replacement method is adopted, automatically supplement the description for high-frequency negative keywords (for example, "slow logistics" → add "default SF Express, 3-5 days to reach remote areas"), and synchronously generate quality inspection work orders and push them to the supply chain (for example, "size does not match" → trigger a re-inspection of product specifications, and the work order includes the keyword occurrence frequency and the associated order ID).

[0039] The synergy of the three realizes the real-time linkage between user feedback and product optimization, not only optimizes the user experience through behavior data, but also forces the supply chain to upgrade through negative review analysis, forming a closed loop of "user feedback - platform adjustment - supply chain improvement".

[0040] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention are also within the protection scope of the present invention.

Claims

1. A product management system for an online marketing platform, characterized in that, include: The multi-source information fusion module integrates basic product information, cross-platform data, and real-time market data to build a dynamic product information database. The dynamic attribute tagging module uses NLP technology to analyze user reviews and social discussions, generating and updating dynamic product tags in real time. The association recommendation module combines dynamic product tags with user behavior data to construct a three-dimensional association model of "product-scenario-user". The inventory and market linkage module uses the LSTM neural network algorithm for sales forecasting and, combined with supply chain data, automatically triggers replenishment and promotion strategies when inventory is abnormal or market popularity changes. The feedback optimization module monitors user behavior and negative reviews on the product display page, automatically adjusts the focus of product descriptions, and synchronizes optimization actions with supply chain quality inspection.

2. The product management system for an online marketing platform according to claim 1, characterized in that: The multi-source information fusion module includes: a data acquisition module, a preprocessing module, and a dynamic storage module; Data acquisition module: It uses distributed web crawlers to collect basic product information from e-commerce platforms, product mentions on social media, and user discussions. At the same time, it connects to third-party evaluation platforms and supply chain systems through API interfaces to obtain product quality scores, inventory data, and production cycle information. Preprocessing module: Cleans and standardizes the collected data; Dynamic storage module: It uses a time-series database to store processed data, maintains a dynamic product information database with an hourly incremental update mechanism, and supports multi-dimensional retrieval.

3. The product management system for an online marketing platform according to claim 1, characterized in that: The dynamic attribute label module includes: an information parsing module, a label generation and weight calculation module, and a label update module; Information parsing module: Based on NLP technology, it parses user reviews and social discussion texts, and extracts keywords describing product features through word segmentation and semantic recognition; Tag generation and weight calculation module: Combines the sentiment analysis results of the BERT model to generate dynamic tags for products, and calculates tag weights based on keyword frequency and sentiment intensity; Tag update module: The tag library is updated based on new data every 24 hours, and low-weight tags that have not been mentioned for 7 consecutive days are automatically removed.

4. The product management system for an online marketing platform according to claim 1, characterized in that: The association recommendation module includes: an association mining module, an association model construction module, and a recommendation module; Association Mining Module: Utilizes the Apriori algorithm to analyze sales data from the past 30 days and mine association rules for scenario-based product combinations; Association Model Construction Module: Based on the mined association rules, a three-dimensional association model of "product-scenario-user" is constructed, and product co-occurrence frequency, scenario matching degree and user preference tags are entered; Recommendation module: Receives user behavior trigger signals, calls 3D correlation model data to generate a list of related product recommendations, filters out products with insufficient stock or those already purchased by the user, and outputs the final recommendation result.

5. The product management system for an online marketing platform according to claim 1, characterized in that: The inventory and market linkage module includes: a sales forecasting module, an early warning and replenishment module, and a market heat response module; Sales forecasting module: Based on LSTM neural network, it takes historical sales data, holiday factors and competitor price fluctuation data as input to generate a 7-day short-term sales forecast curve; Early warning and replenishment module: Compares predicted sales with current inventory. When inventory is lower than the predicted value, a replenishment reminder is triggered and pushed to the supply chain system, and the procurement priority is updated synchronously. Market popularity response module: Real-time monitoring of the trending search rankings of product-related keywords. When the popularity reaches a preset threshold, it automatically activates homepage banner display and limited-time promotional strategies such as issuing discount coupons.

6. The product management system for an online marketing platform according to claim 1, characterized in that: The feedback optimization module includes: a user behavior monitoring module, a negative review keyword extraction module, and a strategy adjustment module; User behavior monitoring module: Through page tracking technology, it collects user behavior data in real time, such as dwell time, click location, and bounce rate on product display pages, and generates a user attention heatmap; Negative review keyword extraction module: This module segments and counts the frequency of negative product reviews, extracts high-frequency negative keywords, and tags the corresponding product issues. Strategy Adjustment Module: Optimizes the display logic of product pages based on user behavior data, and updates the focus of product descriptions based on negative review keywords.

7. The product management system for an online marketing platform according to claim 2, characterized in that: The data cleaning process involves removing redundant data based on a rule engine; the standardization process involves using standardized field formats for product data.

8. The product management system for an online marketing platform according to claim 6, characterized in that: The page tracking technology involves embedding lightweight monitoring code in key interactive areas of the product display page, and collecting timestamps, click coordinates, and page scroll depth data of user behavior through asynchronous data transmission.

Images