Searching method and device based on e-commerce platform

Through natural language processing and semantic understanding technology, combined with user behavior data and multi-modal input conversion, the problem that traditional search methods cannot accurately understand user intentions is solved, and the accurate, fast and personalized search services of the e-commerce platform are realized, which improves the User experience and satisfaction.

CN119089029AInactive Publication Date: 2024-12-06SHENZHEN GLOBALBRANDS TECH CO LTD

Patent Information

Application Number
CN202411564192.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-11-05
Publication Date
2024-12-06
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The search methods of traditional e-commerce platforms cannot accurately understand user intentions and are difficult to provide personalized and fast search results, especially when dealing with complex user needs and multimodal inputs.

Method used

Using natural language processing, user behavior data, multi-modal input conversion, semantic understanding and dynamic inventory verification technologies, we can analyze user query requests, identify user actual needs, combine user historical behavior and context information for personalized recommendations, and dynamically Adjust search results to ensure real-time and relevance.

Benefits of technology

It realizes accurate and fast product search services, improves user experience and satisfaction, provides personalized recommendations and forward-looking product predictions, ensuring the credibility, practicality and smooth experience of search results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089029A_ABST
    Figure CN119089029A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of computers, in particular to a search method based on an e-commerce platform, which comprises the following steps: receiving a query request from a user, analyzing the query request through a natural language processing technology, calling related data of the user, and enhancing the analyzed query request based on the related data, obtaining corresponding query request content; calling a corresponding conversion tool, performing corresponding format conversion on the query request content based on the conversion tool, performing semantic understanding on the converted query request content, and identifying the actual demand of the user based on a pre-trained identification model; the method comprises the following steps: determining a corresponding candidate commodity set based on an actual demand, checking whether the inventory state and price information of each commodity are latest or not before the candidate commodity set is displayed to a user, sorting candidate commodity lists in the candidate commodity set to determine a search result, and presenting the search result to the user. According to the invention, accurate and rapid search service of the e-commerce platform can be realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of computers, and in particular to a search method and device based on an e-commerce platform. Background Art

[0002] With the rapid development of the internet, e-commerce platforms have become one of the primary shopping channels for consumers. On these platforms, search functions are a crucial tool for users to find the products they need. Traditional search methods typically rely on keyword matching. While simple and straightforward, this approach has numerous shortcomings when dealing with complex user needs, such as an inability to accurately understand user intent, difficulty providing personalized search results, and limited support for multimodal input.

[0003] Furthermore, with the advancement of big data and artificial intelligence technologies, users have higher expectations for search experience, hoping for more accurate and fast search services. As can be seen from the above, the problem of how to achieve accurate and fast search services on e-commerce platforms remains to be solved. Summary of the Invention

[0004] In order to achieve accurate and fast search services on e-commerce platforms, the present application provides a search method and device based on e-commerce platforms.

[0005] In a first aspect, the present application provides a search method based on an e-commerce platform, which adopts the following technical solutions: A search method based on an e-commerce platform comprises: receiving a query request from a user, parsing the query request using natural language processing technology, retrieving relevant data of the user, enhancing the parsed query request based on the relevant data, and obtaining corresponding query request content, wherein the relevant data includes browsing history, purchase history and geographic location; retrieving a corresponding conversion tool, converting the query request content into a corresponding format based on the conversion tool, performing semantic understanding on the converted query request content, and identifying the user's actual needs based on a pre-trained recognition model; determining a corresponding set of candidate products based on the actual needs, verifying whether the inventory status and price information of each product are up to date before displaying them to the user, sorting a list of candidate products in the candidate product set to determine as search results, and presenting the search results to the user.

[0006] By adopting the above technical solutions, we achieve accurate and fast product search services by integrating natural language processing, user behavior data, multimodal input conversion, semantic understanding, and dynamic inventory verification. This approach not only accurately understands the user's query intent but also combines the user's historical behavior and contextual information to provide personalized recommendations, ensuring the relevance and real-time nature of search results, significantly improving the user experience.

[0007] Optionally, in the process of performing semantic understanding on the converted query request content, the method also includes: retrieving the converted query request content, identifying and expanding synonym data of the query request content; determining corresponding semantic understanding data based on the query request content and the synonym data; and extracting entity data from the semantic understanding data, wherein the entity data includes brand name and model.

[0008] By employing this technical solution, during the semantic understanding of the converted query content, the system retrieves the converted query content, identifies and expands the synonym data within it, combines the query content and synonym data to determine the corresponding semantic understanding data, and extracts entity data (such as brand name and model number) from it. This allows for more accurate capture of users' actual needs and improves the relevance and precision of search results. This not only enhances understanding of user intent but also ensures more precise and targeted search results by identifying specific entity information.

[0009] Optionally, the method also includes: conducting a conversation with the user and feeding back the actual needs to the user in the form of questions; obtaining corresponding feedback data from the user based on the user's answers to the questions; in multiple rounds of conversations, collecting contextual data in the user feedback data, and optimizing search results based on the feedback data and the contextual data.

[0010] By employing this technical solution, we engage in multiple rounds of conversations with users, providing feedback on identified needs in the form of questions. We then collect feedback data and contextual information based on their responses. By continuously optimizing search results, we not only gradually clarify the user's query intent but also dynamically adjust and improve search results during the conversation, providing more accurate and personalized product recommendations and significantly improving user experience and satisfaction.

[0011] Optionally, the method also includes: retrieving the user's search history data and purchase behavior data; identifying the user's preference data based on the search history data and the purchase behavior data, wherein the preference data includes high-frequency purchase periods and common product categories; retrieving current seasonal data and the preference data, predicting the user's demand change data within a specific time period, and updating the search results based on the demand change data.

[0012] By employing this technical solution, we can identify user preferences (such as high-frequency purchase periods and common product categories) by accessing user search history and purchase behavior data. This information, combined with current seasonality data, can be used to predict changes in user demand over specific time periods. Based on this demand change data, the system can dynamically update search results, providing users with relevant product recommendations in advance, improving the timeliness and relevance of search results.

[0013] Optionally, in the process of sorting the candidate product list in the candidate product set to determine it as search results, the method also includes: collecting and integrating content data generated by users on the e-commerce platform, wherein the content data includes product reviews, ratings, questions and answers; filtering and collecting the content data to determine the corresponding content set; and sorting the candidate product list based on the content set to update the search results.

[0014] By employing the aforementioned technical solution, we collect and integrate user-generated content data on e-commerce platforms (such as product reviews, ratings, and Q&A), filter and process this content data to form content collections. Based on this content collection, the system re-ranks the list of candidate products, thereby updating search results and making the products presented more credible and useful.

[0015] Optionally, the method also includes: obtaining real-time background data of the user, wherein the real-time background data includes click-through rate, dwell time, and purchase conversion rate; retrieving pre-set factor weight data, and adjusting corresponding factor weight data based on the real-time background data, wherein the factor weight includes price, rating, and sales volume; and reordering search results based on the adjusted factor weight data.

[0016] By employing this technical solution, we capture real-time user data (including click-through rate, dwell time, and purchase conversion rate) and draw upon pre-set factor weightings (such as price, rating, and sales volume). Based on this real-time feedback, we dynamically adjust the weightings of various factors. The system reorders search results based on these adjusted factor weightings, optimizing the order of product display and ensuring that search results better align with users' actual needs and preferences, improving search relevance and user experience.

[0017] Optionally, the method also includes: obtaining the real-time network connection status corresponding to the user, wherein the real-time network connection status includes bandwidth, delay, and packet loss rate; retrieving a mapping relationship table between the network connection status preset in the e-commerce platform and the loading data, wherein the loading data includes loading speed and loading quality; determining target loading data based on the real-time network connection status and the mapping relationship table, and adjusting the loading data of the search results based on the target loading data.

[0018] By adopting this technical solution, the user's real-time network connection status (including bandwidth, latency, and packet loss rate) is obtained, and a mapping table pre-set in the e-commerce platform between network connection status and loading data (loading data includes loading speed and loading quality) is retrieved. The target loading data is determined based on the real-time network status, thereby dynamically adjusting the loading speed and quality of search results. This ensures a smooth and high-quality search experience under various network conditions, avoiding slow page loading or lag caused by network issues, and improving the overall user experience.

[0019] In a second aspect, the present application provides a search device based on an e-commerce platform, which adopts the following technical solutions: A search device based on an e-commerce platform, comprising: a query request content acquisition module that receives a query request from a user, parses the query request using natural language processing technology, retrieves relevant user data, and enhances the parsed query request based on the relevant data to obtain the corresponding query request content, wherein the relevant data includes browsing history, purchase history, and geographic location; The actual demand identification module calls the corresponding conversion tool, converts the query request content into the corresponding format based on the conversion tool, performs semantic understanding on the converted query request content, and uses the pre-trained recognition model to identify the user's actual demand; The search result determination module determines the corresponding candidate product set based on the actual demand, verifies whether the inventory status and price information of each product are up to date before displaying them to the user, sorts the list of candidate products in the candidate product set to determine them as search results, and presents the search results to the user.

[0020] In a third aspect, the present application provides a search method based on an e-commerce platform, which adopts the following technical solutions: A search method based on an e-commerce platform includes a processor running a program of any one of the above-mentioned search methods based on an e-commerce platform.

[0021] In a fourth aspect, the present application provides a storage medium, which adopts the following technical solution: A storage medium stores a program of any one of the above-mentioned search methods based on an e-commerce platform.

[0022] In summary, this application includes at least one of the following beneficial technical effects: Precise search: Accurately capture user query intent through semantic understanding and synonym expansion.

[0023] Personalized recommendations: Provide more personalized and relevant product recommendations through multi-round conversations and user historical data analysis.

[0024] Forward-looking recommendations: Predict future demand based on user historical behavior and seasonal trends, and provide users with relevant products in advance.

[0025] Credibility and practicality: Integrate user-generated content data to ensure that search results are more credible and practical.

[0026] Smooth experience: Dynamically adjust loading speed and quality based on the user's real-time network conditions to ensure a smooth search experience.

[0027] Continuous optimization: Dynamically adjust factor weights based on real-time user feedback to continuously optimize the relevance of search results and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 The figure is a flow chart showing a search method based on an e-commerce platform according to an exemplary embodiment.

[0029] Figure 2 The figure is a structural block diagram of a search method and apparatus based on an e-commerce platform according to an exemplary embodiment. DETAILED DESCRIPTION

[0030] Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings.

[0031] Throughout this specification, reference to the terms "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with the embodiment or example is included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0032] The present application embodiment discloses a search method based on an e-commerce platform, referring to Figure 1 ,include: S100, receiving a query request from a user, parsing the query request using natural language processing technology, retrieving relevant data of the user, enhancing the parsed query request based on the relevant data, and obtaining corresponding query request content.

[0033] Among them, users enter query requests through the search box of the e-commerce platform, which can be in the form of text, voice or pictures. Natural language processing (NLP) technology is used to preliminarily analyze the user's query requests and extract keywords and phrases; for example, if the user enters "the latest Apple mobile phone", the system will extract keywords such as "latest model", "Apple", and "mobile phone".

[0034] The system also retrieves relevant user data from the database, including browsing history, purchase history, and geographic location. Based on the user's browsing and purchase history, the system identifies the user's preferences and interests. Combined with the user's geographic location, the system further refines the query request. For example, if the user is in a certain city, the system may prioritize inventory availability in that city.

[0035] This contextual information is combined with the parsed query request to generate a more specific query request content; by combining the user's personal behavior data, the query request is made more in line with the user's actual needs and the relevance of the search results is improved.

[0036] S110, calling a corresponding conversion tool, converting the query request content into a corresponding format based on the conversion tool, performing semantic understanding on the converted query request content, and identifying the user's actual needs based on a pre-trained recognition model.

[0037] If a query request contains non-text data (such as images or audio), the system invokes appropriate conversion tools to convert the non-text data into a processable text format. For example, image recognition technology can be used to extract text from an image or to transcribe audio into text. This ensures that all query types can be correctly processed by the system, supports multimodal input, and improves the user experience.

[0038] Secondly, the system conducts in-depth semantic understanding of the converted query content, including synonym expansion, contextual analysis, and entity recognition. For example, it recognizes that "latest model" likely refers to a recently released model, "Apple" refers to the brand, and "mobile phone" refers to the product category. The system then uses pre-trained recognition models (such as deep learning models) to identify the user's actual needs. These recognition models have been trained with extensive data and are capable of understanding complex query intent. Through semantic understanding and need recognition, the system can more accurately capture the user's query intent, thereby providing more precise search results.

[0039] S120, determining a corresponding set of candidate products based on actual needs, verifying whether the inventory status and price information of each product are up to date before displaying them to the user, sorting the candidate product list in the candidate product set to determine them as search results, and presenting the search results to the user.

[0040] Among them, based on the identified actual needs, a relevant set of candidate products is screened out from the product database. For example, the system will find all the "latest Apple mobile phones" that meet the conditions; the products that best meet the user's needs are screened out from a large number of products, narrowing the search scope and improving search efficiency.

[0041] In addition, before displaying the product to the user, the system checks the inventory status of each product to ensure that the product is in stock. It also checks whether the price information of each product is up to date to avoid displaying outdated price information. This ensures that the product information seen by the user is up to date and avoids a negative experience caused by insufficient stock or inaccurate prices.

[0042] At the same time, the candidate product list is sorted based on a combination of factors, such as relevance, ratings, sales volume, price, etc. The sorting weight can be dynamically adjusted based on real-time user feedback to ensure that the search results are more in line with user preferences. The sorted search results are then presented to the user, providing an intuitive and relevant product list. Through reasonable sorting, users are ensured to see the most relevant and attractive products first, improving their shopping experience and satisfaction.

[0043] Based on the above solution, natural language processing and semantic understanding technologies are used to accurately capture user query intent. Personalized recommendations are provided by combining historical user behavior data to provide more personalized product recommendations. Multiple query input methods, such as text, voice, and images, are supported to enhance the user experience. Before displaying search results, the inventory status and price of each product are verified to ensure real-time and accurate information. Search results are sorted by comprehensively considering multiple factors to ensure that users see the most relevant and attractive products first. This enables e-commerce platforms to provide more accurate, fast, personalized, and reliable search services, significantly improving user shopping experience and satisfaction.

[0044] In an embodiment of the present application, during the process of semantically understanding the converted query request content, the method further includes: S111 , retrieve the converted query request content, identify and expand synonym data of the query request content.

[0045] The system retrieves the query content from the previous step, which has been formatted. For example, if the user inputs voice or images, the system has already converted them into text. Natural language processing techniques (such as word vector models and synonym dictionaries) are used to identify keywords in the query. For example, for the query "latest iPhone," the system identifies the keywords "latest," "apple," and "phone."

[0046] Perform synonym expansion on identified keywords. Use pre-trained word vector models (such as Word2Vec and GloVe) or synonym dictionaries to find synonyms for these keywords. For example, synonyms for "latest model" might include "new model," "latest release," and "newly launched"; synonyms for "apple" might include "Apple" and "iPhone"; and synonyms for "mobile phone" might include "smartphone" and "mobile phone."

[0047] S112: Determine corresponding semantic understanding data based on the query request content and synonym data.

[0048] The original query content and the expanded synonym data are combined to form a richer semantic understanding dataset. For example, "latest Apple phone" can be expanded to "latest / new model / latest release / newly launched Apple / Apple / iPhone phone / smartphone / mobile phone."

[0049] Semantic representation technology is used to perform deep semantic representation on the integrated data. It can capture the complex relationships and contextual information between words and generate more accurate semantic representations.

[0050] S113, extracting entity data from the semantic understanding data, wherein the entity data includes brand name and model.

[0051] Named Entity Recognition (NER) technology is used to extract specific entity information from semantic understanding data. NER models can identify entities such as brand names, models, and product categories. For example, in the query "latest Apple phone," the system will identify "Apple" as the brand name and "phone" as the product category. The identified entities are then normalized to ensure consistency and standardization of entity names; for example, "Apple" is unified into "Apple" and "phone" into "smartphone."

[0052] Through the execution steps from S111 to S113 above, the following comprehensive effects can be achieved in the process of semantic understanding of the converted query request content: Expand the query scope: through synonym expansion, ensure that the search results can cover more relevant products; Deep semantic understanding: use advanced semantic representation technology to more deeply understand the user's query intention; Accurate entity matching: through entity recognition and normalization, ensure that the search results are highly matched with the user's actual needs.

[0053] It should be noted that the solution of this application also provides a dialogue mode with the user, including the following steps: S1211: Communicate with users and provide feedback on actual needs in the form of questions.

[0054] After parsing and understanding the user's initial query, the system initiates an interactive conversational assistant. Using natural language processing (NLP) technology, the assistant generates questions to further clarify the user's specific needs. Based on preliminary semantic understanding and synonym expansion, the system generates specific questions. For example, if a user queries for "latest Apple phone," the system might ask, "Are you interested in the latest iPhone models?" or "Do you have any specific feature requirements?" The generated questions are then fed back to the user in text or voice format, awaiting their response. Through this conversation, the system can gradually clarify the user's query intent, ensuring that search results better meet the user's specific needs.

[0055] S1212: Obtain corresponding feedback data from the user based on the user's answer to the question.

[0056] Users answer questions posed by the system through text, voice, or other means. The system then uses NLP technology to parse the user's responses and extract key information. The user's responses are then recorded as part of the feedback data, which may include preferences for specific brands, models, features, and more. The user's profile can also be updated based on their responses. For example, if a user expresses interest in a particular iPhone model, the system will add this preference information to the user profile.

[0057] By collecting direct feedback from users, the system can more accurately capture users' actual needs and provide a basis for subsequent search result optimization.

[0058] S1213, in multiple rounds of conversations, collecting context data from user feedback data, and optimizing search results based on the feedback data and the context data.

[0059] The system engages in multiple rounds of conversation with the user, asking questions and receiving responses. Each round may cover different aspects, such as price range, color preferences, and functional requirements. During these rounds, the system not only records the user's direct responses but also collects contextual information, such as whether the user's answers are consistent and whether any new requirements have emerged. The system maintains a conversation history containing the questions and answers from each round. This helps maintain the coherence and consistency of the conversation. Through multiple rounds of conversation, the system can gradually accumulate rich contextual information, better understand the user's complex needs, and improve the relevance of search results.

[0060] Furthermore, feedback and contextual data collected from multiple rounds of conversations are integrated. This data may include specific user preferences, excluded options, and new needs. Search parameters are adjusted based on this integrated data. For example, if a user explicitly expresses disinterest in a particular brand, the system will exclude products from that brand. If a user mentions a particular feature of particular interest, the system will prioritize products with that feature.

[0061] Based on the adjusted search parameters, the candidate product list is re-ranked. The ranking algorithm may combine multiple factors, such as relevance, ratings, and sales volume, and dynamically adjust the weights based on user feedback. The re-ranked search results are then presented to the user. These results better meet the user's actual needs and preferences, improving the user experience.

[0062] By executing steps S1211 to S1213 above, the user's query intent can be gradually clarified and refined over multiple rounds of conversations with the user, and search results can be dynamically optimized based on user feedback and contextual data. This not only improves the accuracy and relevance of search results, but also enhances the interactivity and personalization of the user experience, allowing users to find products that meet their needs more quickly.

[0063] In the process of sorting the candidate product list in the candidate product set and determining it as the search result, the method further includes: S1221, collect and integrate content data generated by users on the e-commerce platform.

[0064] The system collects user-generated content data from the e-commerce platform's database. This data includes, but is not limited to, product reviews, ratings, and Q&A. Data scraping tools or APIs are used to extract this content data from the platform periodically or in real time.

[0065] Integrate the collected content data into a unified dataset. This may involve merging data from different sources (such as reviews, ratings, and Q&A) into a single data structure. For example, create a table containing all product IDs, corresponding review text, rating values, and Q&A records.

[0066] S1222: Filter and collect content data to determine a corresponding content set.

[0067] The collected content data is pre-processed, including removing noise data (such as advertisements and spam), formatting data (such as standardizing date formats and cleaning HTML tags), etc. For example, regular expressions or natural language processing techniques are used to identify and delete spam comments.

[0068] The quality of each piece of content is then assessed. Natural language processing techniques (such as sentiment analysis and topic modeling) can be used to identify high-quality reviews and useful questions and answers. For example, sentiment analysis can be used to determine the emotional leaning (positive, negative, neutral) of a review, while topic modeling can be used to identify the main topics of the reviews.

[0069] Finally, based on the results of the quality assessment, low-quality or useless content data is filtered out, and high-quality content data is collected to form a filtered content collection that contains the most valuable user feedback information.

[0070] S1223 , sorting the candidate product list based on the content set to update the search results.

[0071] Key features are extracted from the content collection, such as average rating, number of reviews, percentage of positive reviews, and popular questions and answers. For example, the average rating, total number of reviews, and percentage of positive reviews are calculated for each product. Weights are assigned to each feature. These weights can be determined based on historical data and user behavior analysis. For example, if users are more interested in products with high ratings, a higher weight can be assigned to the average rating. Based on the extracted features and assigned weights, a comprehensive score is calculated for each candidate product. For example, a weighted sum method is used to calculate the comprehensive score: Comprehensive score = Average rating * Weight 1 + Number of reviews * Weight 2 + Percentage of positive reviews * Weight 3.

[0072] The candidate product list is reordered based on the calculated comprehensive scores. For example, the product list is sorted from high to low by comprehensive score, ensuring that the products with the highest scores are ranked first. The reordered candidate product list is then presented to the user as the final search result, ensuring that the user sees the most relevant products that are most recommended by other users.

[0073] By executing steps S1221 to S1223 as described above, the following comprehensive effects can be achieved in the process of sorting the list of candidate products in the candidate product set and determining them as search results: by integrating and utilizing user-generated content data, the system can better understand the actual performance of the products and the real feedback from users; based on high-quality user reviews, ratings, and question-and-answer information, it ensures that the products in the search results are recognized and recommended by users; by dynamically adjusting the sorting of search results, it ensures that users see the most relevant and valuable products first, thereby improving user shopping satisfaction.

[0074] Furthermore, the method further comprises: S1231, obtain the user's real-time background data.

[0075] The system collects real-time user behavior data through logging and analysis tools. This data includes, but is not limited to, click-through rate, dwell time, and purchase conversion rate. For example, the system can record how long a user stays on a product page, how often they click on a product link, and whether they ultimately complete a purchase.

[0076] S1232, retrieve pre-set factor weight data, and adjust the corresponding factor weight data based on real-time background data.

[0077] The system pre-sets weights for a series of factors, including price, rating, and sales volume. Initial weights may be determined based on historical data analysis or expert experience; for example, the initial weights might be 0.3 for price, 0.4 for rating, and 0.3 for sales volume. These preset weights are stored in a configuration file or database table, allowing the system to easily access and update them at any time.

[0078] Analyze collected real-time backend data to extract key metrics, such as the average click-through rate, average dwell time, and purchase conversion rate for each product. Use statistical methods or machine learning models (such as linear regression and random forests) to analyze the relationship between these metrics and various factors. Based on the analysis results, dynamically adjust the weight of each factor. For example, if you find that users are more likely to click on and purchase products with high ratings, you can increase the weight of the rating; if users are more concerned about price, you can increase the weight of price. For example, if the analysis results show that products with high ratings have higher click-through rates and purchase conversion rates, you can adjust the weight of the rating from 0.4 to 0.5, while appropriately reducing the weight of other factors.

[0079] S1233, reorder the search results based on the adjusted factor weight data.

[0080] Based on the adjusted weights, the comprehensive score for each candidate product is recalculated. The comprehensive score is typically a weighted sum. For example, comprehensive score = price * weight 1 + score * weight 2 + sales volume * weight 3. Based on the calculated comprehensive score, the candidate product list is reordered, typically from highest to lowest comprehensive score. For example, the product with the highest comprehensive score is placed first, and so on.

[0081] The re-ranked candidate product list is presented to the user as the final search results, ensuring that the user sees the most relevant products that best match their current preferences.

[0082] Through the above steps, this method can achieve the following combined effects in the process of dynamically adjusting the search result ranking based on the user's real-time background data: by analyzing the user's click-through rate, dwell time, and purchase conversion rate in real time, the system can better understand the user's current preferences and adjust the search result ranking accordingly; dynamically adjust the weights to make the search results more consistent with the user's actual needs, improving the user's shopping experience and satisfaction; the system continuously collects and analyzes real-time data and continuously adjusts the weights to ensure that the search results are always in the best state and adapt to changes in user behavior. Based on the above effects, e-commerce platforms can provide more accurate and personalized search services, significantly improving the user's shopping experience and satisfaction.

[0083] In addition, the search method in the embodiment of the present application also has a certain prediction function, and the specific method includes: S1241, retrieve the user's search history data and purchase behavior data.

[0084] The integrated data is analyzed and key indicators are extracted. For example, the user's search frequency, click frequency and purchase frequency in different time periods are calculated, and statistical methods or machine learning models (such as clustering algorithms and association rule mining) are used to identify users' high-frequency purchase periods and common product categories. For example, clustering algorithms can be used to find that users often shop between 8 and 10 pm; and association rule mining can be used to find that users often buy electronic products and sports equipment.

[0085] S1242, identifying user preference data based on search history data and purchase behavior data.

[0086] Based on the data analysis results, user preference data is generated. This preference data can include high-frequency purchase periods, common product categories, brand preferences, etc. For example, the following preference data may be generated: high-frequency purchase period: 8:00 PM to 10:00 PM; common product categories: electronics, sports equipment; brand preferences: Apple, Nike.

[0087] S1243, retrieve current seasonal data and preference data, predict user demand change data within a specific time period, and update search results based on the demand change data.

[0088] Among them, retrieve the current seasonal data, including holidays, seasonal changes, promotions, etc. This data can be obtained from external data sources or extracted from the platform's historical data; for example, it is currently winter, Christmas is approaching, and the platform has related promotions.

[0089] Combine user preference data with current seasonality data for comprehensive analysis. Use time series analysis, machine learning models (such as ARIMA and LSTM), or other forecasting models to predict changes in user demand within a specific time period. For example, by combining the high-frequency purchase period between 8:00 PM and 10:00 PM and common winter promotions, you can predict increased demand for electronics and sports equipment before Christmas.

[0090] Based on the results of comprehensive analysis, we generate data on user demand changes within a specific time period. This data can include expected purchase frequency, expected product categories, and more. For example, we might generate the following demand change data: Expected purchase frequency: Purchase frequency between 8:00 PM and 10:00 PM will increase by 20%; Expected product categories: Electronics (especially new iPhones) and sports equipment (especially warm clothing).

[0091] In addition, search parameters can be adjusted based on predicted demand change data. For example, if it is predicted that users’ demand for electronic products will increase, the weight of electronic products can be increased. If it is predicted that users’ purchase frequency will increase between 8 and 10 p.m., relevant products can be displayed preferentially during that time period.

[0092] For example, search parameters can be adjusted as follows: electronics and sports equipment are weighted more highly; promotional items are prioritized between 8 PM and 10 PM. Reorder search results: Based on the adjusted search parameters, the list of candidate products is reordered. The sorting algorithm may combine multiple factors, such as relevance, ratings, and sales volume, and dynamically adjust weights based on predicted changes in demand. For example, the list of products may be sorted from high to low based on the adjusted overall rating, ensuring that products that meet predicted demand are ranked first. Finally, the reordered list of candidate products is presented to the user as the final search results.

[0093] Through the above steps, the method can achieve the following comprehensive effects in the process of predicting demand changes and updating search results based on users' search history data and purchase behavior data: by analyzing users' search history and purchase behavior, the system can better understand users' interests and behavior patterns, and predict future demand changes accordingly; dynamically adjust search results based on predicted demand change data, making it easier for users to find products that meet their current needs, thereby improving shopping experience and satisfaction; by combining seasonal data and user preferences, the system can predict users' demand changes in advance, provide forward-looking product recommendations, and enhance users' sense of trust.

[0094] When users search on e-commerce platforms, the current network conditions are also taken into account. The methods also include: S1251, obtaining the real-time network connection status of the user.

[0095] The system uses network monitoring tools to periodically or in real time monitor the user's network connection status. These tools can measure key metrics such as bandwidth, latency, and packet loss. For example, the system can use JavaScript or other front-end technologies to perform network speed tests in the user's browser, or call third-party APIs to obtain this information. The system collects and stores the detected real-time network connection status data (including bandwidth, latency, and packet loss) in the system database. This data can be used for subsequent analysis and processing.

[0096] S1252, retrieve a mapping relationship table between network connection status and loading data preset in the e-commerce platform.

[0097] The system pre-sets a mapping table between network connection status and loading data, which is stored in a configuration file or database for easy access and updating by the system. This table defines the loading speed and quality to be used under different network conditions.

[0098] S1253: Determine target loading data based on the real-time network connection status and the mapping relationship table, and adjust the loading data of the search results based on the target loading data.

[0099] Among them, the system will match the collected real-time network connection status data (bandwidth, latency, packet loss rate) with the mapping relationship table to determine the loading speed and loading quality corresponding to the current network status; for example, if the user's bandwidth is 30Mbps, the latency is 70 ms, and the packet loss rate is 2%, then according to the mapping relationship table, the system will determine the loading speed as "medium speed" and the loading quality as "medium quality".

[0100] Generate target loading data based on the matching results. This data will be used to adjust the search result loading strategy. For example, generate the following target loading data: Loading speed: medium speed; Loading quality: medium quality.

[0101] Adjust the search result loading strategy based on the target data loading target. Specific measures may include: image resolution adjustment: lowering image resolution to reduce data transmission and speed up loading; resource loading order optimization: prioritizing text content and important information, delaying the loading of large media files; data compression: compressing transmitted data to reduce data transmission and increase loading speed; lazy loading: implementing lazy loading, loading images and other resources only when the user scrolls to the corresponding position; and reducing the number of recommended products: reducing the number of products displayed in search results when network conditions are poor to reduce data transmission.

[0102] Apply the adjusted loading strategy to the front-end page. For example, for the "Medium Speed" and "Medium Quality" loading strategies, the system may reduce image resolution and prioritize loading text content. The system continuously monitors the user's network status and updates the loading strategy in real time based on changes, ensuring that users always receive the best search experience.

[0103] Through the above steps, the following comprehensive effects can be achieved in the process of adjusting the search result loading data based on the user's real-time network connection status: by dynamically adjusting the loading strategy, the system can significantly improve the page loading speed, especially under poor network conditions; even under limited network conditions, the system can ensure the quality of search results through reasonable loading strategies and provide a good user experience; by optimizing the loading strategy, the system can reduce the user's waiting time and improve the user's shopping experience and satisfaction; the system can dynamically adjust according to the user's real-time network status to ensure that the search results always meet the current network conditions and provide a consistent and high-quality user experience.

[0104] The present application embodiment discloses a search device based on an e-commerce platform, referring to Figure 2 , devices include but are not limited to: The query request content acquisition module 200 receives a query request from a user, parses the query request using natural language processing technology, retrieves relevant user data, and enhances the parsed query request based on the relevant data to obtain the corresponding query request content, wherein the relevant data includes browsing history, purchase history, and geographic location; The actual demand identification module 210 calls a corresponding conversion tool, converts the query request content into a corresponding format based on the conversion tool, performs semantic understanding on the converted query request content, and uses a pre-trained recognition model to identify the user's actual demand; The search result determination module 220 determines the corresponding candidate product set based on actual needs, verifies whether the inventory status and price information of each product are up to date before displaying them to the user, sorts the candidate product list in the candidate product set to determine them as search results, and presents the search results to the user.

[0105] Furthermore, the device includes but is not limited to: A query request content retrieval module is used to retrieve the converted query request content, identify and expand synonym data of the query request content; A semantic understanding data determination module is used to determine corresponding semantic understanding data based on the query request content and synonym data; The entity data extraction module is used to extract entity data from the semantic understanding data, where the entity data includes brand name and model.

[0106] Furthermore, the device includes but is not limited to: The feedback module is used to communicate with users and feedback their actual needs in the form of questions; A feedback data acquisition module is used to obtain the user's corresponding feedback data based on the user's answer to the question; The search result optimization module collects contextual data from user feedback data during multiple rounds of conversations and uses the feedback data and contextual data to optimize search results.

[0107] Furthermore, the device includes but is not limited to: Search history data retrieval module, used to retrieve user search history data and purchase behavior data; A preference data identification module is used to identify user preference data based on search history data and purchase behavior data, where the preference data includes high-frequency purchase periods and common product categories; The demand change data prediction module retrieves current seasonal data and preference data to predict the user's demand change data within a specific time period and updates the search results based on the demand change data.

[0108] Furthermore, the device includes but is not limited to: The content data collection module is used to collect and integrate content data generated by users on the e-commerce platform, where the content data includes product reviews, ratings, and questions and answers; A content set determination module filters and collects content data to determine a corresponding content set; Search result update module: sorts the candidate product list based on the content set to update the search results.

[0109] Furthermore, the device includes but is not limited to: Real-time background data acquisition module, used to obtain real-time background data of users, where real-time background data includes click-through rate, dwell time, and purchase conversion rate; The factor weight data retrieval module is used to retrieve the pre-set factor weight data and adjust the corresponding factor weight data based on real-time background data. The factor weight includes price, rating, and sales volume. The search result ranking module is used to re-rank the search results based on the adjusted factor weight data.

[0110] Furthermore, the device includes but is not limited to: A real-time network connection status acquisition module is used to obtain the real-time network connection status of the user, wherein the real-time network connection status includes bandwidth, delay, and packet loss rate; A mapping relationship table retrieval module is used to retrieve a mapping relationship table between network connection status and loading data preset in the e-commerce platform, wherein the loading data includes loading speed and loading quality; The target loading data determination module is used to determine the target loading data based on the real-time network connection status and the mapping relationship table, and adjust the loading data of the search result based on the target loading data.

[0111] An embodiment of the present application further discloses a search method based on an e-commerce platform, comprising a processor in which a program of any one of the above-described search methods based on an e-commerce platform is running.

[0112] An embodiment of the present application further discloses a storage medium storing a program of any one of the above-mentioned search methods based on an e-commerce platform.

[0113] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A search method based on an e-commerce platform, characterized in that: include: Receiving a query request from a user, parsing the query request using natural language processing technology, retrieving relevant data of the user, enhancing the parsed query request based on the relevant data, and obtaining corresponding query request content, wherein the relevant data includes browsing history, purchase history, and geographic location; Retrieving a corresponding conversion tool, converting the query request content into a corresponding format based on the conversion tool, performing semantic understanding on the converted query request content, and identifying the actual needs of the user based on a pre-trained recognition model; A corresponding set of candidate products is determined based on the actual demand. Before displaying the products to the user, the inventory status and price information of each product are verified to be up to date. The candidate product list in the candidate product set is sorted to determine the search results, and the search results are presented to the user.

2. The search method based on the e-commerce platform according to claim 1, characterized in that: In the process of semantically understanding the converted query request content, the method further includes: Retrieving the converted query request content, identifying and expanding synonym data of the query request content; Determining corresponding semantic understanding data based on the query request content and the synonym data; Entity data is extracted from the semantic understanding data, wherein the entity data includes brand name and model.

3. The search method based on the e-commerce platform according to claim 2, characterized in that: The method also includes: Have a dialogue with the user and feedback the actual needs to the user in the form of questions; Obtaining corresponding feedback data from the user based on the user's answer to the question; In multiple rounds of conversations, context data in user feedback data is collected, and search results are optimized based on the feedback data and the context data.

4. The search method based on the e-commerce platform according to claim 3, characterized in that: The method also includes: Retrieve user search history data and purchase behavior data; Identifying user preference data based on the search history data and the purchase behavior data, wherein the preference data includes high-frequency purchase time periods and common commodity categories; Retrieve current seasonal data and the preference data, predict the user's demand change data within a specific time period, and update the search results based on the demand change data.

5. The search method based on the e-commerce platform according to claim 1, characterized in that: In the process of sorting the candidate commodity list in the candidate commodity set to determine it as the search result, the method further includes: Collect and integrate content data generated by users on the e-commerce platform, wherein the content data includes product reviews, ratings, and questions and answers; Filtering and collecting the content data to determine a corresponding content set; The candidate product list is sorted based on the content set to update the search result.

6. The search method based on the e-commerce platform according to claim 1, characterized in that: The method also includes: Obtaining real-time background data of users, wherein the real-time background data includes click-through rate, dwell time, and purchase conversion rate; Retrieving pre-set factor weight data, and adjusting the corresponding factor weight data based on the real-time background data, wherein the factor weight includes price, rating, and sales volume; Re-rank search results based on adjusted factor weight data.

7. The search method based on the e-commerce platform according to claim 1, characterized in that: The method also includes: Obtain the real-time network connection status of the user, including bandwidth, delay, and packet loss rate; Retrieving a mapping relationship table between a preset network connection status and loading data in the e-commerce platform, wherein the loading data includes loading speed and loading quality; Target loading data is determined based on the real-time network connection status and the mapping relationship table, and loading data of the search results is adjusted based on the target loading data.

8. A search device based on an e-commerce platform, characterized in that: include: A query request content acquisition module receives a query request from a user, parses the query request using natural language processing technology, retrieves relevant data of the user, and enhances the parsed query request based on the relevant data to obtain corresponding query request content, wherein the relevant data includes browsing history, purchase history and geographic location; The actual demand identification module calls the corresponding conversion tool, converts the query request content into a corresponding format based on the conversion tool, performs semantic understanding on the converted query request content, and uses the pre-trained recognition model to identify the actual needs of the user; The search result determination module determines the corresponding candidate product set based on the actual demand, verifies whether the inventory status and price information of each product are up to date before displaying them to the user, sorts the candidate product list in the candidate product set to determine them as search results, and presents the search results to the user.

9. A search method based on an e-commerce platform, characterized in that: It comprises a processor, in which runs a program of a search method based on an e-commerce platform as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: A program storing the search method based on the e-commerce platform as described in any one of claims 1 to 7.

Citation Information

Patent Citations

  • Commodity query redirection method and device, equipment, medium and product

    CN115203514A

  • Method and device for determining search result of shopping platform and electronic equipment

    CN117454024A

  • Shopping guide method and system based on large language model

    CN118247007A

  • Commodity information processing method and commodity searching method

    CN118537091A

  • On-screen price comparison system and device based on travel management platform, and medium

    CN118552285A

Cited By

  • Sorting recommendation method and device for commodity search results

    CN119515512A

  • Search optimization method, device and medium based on large model

    CN119782494A