Method, device and equipment for determining commodity search result

Abstract

This application provides a method, apparatus, and device for determining product search results. The method includes: acquiring a target search term input by a user in a target application software based on an intelligent agent; invoking a preset search engine to perform search processing on the target search term to obtain search results; wherein the search results are used to supplement background information of the target search term; retrieving example information from a preset knowledge base based on the target search term; wherein the knowledge base includes the correspondence between historical search terms and product categories; the example information is the correspondence between historical search terms matching the target search term and product categories; invoking a preset large model to determine the target product category corresponding to the target search term from the example information based on the target search term and the search results; and determining the product search results corresponding to the target search term based on the target product category. This method aims to improve the accuracy of product search results.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, and device for determining product search results. Background Technology

[0002] In the operation of e-commerce platforms, user search behavior is a core element of product exposure and transaction conversion.

[0003] Users can search for target products by entering search terms (Query), but in real-world scenarios, the search terms entered by users are generally brief, semantically ambiguous, and lack context, making it difficult for the platform to accurately identify the user's true search intent. Consequently, the platform cannot accurately align the search results to the correct product categories, resulting in inaccurate product search results that affect transaction conversion and user experience. Summary of the Invention

[0004] The method, apparatus, and equipment for determining product search results provided in this application are intended to improve the accuracy of product search results.

[0005] In a first aspect, embodiments of this application provide a method for determining product search results. The method is applied to an electronic device, which has target application software installed. The target application software integrates an intelligent agent, including:

[0006] Based on the intelligent agent, the target search terms entered by the user in the target application software are obtained;

[0007] A preset search engine is invoked to perform a search on the target search term to obtain search results; wherein, the search results are used to supplement the background information of the target search term;

[0008] The system retrieves example information by searching a preset knowledge base based on the target search term; wherein, the knowledge base includes the correspondence between historical search terms and product categories; and the example information is the correspondence between historical search terms that match the target search term and product categories.

[0009] The system invokes a pre-defined large model and, based on the target search term and the search results, determines the target product category corresponding to the target search term from the example information.

[0010] Based on the target product category, determine the product search results corresponding to the target search term.

[0011] In one possible implementation, the step of invoking a preset large model to determine the target product category corresponding to the target search term from the example information based on the target search term and the search results includes:

[0012] The large model is invoked to summarize and process the search results to obtain the background information of the target search term;

[0013] Construct suggestion words based on the target search term, the background information, and the example information;

[0014] Based on the prompt words and the large model, the target product category is determined.

[0015] In one possible implementation, the step of retrieving sample information matching the target search term from a preset knowledge base includes:

[0016] For each correspondence in the knowledge base, determine the similarity between the historical search term and the target search term in the corresponding correspondence;

[0017] In response to the similarity being greater than a preset threshold, the correspondence is determined as example information matching the target search term.

[0018] In one possible implementation, the method further includes:

[0019] Obtain historical search terms and the click behavior information corresponding to those historical search terms;

[0020] For each historical search term, the product category to which the historical search term belongs is determined based on the product with the highest click-through rate in the click behavior information corresponding to that historical search term;

[0021] The knowledge base is constructed based on the correspondence between the historical search terms and the product categories to which they belong.

[0022] In one possible implementation, the step of invoking a preset search engine to perform search processing on the target search term and obtaining search results includes:

[0023] Based on preset feature extraction rules, the feature information of the target search term is extracted;

[0024] Based on the feature information and the preset parameter mapping rules, the query parameters corresponding to the target search term are determined;

[0025] Based on the target search term and the query parameters, at least one search engine is invoked to perform a search on the target search term and obtain search results.

[0026] In one possible implementation, the feature information includes at least one of entity features, length features, timeliness features, and ambiguity features;

[0027] The query parameters include at least one of the following: vertical search channel, matching mode, time filter parameters, category limitation parameters, deduplication filter parameters, and sorting weight parameters.

[0028] In one possible implementation, the method further includes:

[0029] The product search results are output to the target application software for display.

[0030] Secondly, embodiments of this application provide a device for determining product search results. The device is applied to an electronic device, which has target application software installed. The target application software integrates an intelligent agent. The device includes:

[0031] The acquisition module is used to acquire, based on the intelligent agent, the target search terms entered by the user in the target application software;

[0032] The search module is used to call a preset search engine to perform search processing on the target search term and obtain search results; wherein, the search results are used to supplement the background information of the target search term;

[0033] The retrieval module is used to retrieve example information from a preset knowledge base based on the target search term; wherein, the knowledge base includes the correspondence between historical search terms and product categories; the example information is the correspondence between historical search terms and product categories that match the target search term;

[0034] The prediction module is used to call a preset large model to determine the target product category corresponding to the target search term from the example information based on the target search term and the search results.

[0035] The determination module is used to determine the product search results corresponding to the target search term based on the target product category.

[0036] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0037] The memory stores computer-executed instructions;

[0038] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0039] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0040] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0041] The method, apparatus, and device for determining product search results provided in this application embodiment acquire the target search term input by the user through an intelligent agent integrated into the target application software. It then calls a preset search engine to obtain search results to supplement the background information of the target search term. Based on the target search term, it retrieves example information from a knowledge base containing the correspondence between historical search terms and product categories. Finally, it calls a preset large model to combine the target search term and the search results to determine the target product category from the example information. Based on the target product category, it generates the corresponding product search results. This method effectively solves the problems of semantic ambiguity and inaccurate intent parsing by using a search engine to supplement background information. It also achieves accurate alignment between user intent and the platform's own categories by relying on historical example information in the knowledge base. Combined with the fusion reasoning capability of the large model, it improves the accuracy of category prediction and result matching, thereby enhancing the accuracy of user intent recognition and product search results, and optimizing the search experience and transaction conversion efficiency of e-commerce platforms. Attached Figure Description

[0042] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0043] Figure 1 Flowchart of the method for determining product search results provided in this application Figure 1 ;

[0044] Figure 2 Flowchart of the method for determining product search results provided in this application Figure 2 ;

[0045] Figure 3 A schematic diagram illustrating one application scenario provided in this application;

[0046] Figure 4 A schematic diagram of the structure of the device for determining product search results provided in this application;

[0047] Figure 5 A schematic diagram of the structure of the electronic device provided in this application.

[0048] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0049] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0050] First, the terms used in the embodiments of this application will be explained:

[0051] Long-tail search terms: These are search terms that appear less frequently (e.g., ≤3 times per week). Statistics show that e-commerce platforms generate an average of about 3,000 long-tail search terms per day.

[0052] Ambiguous search terms: These are search terms whose intent is unclear and requires background knowledge to interpret correctly. For example, a user searching for "Brand A butt bag" actually refers to the "A" series products under Brand A, which are commonly known as "butt bags" due to their rounded bottom design.

[0053] In product search scenarios, existing retrieval solutions on e-commerce platforms primarily rely on traditional search engines to interpret user query intent and match products. These solutions often identify user search needs based on keyword matching and frequency statistics, then combine general category classification rules to retrieve products and push results. However, these traditional solutions have significant technical shortcomings in practical applications. On the one hand, e-commerce platforms generate a large number of low-frequency long-tail search terms daily. Traditional search engines, lacking corresponding statistical data and contextual information, cannot effectively analyze these low-frequency search terms. Furthermore, ambiguous or vaguely expressed search terms, relying solely on keyword matching, directly leads to unclear user search intent identification and misinterpretation; for example, the common name for "A brand's butt bag" cannot be associated with the corresponding model. On the other hand, e-commerce platforms' product category classification is based on their own business rules, exhibiting strong subjectivity and customization. General category matching logic cannot accurately align with the platform's proprietary category system, easily leading to category matching errors. Ultimately, both result in low relevance of product search results, severely impacting user search experience and platform transaction conversion efficiency.

[0054] Therefore, starting from the pain points of retrieval intent parsing, the inventors of this application conceived of a function call capability based on an intelligent agent. This integrates external general-purpose search engines and large-scale models. First, the external general-purpose search engine is used to retrieve the target search term, supplementing the background information of the search term with external search results, thereby eliminating semantic ambiguity and intent ambiguity. Second, addressing the difficulty in aligning platform-owned categories, the inventors abandoned general category matching logic and instead relied on the platform's real historical retrieval data to construct a knowledge base containing the correspondence between historical search terms and product categories. By retrieving historical example information matching the target search term, a reference basis that aligns with the platform's business rules is provided for category prediction. Finally, to achieve integrated decision-making based on background information and business examples, a large-scale model is used to integrate the target search term, supplemented background information, and matched example information to accurately determine the platform's standard product category corresponding to the target search term. Then, based on this category, matching product search results are output, improving the accuracy of product search results and thus enhancing user experience and platform transaction conversion rates.

[0055] It should be noted that in the embodiments of this application, "large model" refers to "large language model" (LLM), and the two have the same meaning and are not distinguished here.

[0056] It should be noted that the user information and user data involved in the embodiments of this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with relevant laws, regulations and standards, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0057] The executing entity of this application embodiment can be an electronic device with processing capabilities, such as a computer, server, mobile terminal, wearable smart device, etc., and this application embodiment is not limited thereto. The mobile terminal can be, for example, a mobile phone or tablet. The electronic device has a target application software (APP) installed, which can be the application software corresponding to various e-commerce platforms, and this application embodiment does not limit the type and content of the application software. The target application software integrates an intelligent agent, and the electronic device can implement the technical solution of this application embodiment through the target application software integrating the intelligent agent.

[0058] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0059] Figure 1Flowchart of the method for determining product search results provided in this application Figure 1 ,like Figure 1 As shown, the method includes:

[0060] S101. Based on the intelligent agent, obtain the target search terms entered by the user in the target application software.

[0061] For example, the intelligent agent is an intelligent program module integrated into the target application software, which has the ability to interact with data and process logic, and is responsible for undertaking and executing the entire process of determining the product search results of this application.

[0062] The target application software is an application program corresponding to an e-commerce platform, allowing users to directly input their search requirements. It should be noted that this application embodiment does not limit the type of the target application software; it can be a mobile application, a desktop client application, or a web application, as long as it is compatible with the electronic device.

[0063] The target search term is the text content entered by the user on the search interface of the target application software to express the user's need to find products. It can include various forms of expression such as product abbreviation, model, characteristics, and common names. This application embodiment does not limit the length, expression form, or content type of the target search term.

[0064] In one example, the electronic device uses an agent to monitor the search input events of the target application software in real time. When the user completes text input in the search box and triggers the search operation, the agent automatically extracts the text content in the input box as the target search term.

[0065] S102. Call the preset search engine to perform search processing on the target search terms and obtain search results.

[0066] For example, the preset search engine is a general web search engine used to obtain external information related to the target search term. It should be noted that this application embodiment does not limit the type of search engine.

[0067] Search results are external information returned by the search engine that matches the target search term. This information may include category definitions, brand attributes, product models, industry knowledge, etc., to supplement the background information of the target search term and compensate for its semantic ambiguity and lack of context. It should be noted that this application embodiment does not limit the content or amount of search results.

[0068] In one example, an electronic device can generate request information based on a target search term using an intelligent agent, and send a retrieval request to a pre-defined search engine through a standardized interface protocol. After the search engine completes the retrieval based on the request information, it synchronously transmits the returned search results to the intelligent agent.

[0069] S103. Search the preset knowledge base according to the target search term to obtain example information.

[0070] For example, the preset knowledge base is a pre-built database that stores the correspondence between historical search terms and product categories of the e-commerce platform. This correspondence is generated based on the historical search behavior of real users of the e-commerce platform and the product category annotation results, and can conform to the business rules of the platform.

[0071] The example information consists of the correspondence between historical search terms that match the target search term and product categories, obtained by retrieving information from a knowledge base. It should be noted that this application embodiment does not limit the number of correspondences included in the example information.

[0072] In one example, an electronic device can vectorize the target search term based on an intelligent agent, calculate the semantic similarity between the semantic vector of the target search term and the semantic vector of the historical search terms in each corresponding relationship in the knowledge base, and select the corresponding relationships with similarity higher than a preset threshold as example information.

[0073] It is understandable that this example information can serve as a reference for the large model to perform contextual learning. Furthermore, through the aforementioned steps, the electronic device can automatically retrieve historical search terms and their accurate categories that are highly similar to the target search term from the knowledge base based on the intelligent agent, thereby helping to effectively guide the large model to make accurate predictions of subsequent product categories. For example, Table 1 is an example of example information provided in an embodiment of this application.

[0074] Table 1 Example Information

[0075]

[0076] In some embodiments, the electronic device can acquire historical search terms and corresponding click behavior information; for each historical search term, determine the product category to which the historical search term belongs based on the product with the highest click rate in the click behavior information corresponding to the historical search term; and construct a knowledge base based on the correspondence between historical search terms and the product categories to which the historical search terms belong.

[0077] Historical search terms refer to the search text entered by users on the e-commerce platform within a historical period, covering various types such as regular search terms, long-tail search terms, and ambiguous search terms. Click behavior information records user interactions such as clicks, views, and add-to-cart actions when a user enters a historical search term, reflecting the user's true search preferences. Each click behavior record may include search term identifiers, product identifiers, number of clicks, and number of impressions. Click-through rate (CTR) is the ratio of clicks to impressions for a single product corresponding to the same historical search term, used to quantify the user's level of acceptance and relevance to the product. Product categories are standardized product classification units divided by the e-commerce platform according to business rules; each product is uniquely assigned a category when it is listed on the platform.

[0078] For example, electronic devices can extract all historical search terms within a preset statistical period from the user behavior log database through the platform's backend data interface. Simultaneously, they can extract click behavior information such as the number of clicks and product exposures for each candidate product under each historical search term. For a single historical search term, the click-through rate (CTR) of each candidate product under that historical search term is calculated based on the click behavior information. Then, the target product with the highest CTR is selected through numerical comparison. Next, the category information of the target product, pre-marked in the platform's product database, is retrieved, and this category is determined as the product category to which the current historical search term belongs. All matched historical search terms and corresponding product categories are structured and organized to form one-to-one mapping pairs between historical search terms and product categories. After removing duplicate and abnormal low-quality mapping pairs, embedding technology is used to transform the historical search terms in each mapping pair into high-dimensional semantic vectors, which are then stored in an efficient vector retrieval database (such as Faiss), thus obtaining the knowledge base.

[0079] As mentioned earlier, in the e-commerce field, due to differences in the definition standards of product category systems among various e-commerce platforms, it is difficult to guarantee consistency with specific business definitions when directly using large-scale models for category prediction. Therefore, this approach can leverage the platform's historical search data to build a high-confidence knowledge base between historical search terms and product categories, providing concrete business examples for subsequent large-scale models.

[0080] S104. Call the preset large model, and determine the target product category corresponding to the target search term from the example information based on the target search term and the search results.

[0081] For example, the preset large model is a natural language processing model fine-tuned for a product category prediction task, possessing semantic understanding, information fusion, and logical reasoning capabilities, and can accurately infer the target product category based on multi-source input information. It should be noted that this application does not limit the type of the large model; for example, it can be an Ernie 3.5 model or a DeepSeek-R1 model.

[0082] The target product category is a standardized classification unit preset by the e-commerce platform for the standardized management of products. It is the final platform-side definition that carries the user's search intent.

[0083] In one example, the electronic device integrates the target search term, search results, and example information based on the intelligent agent, combining them into a model input prompt. This prompt is then input into a pre-defined large model. The large model, by fusing the category mapping logic in the target search term, search results, and example information, infers from the example information to determine the target product category corresponding to the target search term and conforming to the platform's business rules, and feeds back the target product category to the intelligent agent. For example, the prompt could be: "Please first summarize the [search results] into a sentence to obtain background information, and based on the [target search term] and [background information], select and output the most matching product category from the given [example information]." It should be noted that the content of the prompt is not limited in this embodiment.

[0084] This method provides concrete business examples to a large model by inputting example information that matches the target search term into the model. This allows the model to learn and internalize the unique category division logic and boundaries of the e-commerce platform (target application software), thereby significantly improving the consistency between the target product category corresponding to the target search term and the proprietary category system.

[0085] S105. Based on the target product category, determine the product search results corresponding to the target search term.

[0086] For example, the product search results are a list of products that match the target product category and meet the user's search intent. For example, they may include product names, prices, images, etc. This application embodiment does not limit the content of the product search results.

[0087] In one example, an electronic device can retrieve all valid products under a target product category determined by an agent based on a large model from the product database of the e-commerce platform. The retrieved products are then sorted and filtered according to preset rules such as product relevance, price range, and user ratings. Finally, product search results corresponding to the target search terms are generated for the target application software to display to the user.

[0088] The method for determining product search results provided in this application embodiment obtains the target search term input by the user through an intelligent agent integrated into the target application software, calls a preset search engine to obtain search results to supplement the background information of the target search term, retrieves example information based on the target search term in a knowledge base containing the correspondence between historical search terms and product categories, and then calls a preset large model to combine the target search term and search results to determine the target product category from the example information. Finally, the corresponding product search results are generated based on the target product category. In this way, the use of a search engine to supplement background information effectively solves the problems of semantic ambiguity of search terms and inaccurate intent parsing, and relies on historical example information in the knowledge base to achieve accurate alignment between user intent and the platform's own categories. Combined with the fusion reasoning capability of the large model, the accuracy of category prediction and result matching is improved, thereby improving the accuracy of user intent recognition and product search results, and optimizing the search experience and transaction conversion efficiency of e-commerce platforms.

[0089] Figure 2 Flowchart of the method for determining product search results provided in this application Figure 2 , Figure 3 A schematic diagram illustrating an application scenario provided in this application is shown below. Figure 2 and Figure 3 As shown, in this embodiment... Figure 1 Based on the examples, the method for determining product search results is described in detail, and the method includes:

[0090] S201. Based on the intelligent agent, obtain the target search terms input by the user.

[0091] It should be noted that this step is similar to the aforementioned step S101, and will not be repeated here.

[0092] S202. Based on the preset feature extraction rules, extract the feature information of the target search term.

[0093] For example, feature extraction rules are pre-defined text processing rules used to parse the semantics and attributes of search terms, and may include processing logic such as keyword extraction, semantic classification, and attribute annotation.

[0094] Feature information is attribute identifiers extracted from target search terms. For example, it may include at least one of entity features, length features, timeliness features, and ambiguity features, which are used to characterize the core attributes and retrieval characteristics of the target search terms.

[0095] Among them, entity features refer to noun-like information extracted from the target search term that has actual referential meaning. For example, it can include key attribute content that can directly point to the product itself, such as brand, category, model, style, material, and function. It is the basic feature that represents the core retrieval object of the search term.

[0096] Length features are text length attributes categorized based on the number of characters and words in the target search term. They are used to distinguish between short text search terms, regular-length search terms, and long text search terms, reflecting the conciseness of the search term's expression and serving as an important basis for judging the completeness of the search term information.

[0097] Timeliness features refer to time-related attribute information contained in the target search terms, such as time-related descriptions like season, holiday, year, new model, limited edition, and expired. These features are used to identify the time attributes or timeliness preferences of users' search needs and provide a time-dimensional constraint basis for configuring search parameters.

[0098] Ambiguity features are attributes used to identify situations where the target search term has unclear semantics, multiple meanings, or colloquialisms. They are used to determine whether the search term needs additional background information to determine the true intent and are key features for distinguishing between regular, clear search terms and ambiguous search terms.

[0099] In one example, an electronic device can perform text splitting, semantic recognition, and attribute labeling on target search terms based on a smart agent according to preset feature extraction rules. It can extract entity features such as brand, category, and model from the target search terms, determine the character length features of the search terms, whether they have timeliness attributes, and whether there is semantic ambiguity, and finally integrate them to obtain the feature information of the target search terms.

[0100] Specifically, the target search terms can be segmented using natural language processing tools (BERT model), and the semantic type of each segmented word in the product search scenario can be identified, such as whether it belongs to entity attributes such as brand words, category words, model words, material words, and function words. The identified words are then matched with preset entity dictionaries (such as brand dictionary, category dictionary, and model dictionary) to obtain entity features.

[0101] Specifically, the length characteristics of the target search term can be determined by calculating the total number of characters in the target search term and based on a preset length range (e.g., short text ≤ 5 characters, regular length 6–15 characters, long-tail text > 15 characters).

[0102] Specifically, the system detects whether time-related words exist in the target search terms, such as "2024," "new model," "limited edition," "spring," and "19 models." If time-related words are identified, the target search term is determined to have time-sensitive characteristics, and the time-sensitive type is further labeled (such as year, season, version, new / old, etc.). If no time-related semantics are detected, it is determined to have no time-sensitive characteristics.

[0103] Specifically, the target search term is matched with a preset ambiguity dictionary (such as colloquial terms, polysemous terms, and industry slang). If a match is successful, it is determined that there is semantic ambiguity.

[0104] For example, if the target search term is "2024 A15 charger", where A is the brand name, the feature extraction of the target search term based on the preset feature extraction rules can yield the following feature information: entity features include brand A, mobile phone accessories, and charger category; length features are regular text; timeliness features include the time attribute of "2024 model"; and there are no ambiguous features.

[0105] S203. Determine the query parameters corresponding to the target search term based on the feature information and the preset parameter mapping rules.

[0106] For example, the preset parameter mapping rules are pre-established correspondence matching rules between search term features and retrieval parameters, which are used to adapt the optimal retrieval strategy according to different search term features.

[0107] Query parameters are the configuration parameters required when calling a search engine. For example, they may include at least one of the following: vertical search channel, matching mode, time filter parameters, category limitation parameters, deduplication filter parameters, and sorting weight parameters.

[0108] Vertical search channels are exclusive search paths that search engines divide according to information domains and content types, such as encyclopedia channels, product information channels, and Q&A channels. Different channels correspond to exclusive search data sources. Choosing the appropriate vertical channel can make search results more closely match the domain attributes of the search terms and improve the accuracy of background information.

[0109] Matching patterns are keyword association rules set by search engines for target search terms. These can include types such as exact match, fuzzy match, semantic association match, and prefix-suffix match. They are used to specify the degree of association between search results and target search terms and determine the matching granularity of search results.

[0110] The time filtering parameter is a configuration parameter used to limit the time range in which search results are generated or updated. It can be set to the past week, the past month, or all time periods, etc., to filter out results that meet the time requirements for search terms with time-sensitive attributes, ensuring the timeliness of background information.

[0111] Category limitation parameters are used to constrain the product or information category to which the search scope belongs. They can limit the search scope to specified categories such as digital products, apparel, and home furnishings, avoiding interference from search results in unrelated fields and improving the relevance of the background information obtained.

[0112] The deduplication filter parameter is used to remove duplicate content, redundant information, advertising content, and low-quality invalid web pages from search results. This parameter can purify search results, retaining valid and unique core information, and providing clean data for subsequent background information extraction.

[0113] The ranking weight parameter is used to set the weight configuration for the ranking of search results. It can adjust the weight ratio of dimensions such as relevance, information authority, and content completeness, so that more valuable and relevant background information is presented first.

[0114] In one example, an electronic device can match the extracted feature information of the target search term with preset parameter mapping rules based on the intelligent agent. According to the feature information such as the entity type and degree of ambiguity of the target search term, the corresponding retrieval channel, keyword matching method and filtering rules are selected, and finally the query parameters suitable for the target search term are determined.

[0115] For example, if the target search term is "2024 A15 charger", according to the preset parameter mapping rules, the query parameter matching results corresponding to this type of time-sensitive + digital entity characteristics can be: vertical search channel mapped to digital product information channel and e-commerce product channel, matching mode mapped to exact match, time filter parameter mapped to the past two years, category limitation parameter mapped to mobile phone accessories category, deduplication filter parameter mapped to enable duplicate information filtering, and sorting weight parameter mapped to prioritize timeliness weight and product relevance weight.

[0116] S204. Based on the target search term and query parameters, call at least one search engine to perform search processing on the target search term and obtain search results.

[0117] For example, a search engine is a general-purpose retrieval tool with the ability to retrieve information from the entire web or a vertical field, and can return relevant information based on the input search terms and query parameters.

[0118] At least one search engine indicates that it can invoke one or more different types of search engines to enrich the source of results.

[0119] In one example, an electronic device can combine target search terms with predetermined query parameters based on an intelligent agent, and initiate a search request by calling one or more general search engines through an interface. After the search engine completes the search according to the configuration of the query parameters, it returns search results containing information such as product definition, category classification, and background knowledge. The intelligent agent receives the search results.

[0120] This method, which involves using at least two different search engines to jointly retrieve and process target search terms, effectively broadens the channels and coverage of background information acquisition. It overcomes the limitations of single search engines, such as limited data sources, singular information dimensions, and insufficient coverage of long-tail and ambiguous search terms. By integrating the retrieval advantages of different search engines, differentiated and complementary background content can be obtained from multiple information sources. This avoids the problems of incomplete or inaccurate background information extraction due to single-engine retrieval bias or information gaps. Simultaneously, it enhances the richness and authority of search results, providing more ample high-quality data support for subsequent large-scale model summarization and generation of accurate and reliable target search term background information. This strengthens the ability to analyze fuzzy and ambiguous search intents, providing a more solid guarantee for accurate category alignment and accurate matching of product search results.

[0121] S205. For each correspondence in the knowledge base, determine the similarity between historical search terms and target search terms in the correspondence.

[0122] For example, similarity is a quantitative value used to measure the degree of semantic matching between the target search term and the historical search terms in the corresponding relationships in the knowledge base. The higher the value, the more similar the search intentions of the two are.

[0123] In one example, as mentioned above, the knowledge base includes multiple sets of historical search terms and their corresponding product categories. The agent first converts the target search term into a semantic vector, and then performs similarity calculations between this vector and the high-dimensional vectors of each historical search term in the knowledge base. This process determines the semantic similarity between the target search term and each set of historical search terms in the knowledge base. It should be noted that this application does not limit the method of similarity calculation; for example, it can use cosine similarity.

[0124] S206. In response to a similarity greater than a preset threshold, the corresponding relationship is determined as example information that matches the target search term.

[0125] For example, the preset threshold is a pre-set similarity threshold value, such as 0.8. This preset threshold can be used to filter historical mapping data with high matching degree and remove invalid information with low semantic relevance.

[0126] In one example, the agent compares the calculated similarities with preset thresholds, filters out historical search terms and product category correspondences with similarities exceeding the preset thresholds, and determines these filtered mapping data as example information for this category prediction.

[0127] S207. Call the large model to summarize and process the search results to obtain the background information of the target search term.

[0128] Understandably, one of the main reasons why traditional platform search engines struggle to correctly parse search terms is the lack of background knowledge. For example, when a user enters the target search term "xxx model," if the search engine knows that "xxx model" refers to the electric vehicle of Company B, it will be more effective in accurately understanding the user's query intent. Therefore, this step uses a large model to summarize the real-time search results returned by the search engine to obtain the background information of the target search term.

[0129] For example, summary processing is the process by which a large model sorts out and extracts the core content from redundant and scattered search results.

[0130] Background information is key information extracted from search results, including core definitions, category attributes, and product associations, used to eliminate semantic ambiguity of the target search terms.

[0131] In one example, the agent inputs the raw search results returned by the search engine into a large model. The large model removes redundancy, advertisements, and irrelevant content from the search results, extracting core content such as product definitions, category affiliations, and common names corresponding to the target search terms, generating concise and accurate background information. Table 2 shows an example of a target search term and background information provided in an embodiment of this application.

[0132] Table 2 Target Search Terms and Background Information

[0133]

[0134] As mentioned above, some search terms require background knowledge to be accurately interpreted. For example, a charger for brand A should be understood as a mobile phone accessory rather than a regular plug. In this way, electronic devices can integrate search engine results based on intelligent agents and use LLM to extract key information and analyze intent from the search results to obtain the background information shown in Table 2, thereby enhancing the semantic understanding of the target search terms.

[0135] S208. Construct suggestion words based on the target search term, background information, and example information.

[0136] For example, prompt words are standardized input texts that are fed into the large model to guide it in completing the category prediction task. They integrate three core types of content: user search needs, supplementary background information, and business reference examples, and can provide the large model with complete reasoning basis.

[0137] In one example, the agent, following a pre-defined prompt template, systematically concatenates and formats the target search term, refined background information, and selected example information to construct a logically clear and complete prompt. For example, this prompt template could be:

[0138] my_prompt='''

[0139] #Task Commands

[0140] Based on the following three types of information, select the most matching product category from the [Category List of Example Information]:

[0141] 1. Target search term: {myquery}

[0142] 2. Background information:\n{query_argument}\n

[0143] 3. Example Information Top 20 Examples:

[0144] - Format: -query→category→category ID -

[0145] {examples}\n

[0146] #Analysis Steps

[0147] 1. Keyword extraction Identify the core product terms and attribute terms in the target search terms.

[0148] 2. Search result verification :Analyze the high-frequency category-related words appearing in the search results text

[0149] 3. Example Match Compare the category distribution patterns in the example information.

[0150] 4. Confidence assessment Based on the given information, estimate the confidence score for the product category, with a score ranging from 0 to 1.

[0151] #Output Requirements

[0152] json

[0153] {

[0154] "query":"",

[0155] "Matching category": "",

[0156] "Matching Category ID": "",

[0157] "Confidence level": "",

[0158] "Reasons for Confidence":[

[0159] Keyword matching: "

[0160] Search results clue:

[0161] Example similarity: ]

[0163] }

[0164] '''.

[0165] S209. Based on prompts and the big model, determine the target product category.

[0166] In one example, the intelligent agent inputs the constructed prompt words into the large model. The large model combines the target search term requirements, background information interpretation, and category mapping logic of the example information in the prompt words to perform semantic reasoning and category matching, and finally determines and outputs the target product category that conforms to the platform's business rules.

[0167] As mentioned earlier, each e-commerce platform sets its own category rules based on its own business rules, which can be subjective. For example, "printer" might be categorized as "office supplies" or "home appliances," depending on the platform's specific category rules. This approach leverages a large model to enhance the contextual information (background information) of the target search term and combines it with dynamic example information to clarify business category boundaries and mapping relationships. It understands and infers the deeper intent of the target search term, ultimately determining the accurate product category that corresponds to the target search term and conforms to the business's own category system. This solves the problems of ambiguous intent and category alignment, correctly classifying simple descriptions of model numbers or common names into their corresponding product categories, thus improving the accuracy of category prediction in e-commerce scenarios.

[0168] Optionally, referring to the aforementioned prompt word template, the large model can also output the category ID of the target product category, the confidence level, and the reason for confidence analysis. Table 3 is an example of model output information provided in an embodiment of this application.

[0169] Table 3 Model Output Information

[0170]

[0171] S210. Based on the target product category, determine the product search results corresponding to the target search term.

[0172] It should be noted that this step is similar to the aforementioned step S105, and will not be repeated here.

[0173] S211. Output the product search results to the target application software for display.

[0174] For example, display is an interactive operation in which the target application software presents product search results to the user in a visual form such as a list or card, thereby realizing the front-end display of search results.

[0175] In one example, the generated product search results are transmitted to the front-end interface of the target application software based on the intelligent agent. The application software arranges the product information according to the preset display layout and clearly displays the product search results to the user.

[0176] The method for determining product search results provided in this application involves an intelligent agent acquiring the target search term input by the user, extracting the feature information of the target search term and matching it with corresponding query parameters, calling a search engine to obtain search results, and simultaneously filtering example information matching the target search term from a knowledge base. A large model then summarizes the search results to obtain background information, combines the target search term, background information, and example information to construct prompt words and infer the target product category, and finally generates product search results based on the target product category and outputs them to the target application software for display. This method improves the accuracy of external retrieval by adapting query parameters to search term features, ensures the matching effectiveness of business examples through similarity filtering, and achieves accurate alignment of fuzzy search term intent parsing with platform categories through large-scale model fusion of multi-source information. The entire process is coordinated and executed by an intelligent agent, effectively solving the problem of intent recognition for long-tail and ambiguous search terms, ensuring a high degree of matching between search results and the platform's category system, significantly improving the relevance and accuracy of product search results, optimizing the user's search interaction experience, and simultaneously improving the product retrieval efficiency and transaction conversion potential of e-commerce platforms.

[0177] Figure 4 A schematic diagram of the device for determining product search results provided in this application is shown below. Figure 4 As shown, the product search result determination device 300 provided in this embodiment includes:

[0178] The acquisition module 301 is used to acquire the target search terms entered by the user in the target application software based on the intelligent agent;

[0179] The search module 302 is used to call a preset search engine to perform search processing on the target search terms and obtain search results; wherein, the search results are used to supplement the background information of the target search terms.

[0180] The retrieval module 303 is used to retrieve example information from a preset knowledge base based on the target search term; wherein, the knowledge base includes the correspondence between historical search terms and product categories; the example information is the correspondence between historical search terms that match the target search term and product categories;

[0181] The prediction module 304 is used to call a preset large model to determine the target product category corresponding to the target search term from the example information based on the target search term and the search results.

[0182] The determination module 305 is used to determine the product search results corresponding to the target search term based on the target product category.

[0183] In one possible implementation, the prediction module 304 is used for:

[0184] The large model is invoked to summarize and process the search results to obtain background information on the target search terms;

[0185] Construct suggestion words based on the target search term, background information, and example information;

[0186] Based on prompts and a large model, the target product category is determined.

[0187] In one possible implementation, the retrieval module 303 is used for:

[0188] For each correspondence in the knowledge base, determine the similarity between historical search terms and target search terms in the corresponding correspondence;

[0189] If the similarity is greater than a preset threshold, the corresponding relationship is determined as example information that matches the target search term.

[0190] In one possible implementation, the device further includes a building module for:

[0191] Obtain historical search terms and the corresponding click behavior information;

[0192] For each historical search term, the product category to which the historical search term belongs is determined based on the product with the highest click-through rate in the click behavior information corresponding to that historical search term;

[0193] A knowledge base is constructed based on the correspondence between historical search terms and the product categories to which they belong.

[0194] In one possible implementation, the search module 302 is configured to:

[0195] Based on preset feature extraction rules, extract feature information of the target search terms;

[0196] Based on feature information and preset parameter mapping rules, determine the query parameters corresponding to the target search term;

[0197] Based on the target search term and query parameters, at least one search engine is invoked to perform a search on the target search term and obtain the search results.

[0198] In one possible implementation, the feature information includes at least one of entity features, length features, timeliness features, and ambiguity features;

[0199] The query parameters include at least one of the following: vertical search channel, matching mode, time filter parameters, category limitation parameters, deduplication filter parameters, and sorting weight parameters.

[0200] In one possible implementation, the device further includes an output module for:

[0201] Output the product search results to the target application for display.

[0202] The product search result determination device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0203] Figure 5 A schematic diagram of the structure of the electronic device provided in this application. Figure 5 As shown, the electronic device 400 provided in this embodiment includes at least one processor 401 and a memory 402. Optionally, the electronic device 400 further includes a communication component 403. The processor 401, memory 402, and communication component 403 are connected via a bus.

[0204] In a specific implementation, at least one processor 401 executes computer execution instructions stored in memory 402, causing at least one processor 401 to perform the above-described method.

[0205] The specific implementation process of processor 401 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0206] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0207] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0208] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0209] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0210] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0211] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0212] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0213] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0214] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0215] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0216] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0217] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0218] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method of determining a merchandise search result, characterized by, The method is applied to an electronic device, the electronic device having target application software installed, the target application software integrating an intelligent agent, and the method comprising: Based on the intelligent agent, the target search terms entered by the user in the target application software are obtained; A preset search engine is invoked to perform a search on the target search term to obtain search results; wherein, the search results are used to supplement the background information of the target search term; The system retrieves example information by searching a preset knowledge base based on the target search term; wherein, the knowledge base includes the correspondence between historical search terms and product categories; and the example information is the correspondence between historical search terms that match the target search term and product categories. The system invokes a pre-defined large model and, based on the target search term and the search results, determines the target product category corresponding to the target search term from the example information. Based on the target product category, determine the product search results corresponding to the target search term.

2. The method of claim 1, wherein, The process of calling a preset large model, determining the target product category corresponding to the target search term from the example information based on the target search term and the search results, includes: The large model is invoked to summarize and process the search results to obtain the background information of the target search term; Construct suggestion words based on the target search term, the background information, and the example information; Based on the prompt words and the large model, the target product category is determined.

3. The method of claim 1, wherein, The step of retrieving sample information matching the target search term from a preset knowledge base includes: For each correspondence in the knowledge base, determine the similarity between the historical search term and the target search term in the corresponding correspondence; In response to the similarity being greater than a preset threshold, the correspondence is determined as example information matching the target search term.

4. The method of claim 1, wherein, The method further includes: Obtain historical search terms and the click behavior information corresponding to those historical search terms; For each historical search term, the product category to which the historical search term belongs is determined based on the product with the highest click-through rate in the click behavior information corresponding to that historical search term; The knowledge base is constructed based on the correspondence between the historical search terms and the product categories to which they belong.

5. The method of claim 1, wherein, The process of calling a preset search engine to search for the target search term and obtaining search results includes: Based on preset feature extraction rules, the feature information of the target search term is extracted; Based on the feature information and the preset parameter mapping rules, the query parameters corresponding to the target search term are determined; Based on the target search term and the query parameters, at least one search engine is invoked to perform a search on the target search term and obtain search results.

6. The method of claim 5, wherein, The feature information includes at least one of entity features, length features, timeliness features, and ambiguity features; The query parameters include at least one of the following: vertical search channel, matching mode, time filter parameters, category limitation parameters, deduplication filter parameters, and sorting weight parameters.

7. The method according to any one of claims 1 to 6, characterized in that, The method further includes: The product search results are output to the target application software for display.

8. A device for determining product search results, characterized in that, The device is applied to an electronic device, the electronic device having target application software installed, the target application software integrating an intelligent agent, and the device comprising: The acquisition module is used to acquire, based on the intelligent agent, the target search terms entered by the user in the target application software; The search module is used to call a preset search engine to perform search processing on the target search term and obtain search results; wherein, the search results are used to supplement the background information of the target search term; The retrieval module is used to retrieve example information from a preset knowledge base based on the target search term; wherein, the knowledge base includes the correspondence between historical search terms and product categories; the example information is the correspondence between historical search terms and product categories that match the target search term; The prediction module is used to call a preset large model to determine the target product category corresponding to the target search term from the example information based on the target search term and the search results. The determination module is used to determine the product search results corresponding to the target search term based on the target product category.

9. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-7.

Images